<%BANNER%>

Self-Configurable Communication Network for Wireless Multi-Robot Testbed


PAGE 1

SELF-CONFIGURABLE COMMUNICATION NETWORK FOR WIRELESS MULTIROBOT TESTBED By CHUN-HAUR CHAO A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORI DA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2005

PAGE 2

Copyright 2005 by Chun-Haur Chao

PAGE 3

This document is dedicated to my loving family.

PAGE 4

ACKNOWLEDGMENTS The author expresses his sin cere gratitude to his advisor, Dr. Norman G. Fitz-Coy, for his exhortation and motivation to drive this research in its attention to detail. The author also expresses his grat itude to his committee, Dr. Gl oria J. Wiens and Dr. Haniph A. Latchman, for their instruction and guidan ce. The author acknowledges the University of Floridas Mechanical and Aerospace Engineering Department for offering the opportunity and financial support to fini sh the Master of Science degree. The author thanks the friendship and selfless knowledge sharing found in the members of AMAS (Autonomous Multi-Agent System): Andrew Tatsch, Svetlana Gladun, Daniel Jones, Sharanabasaweshwara Asundi. The author would like to express his grat itude for the unconditional support from his parents, especially when his family is in a difficult situation. This research would never have been completed without their ge neral giving both spirit ually and financially. The author also appreciates his girl fri end for encouragement and companionship. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................iv LIST OF TABLES ...........................................................................................................viii LIST OF FIGURES ...........................................................................................................ix CHAPTER 1 INTRODUCTION........................................................................................................1 1.1 Applications of MRS.........................................................................................1 1.1.1 Military Applications...............................................................................2 1.1.2 Civilian Applications...............................................................................5 1.1.2.1 MRS for jet engine inspection.........................................................5 1.1.2.2 Robot soccer competition................................................................7 1.1.2.3 Multi robot search and rescue.........................................................9 1.1.3 Space-Based Applications.....................................................................11 1.2 Fundamental Issues..........................................................................................13 1.2.1 Autonomous Behavior...........................................................................13 1.2.2 Cooperative Operation and Communications........................................15 1.2.3 Hardware Restriction.............................................................................16 1.3 Methodology....................................................................................................17 1.3.1 Autonomous Behavior...........................................................................17 1.3.2 Cooperative Operation and Communication.........................................21 1.4 Motivation and Scope of the Research............................................................24 2 NETWORK COMMUNICATIONS..........................................................................27 2.1 Evolution of Network Communications..........................................................28 2.1.1 Message Switching................................................................................28 2.1.2 Circuit Switching...................................................................................30 2.1.3 Packet Switching....................................................................................32 2.2 Layered Architecture.......................................................................................34 2.2.1 OSI Reference Model............................................................................34 2.2.2 TCP/IP structure.....................................................................................38 2.3 Wireless Communications and Issues..............................................................41 2.3.1 Medium Access Control Protocol..........................................................42 v

PAGE 6

2.3.2 Ad hoc and Infrastructure Topology......................................................46 2.4 Communication Performance..........................................................................48 2.4.1 Bandwidth..............................................................................................49 2.4.2 Transmission Loss.................................................................................49 2.4.3 Throughput.............................................................................................50 2.4.4 Latency...................................................................................................51 3 WIRELESS MULTI-ROBOT TESTBED..................................................................52 3.1 Hardware Architecture of the testbed..............................................................52 3.2 Wireless Mobile Robot....................................................................................54 3.2.1 Power Module........................................................................................56 3.2.2 Communication Module........................................................................57 3.2.3 Hardware Control Unit..........................................................................57 3.2.4 Processing Unit......................................................................................58 3.3 Positioning System...........................................................................................59 3.4 Operational Area..............................................................................................61 4 PROTOCOL SUITE...................................................................................................63 4.1 Limitations and Requirements.........................................................................63 4.2 Local Area Network Architecture....................................................................65 4.3 Protocol Specifications....................................................................................67 4.3.1 Data Link Layer Protocol.......................................................................69 4.3.1.1 Link management..........................................................................71 4.3.1.2 Forward error correction...............................................................72 4.3.1.3 Feedback error correction..............................................................73 4.3.2 Agent Communication Language..........................................................74 5 SELF-CONFIGURABLE TOPOLOGY....................................................................80 5.1 Eligibility List..................................................................................................81 5.2 Self-configuration............................................................................................82 5.3 Test Configuration...........................................................................................84 5.4 Network Initialization......................................................................................85 5.5 Follower Failure...............................................................................................86 5.6 Leader Failure..................................................................................................87 5.7 Control Privilege Transfer...............................................................................88 6 CONCLUSIONS AND FUTURE WORKS...............................................................91 6.1 Conclusions......................................................................................................91 6.2 Future Works...................................................................................................92 ACRONYMS.....................................................................................................................94 LIST OF REFERENCES...................................................................................................99 vi

PAGE 7

BIOGRAPHICAL SKETCH...........................................................................................103 vii

PAGE 8

LIST OF TABLES Table page 1.1 Classifications of motion planning..............................................................................18 1.2 Classifications of MP algorithm..................................................................................18 1.3 Comparison of centralized control and decentralized..................................................25 2.1 Morse code...................................................................................................................29 2.2 ASCII table................................................................................................................ ..30 2.3 List of network protocols.............................................................................................40 3.1 Comparison of different communication channels......................................................57 3.2 Specifications for processing unit................................................................................59 4.1 Proposed agent communication language....................................................................75 5.1 Computer configurations for the test...........................................................................84 viii

PAGE 9

LIST OF FIGURES Figure page 1.1 Key DARPA accomplishments since 1960s..................................................................3 1.2 Robotic evolution.......................................................................................................... .5 1.3 NASA Glenn miniature mob ile sensor platform...........................................................6 1.4 The concept of jet engine inspection.............................................................................7 1.5 Small size league in RoboCup 2004..............................................................................8 1.6 Control diagram of robot soccer....................................................................................9 1.7 Search and rescue operation by MOVER system........................................................11 1.8 Layered multi-robot architecture.................................................................................12 1.9 Planetary Surface Robot Work Crew (RWC)..............................................................16 1.10 The solution path is shown in the bold lines in the visibility graph..........................19 1.11 Object-dependent cell decomposition........................................................................20 1.12 Quadtree motion planning..........................................................................................20 1.13 Function of the potential field....................................................................................21 1.14 The stabilization of formation control.......................................................................23 1.15 Estimation of sensor positions using Kalman filter...................................................23 1.16 Temperature gradients insi de the target building.......................................................23 2.1 Telephone network connections..................................................................................31 2.2 Network Switching......................................................................................................33 2.3 OSI reference model....................................................................................................35 2.4 Comparison of layer definition between OSI model and TCP/IP structure.................38 ix

PAGE 10

2.5 Encapsulation of header and erro r check code into data units.....................................40 2.6 CSMA-CD...................................................................................................................4 4 2.7 Hidden terminal problem for wireless network...........................................................44 2.8 CSMA-CA...................................................................................................................4 5 2.9 Network topologies......................................................................................................47 3.1 Hardware architecture of the testbed...........................................................................53 3.2 WALKER for multi-robot testbed...............................................................................55 3.3 Block diagram for modules on WALKER...................................................................56 3.4 Hardware pictures for power module..........................................................................56 3.5 Hardware pictures fo r communication module............................................................57 3.6 Hardware and hardware control unit............................................................................58 3.7 PC/104 processing unit................................................................................................59 3.8 Block diagram of the PhaseSpace positioning system.................................................60 3.9 Pictures of hardware for PhaseSpace positioning system............................................61 3.10 Geometry of the testbed.............................................................................................62 4.1 Comparison of the interconnectio ns of different networks..........................................65 4.2 Dedicated wireless network layers...............................................................................69 4.3 Bit-wise format of the control field.............................................................................71 4.4 Normal response mode................................................................................................72 4.5 ARQ methods...............................................................................................................7 3 4.6 Examples of ACL messages........................................................................................77 4.7 Comparison between different encoding methods for ACL........................................78 4.8 Comparison of the performance on different message encoding methods..................79 5.1 Flowchart of self-configuration mechanism................................................................83 5.2 Display during the test.................................................................................................85 x

PAGE 11

5.3 Network initialization process.....................................................................................86 5.4 Topology configuration when follower fails...............................................................87 5.5 Topology configuration when the leader fails.............................................................88 5.6 Topology configuration for the control privilege transfer...........................................89 xi

PAGE 12

Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science SELF-CONFIGURABLE COMMUNICATION NETWORK FOR WIRELESS MULTIROBOT TESTBED By Chun-Haur Chao August 2005 Chair: Norman G. Fitz-Coy Major Department: Mechanical and Aerospace Engineering The Multi-Agent Systems (MAS) have been studied over decades. Various issues were well discussed conceptually. The impl ementation of MAS on the physical hardware is an important phase for the research de velopment. Hardware validation process promotes the theoretical concepts to real istic problems. Nevertheless, the hardware implementation is somewhat costly and sophi sticated. A multi-robot testbed is a cost effective solution to implement different concepts for MAS. This research proposes an architecture for the hardware implementation of a MAS. A description about the design of a wireless multi-robot testbed for further MAS research is provided. Relevant research topics including path plan ning and cooperative control are also briefly introduced in the thesis. Meanwhile, the communication is a prerequi site before the validation of MAS. The work in this thesis will mainly focus on the network architecture for the communication between robots. The objectives for the ne twork communications are to simplify the xii

PAGE 13

existing framework and maintain the flexibility for any further revision. Also, in order to enable the inter-cooperation between robots, the agent communication language is used to provide a standard for the conversation. Moreover, the MAS provide a decentralized control scheme for the system. It is more robust for any exceptional incidents and failures. In order to enable a more robust communication environment, the self-configuration of the network topology is proposed in this thesis as well. For such a self-confi gurable network, the failure of the robot for the communication can be accommodated. Dynamic adjustment to an optimal topology could therefore be made. The work presented in this thesis provid es a hardware solution for MAS research. The self-configurable topology offers a flexible network scheme for wireless communication and decentra lized control scenario. xiii

PAGE 14

CHAPTER 1 INTRODUCTION Multi-Robot Systems (MRS) have brought the robotic researches into a new paradigm. Not only the functiona lities but also the cooperative operations between robots have excited the interests and attentions of the robot communities. The widely applicable MRS research could be mainly divided into three categories: space-based, military and civilian uses. Moreover, for so me of the issues, for exampl e, cooperative control, path planning, and communications have become more and more important to the development of MRS. However, the physical implementations of su ch systems may be restricted by the limitations of hardware. The dimension of the robot body, extra payload, the performance of the sensors and actuator s, or the controllers information process capability could all substant ially affect the overall performance. In addition, the communication capability is one of the rest rictions for some implementations like spacecraft communications or smaller size robo ts. In this chapter, the applications for MRS and the corresponding issues will be menti oned. This effort will facilitate a further integration of the hardware implementations pr ovided in the thesis as well as the network communication design into a larger framework of the MRS research. 1.1 Applications of MRS Traditionally the robot communities focused their research interests in the domain of the single robot applications. However, with the growth of the semiconductor and communication technologies, for exampl e, MEMS, wireless network and Global Positioning Systems (GPS), the development of mobile robot technology has been 1

PAGE 15

2 transferred to the multi-agent level. The st udies of the cooperativ e and collaborated control for MRS has been extensively discus sed and implemented since last decade (see JRP [2]). The applications for MRS are fair ly diverse from military unmanned espionage, mine sweeper to rescue or space exploration mission. The following section summaries all the applications into thr ee major application categories military, civilian, and spacebased. Some relevant developments will be mentioned as an example.. 1.1.1 Military Applications The effort to apply the MRS research for military uses has been initiated and performed by several agencies and programs, for example, Defense Advanced Research Project Agency (DARPA) [ 1 ] and Joint Robotics Program (JRP) [ 2 ]. The objectives for the robot researches within these programs include: Increase the autonomous mobility Refine the tactical behavior Design the innovative platform Minimize the robot dimension DARPA has been successfully been merged the cutting edge te chnologies into the robot researches in the past couple of decades, for example, the communications and artificial intelligence enhan ce the controllability and au tonomy on the robot. Figure 1.1 shows the major accomplishments by DARPA since 1960s. With the current achievements on the communication systems, for example, GPS technology, the mature of the network development and the various artificial intelligent algorithms for the autonomous behavior, the Multi-Agent Systems (MAS) could have been able to promote the status from software agents to physical agents. Therefore, as for the military

PAGE 16

3 applications, the MRS including various unman ned vehicles would become new thrusts for the warfare development. Figure 1.1 Key DARPA accomplishments since 1960s. [ 1 ] Moreover, the integration of separate Unmanned Ground Vehicle (UGV) robotics developments and projects had also been made into the Joint Robotics Program [ 2 ] by the Office of the Secretary of Defense (OSD) in 1990 at the recommendation of the Senate Appropriations Committee. This ef fort also advances the successfully global deployments of UGV in some of the military operations in cluding Bosnia, Afghanistan, and Iraq in the past decade. UnManned Systems (UMS) have been validated widely during the mentioned ongoing services. UMS as described in the co ncepts below are envisioned to contribute the increase of mission effectiveness and are planned for integration into service force structures [ 2 ]:

PAGE 17

4 Army Future Force: Future Combat Systems (FCS) Marines/Navy Gladiator Tactical Unmanned Ground Vehicle: Autonomous Operations Air Force Air Expeditionary Warfare: Robotics for Agile Combat Support and the Airborne Explosive Ordnance Disposal Concepts However, the goal of JRP has been to develop a diverse family of UGVs and to foster service initia tives in ground vehicle robotics to meet requirements for greater mission diversity and increasingly more autonomous control architectures [ 2 ]. As in Figure 1.2, the maturation and transition of the technology to the robotic systems will feature the robotic services with more aut onomous capabilities. Therefore, the enhanced object recognition and tactical behavior could enable the use of robotics to a fairly extensive and effective manner. As it is shown in Figure 1.2, the advanced technologies such as route planning, mission planning and ta rget recognition will lead the MRS from a teleoperational service to a more and more autonomous service. As in the current progress of MRS, the system is undergoi ng the development of route planning and heading to a mission planning level, where th e robot autonomy will exceed the level of human intervention. Both the route planning a nd mission planning have the critical need to perform the dispatched tasks cooperativ ely. Such a cooperation behavior requires a reliable and robust communication system. More over, a further advance phase in Figure 1.2 requires higher and higher communication data rate. Therefore, th e importance of the communication system in a MRS will be higher and higher.

PAGE 18

5 Figure 1.2 Robotic evolution. [ 2 ] 1.1.2 Civilian Applications The efficiency and effectiveness of the MRS can not only facilitate for military purposes but also benefit civilian uses. Some times the tasks are too complicated for a single robot and we need multi robots to work cooperatively. For example, the search and rescue mission using multiple vehicles will influentially decrease the time required to complete the mission. Multiple robots can also be used for moving a larger object. Brief descriptions on some of the general applic ations will be discussed in this thesis. 1.1.2.1 MRS for jet engine inspection The maintenance of the aircraft is critical to the aviation safety. Jet engines undergo examinations for detecting potential flaws on the surface of their components, such as

PAGE 19

6 cracking and erosion. Usually inspection methods are usually either the invasive borescopic or a full teardown, both of thes e methods are time-consuming. Full teardown, however, is even more time consuming and costly, and is often applied for only the situations when damage is detected and the replacement of parts is necessary. NASA Glenn Research Center [ 3 ] proposed another approach for the inspection of the jet engine health. Instead of manual in spection procedures, miniature mobile sensor platforms could provide another option. The mobile robots equipped with the vision and communication systems can roam arbitrarily on the surfaces inside the engine. The robots could hence send the internal im age of the jet engine back to the station. Therefore we can go through the engine inspection process with less human power and time. Figure 1.3 and Figure 1.4 show the miniature mob ile robots in the inspection concept. Figure 1.3 NASA Glenn miniature mobile sensor platform. [ 3 ]

PAGE 20

7 Figure 1.4 The concept of jet engine inspection. [ 3 ] 1.1.2.2 Robot soccer competition The concept of robots playing soccer was first introduced by Alan K. Mackworth [ 4 ] in 1993. By using the global vision syst em, the main station could acquire the orientation and position of each different robot. A wireless communication is also required to transmit the commands to the robots An artificial intellig ence is involved to determine the strategy to compete against the other team. In July 1997, the first official conference and games were held in Na goya, Japan. Followed by Paris, Stockholm, Melbourne and Seattle where th e annual events attracted ma ny participants and spectators [ 5 ]. There are five different leagues in th e RoboCupSoccer: simulation league, small size league, middle size league, four legged robot league, and hum anoid league. Comparing to the other MRS applications, robot soccer is highly dynamic and the state change is in real-time. No human intervention is allowed during the period of th e game. Its situation in a MRS development phase contrast to Figur e 1.2 is a more advanced phase including

PAGE 21

8 the object recognition and situational awareness. The research fields cover various from artificial intelligence to robotics. Such areas include real-time sensor fusion, reactive behavior, strategy acquisition, learning, real-time planning, multi-agent systems, context recognition, vision, stra tegic decision-making, motor control, autonomous robo control and more. Figure 1.5 is the picture of a competition in the small size league in RoboCup 2004 at Lisbon, the regula tion restricts each robot must be able to fit inside a 180mm diameter cylinder. In Figure 1.5, differe nt colors and marks at the top of each areas t robot is used to identify the position and orientations of different robots from cameras. then Figure 1.5 Small size league in RoboCup 2004. [ 5] Due to the dimension restriction, the contro l of the whole team is usually somewhat centralized in order to reduce the onboard processing need. A global vision system is used to trace the robots and ball. The c ontrol diagram is in Figure 1.6. The hardware architecture adopted in Figure 1.6 is, however, fairly similar to the wireless multi-robot testbed we will discussed later in the thesis An overhead camera system is used as the

PAGE 22

9 object positioning system for the robot localiz ation. The image processing is dealt by a base station, and proper control commands for the robots are transmitted by a wireless transceiver. Therefore, the robot soccer can also be taken as a platform for the evalua tions of various MAS concepts, as the same purposes for the wireless multi-robot testbed. l diagram of robot soccer. 1.1.2. hich Figure 1.6 Contro 3 Multi robot search and rescue The task of search and rescue can be performed by a single robot. RobotCup [ 5 ] also has a separated domain ca lled RoboCupRescue. The robot is required to be operated in a dedicated scenario autonomously. Howe ver, the cooperative mu lti-robot search and rescue operation could greatly increase the e fficiency or enable some capabilities w couldnt be performed by a single robot. James S. Jennings et al. [ 6 ] discussed the

PAGE 23

10 cooperation of robots for a s earc h and rescue mission. The following capabilities are perform such tasks. lization C The algorithm of the propo us. ue. Step 4: One robot finds the target wh ile all the other r obots still searching. r robots that target has Step 8: T es the target and notifies other robots. Step 9: All the robots manipulate the target toward the intended location collaboratively. required to Navigation and loca Search Object recognition ommunication with other members in the team Cooperative manipulatio n of large objects The MRS demonstrated in Figure 1.7 is named as MOVER sed MOVER system [ 6 ] performs the task in Figure 1.7 by the following steps: Step 1: A workstation and 5 robot s idle for the initial stat Step 2: The workstation initiates the pr ogram to perform the search and resc Step 3: All robots start to search the house shape object. Step 5: The robot that f ound the target notifie s all the othe been found. Step 6: The other robots are heading toward to the target. Step 7: Another robot arrives the target and notifies other robots. he last robot arriv

PAGE 24

11 Figure 1.7 Search and rescue operation by MOVER system. [ 6 ] 1.1.3 Space-Based Applications The MRS researches have also been studi ed in many of the space mission projects, for example, NASAs Mars e xploration project (See NASA [ 7 ]) and the Demonstration of Autonomous Rendezvous Technology (DART) mission (See NASA [ 8 ]). The spacebased applications varied from formation flying, cooperative control for multi-arms, to autonomous mobile robots for space explora tion. Comparing to the terrestrial-based applications, the space-based applications us ually have more serious technical concerns in the following characteristics: Uncertain environments Communication delays Limited sensing and actuation Scalability

PAGE 25

12 Dani Goldberg et al. from Carnegie Mellon University [ 9 ] discussed the synchronization and coordination for mobile robots for the application to space exploration. The Mars explorati on scenario has been set in th e discussion. The distributed layered architecture is proposed for the highest possible scientific return on the given tasks. Due to the limitations on the communications (bandwidth and latency), the centralized control is not reliable. So the robots are responsible for making the decisions based on the priority of the tasks and how the tasks are to be accomplished. The architecture is shown in Fi gure 1.8. The planning layer send s the plans to the executive layer, which could further decompose tasks in to subtasks and dispatch them based on the temporal constraints imposed by the plan. The behavior layer is responsible for the control of the robot or updates the information from the sens ors or the status. Also, the executive layer is responsible for monitoring the tasks status and returning them to the planning layer. This layered architecture can be also used on a single autonomous robot. It is shown in Figure 1.8. Figure 1.8 Layered multi-robot architecture. [ 9 ]

PAGE 26

13 In Figure 1.8, each robot has the mentioned three layered architecture. The communication can occur either vertically be tween either different layers on the same robot or horizontally the same layers on different robots. Th e information of sensor data, plans, or tasks in this architecture could hence be exchanged and coordinated. However, the communication performance in this architecture dominated the system performance. Requirements for a higher bandwidth and lower latency communication in order to coordinate the action are needed, wh ich might not be always allowed. 1.2 Fundamental Issues The various applications pres ented above show different re quirements and restrictions. A good understanding on the potential factors that might affect the performance or feasibility of a MRS application is critic al. A system design or evaluation must be provided under these considera tions. Therefore, the following key factors to the MRS application would be addr essed in the section: Autonomous behavior Cooperative operation and communications Hardware restriction 1.2.1 Autonomous Behavior Some applications only require the base st ation to centrally control different robots. However, since the robots dont take charge of the data anal ysis, the information gathered by various sensors on different robots need to be transmitted to the base station. Also, for the simplicity of the communication, the info rmation will mostly be shared on only the base station. The drawback of the non-autonom ous robot is it cant be operated unless the control command is given from the base st ation. Therefore, the non-autonomous robot itself can be taken as no more than a set of sensor/actuator instead of an intelligent

PAGE 27

14 agent. Nevertheless, the obvious advantage is the design of such a MRS system can hence be greatly simplified and much easie r to be implemented. An example of nonautonomous MRS is the small size robot soccer mentioned earlier. The hardware requirement for an autonomous system is usually depending on the computational and environment sensing needs. From IR sensor, voice detection, thermometer, to relatively complicate systems as well as the vision syst em or laser range detector could all be possible options for an autonomous system. As the earlier discussion, the developmen ts for a MRS toward system autonomy have several different phases from pure tele-operation to full autonomy (See Figure 1.2) during its evolution. Many of th e centralized controlled syst ems are still capable for low level maintenances and fault detections. Ho wever, the high level task schedules or tactical behaviors still need to be assigned by the central station. Multi-robot search and rescue is usually operated under this mode. Each robot in the system can perform its task to search the possible target autonomousl y. The decision making is determined by the base station. Another example is the satelli te system. Navigation, attitude control, or health monitoring modules provided onboard can have the satellite to survive without base station under its autonomous behaviors on the orbit. However, the determination of the flying orbit or docking with other spacec raft is still currently controlled by the commands transmitted from the ground station. The design of the necessary autonomy level for a system is mission dependent. Cost effectiveness could vary case of case. For example, the cost of the implementation for the vision on a small size robot (< 180 mm) is usually expensive. Not only the consumption for the vision module but also the improved capabilities to process the

PAGE 28

15 image data. Communication or power system may also need to be enhanced in order to meet the operation requirement. This may be a big challenge to be implemented on an embedded system and significantly increases the cost of a system. 1.2.2 Cooperative Operation and Communications Another critical issue for MRS is the cooperation of the robot operation. Cooperative operation can be used on a MRS to improve the task performance or enable additional features. The time interval spent to search over a terrai n will be significantly reduced by using multiple robots. Also, the c ooperative localization can help each robot locate itself and understand th e scene better. Sometimes the assigned task is too complicate for a single robot to accomplish and needs to be executed by multiple robots. For example, NASAs Planetar y Surface Robot Work Crew [ 10 ] coordinates the grasp, transport and placement of extended structure using multiple robots, as in Figure 1.9. The formation and cooperative control must occur in such a scheme. A cooperative operation includes many di fferent level problems from task assignment and schedule to the cooperative control and localizati on. One of the most essential components for c ooperative control on MRS is communication. An autonomous robot can have the interactions via environment or sensing without any communication [ 11 ]. However, in most situations, the most effective method to share and update required information with each other is a data networ k. Multi robot search and rescue requires the sharing of information gathered from diffe rent sensors. As the operation procedures mentioned in Figure 1.7, the collaboration requires robots communicate with each other as a network. G. Kantor et al. [ 12 ] discussed the using a netw ork of distributed mobile sensor systems for an emergency response problem. Multiple radio beacons have been deployed in the target building in order to estimate the gradients of temperature in the

PAGE 29

16 building in the discussion. The communica tion network is a prerequisite for many cooperative operations. Figure 1.9 Planetary Surface Robot Work Crew (RWC). [ 10 ] 1.2.3 Hardware Restriction The hardware restriction dominates the capab ility of the MRS. For the mobile robot, the resources are sometimes fairly limited in the system. The power supplied on the robot determines the capabilities of the embedded modules: the controller, all the working sensor and actuator subsystems. Unfortunately, the battery is sometimes relatively large, heavy and a considerable part of the total weight, which might deteriorate the overall performance or disqualify the system requirements. However, the autonomous system or decentralized control often re quires better processing capabilit ies, which results in higher power consumption. For the physi cal restrictions on the mob ile robots, there are some important considerations for a hardware design [ 11 ]: Centralization/decentralization

PAGE 30

17 Differentiation Communication structure Modeling of other agents A careful review on all of th ese factors must be made and taken into consideration while dealing with the design problem fo r the MRS. An example design for small satellite for multi-spacecraft mission can be found in [ 13 ]. 1.3 Methodology In the previous section seve ral issues for the MRS have been addressed. They are critical factors to the MRS. The existing solu tions for these issues will be discussed in this section now. 1.3.1 Autonomous Behavior The autonomy for the robot system i nvolves many different disciplines. Image process, control algorithm, artificial intellig ence, and motion planning may sometimes be needed for an autonomous system. An important aspect of the autonomous behaviors for a mobile robot is Motion Planning (MP). Moreov er, a mobile robot is often required to be able to explore in an uncertain terrain. Hen ce, sensing, obstacle avoiding and planning for the optimal path would be the most cri tical problems for the mobile robot. Y. K. Hwang and N. Ahuja [ 14 ] summarize the recent developments for motion planning problems. Before the discussion of motion planning, we need to classify the type of the problems and problem solving algorithms so appropriate algorithms could be highlighted regarding to speci fic problems. Table 1.1 and 1.2 list the classifications of different problem types and approaches. Also, a proper method to describe the environment, including the robots and obstacles, is essential to MP problems. Instead of using the classical world space representation, which the physical space robots and

PAGE 31

18 obstacles exist in, configur ation space is more frequently mentioned in MP study. Configuration space is a set of parameters that completely specify the positions of every point of the robot or obstacle. Table 1.1 Classifications of motion planning. Yes No Can objects change shape? C onformable Non-conformable Time varying? Time Varying Time invariant Restriction on the motion of robots? Constrained Unconstrained Availability of the obstacle information Dynamic Static Table 1.2 Classifications of MP algorithm. Limited Unlimited Completeness Heuristic Exact Scope Local Global Various approaches have been developed for MP problems. The applicability of each algorithm may be wide or rest ricted. Nevertheless, most of the algorithms could be separated into the following approaches [ 14 ]: skeleton, cell decomposition, potential field. Different approaches are not necessarily exclusive. Between these algorithms, the technique used to solve the MP problem is sometimes a hybrid method. A better performance could sometimes be obtained via the hybrid algorithms. Skeleton The major advantage of skeleton approach is the simplicity of calculation. It simplifies the MP problem into a network of one-dimensional lines so the search can be restricted in the connections between nodes. This algorithm includes three phases. In the first phase, the robot moves from its initial c onfiguration to a node in the skeleton. In the second phase the robot moves from a goal c onfiguration to a node in the skeleton. The third phase connects two point s by using the lines in the skeleton. Figure 1.10 shows the visibility graph of polygon in the plane. S and G in the figure represent the starting

PAGE 32

19 position and goal position respectively. The visibility graph is the collection of lines in the free space that connects di fferent features. There are edges in the visibility graph. is the number of features. 2() On n Figure 1.10 The solution path is shown in th e bold lines in the visibility graph. [ 14 ] Cell Decomposition The cell decomposition, as the name suggeste d, decomposes the free space into a set of simple cells. For the motion planning, we al so need to compute the adjacency of the cells. The decomposition approach can either be object-dependent or object independent. The object-dependent method requires less cells. However, the computation complexity for the boundaries and adjacencies is high. It is shown in the Figure 1.11. First we decide the boundaries and adjacencies of cells by all the sidelines of the obstacles. Then we could determine the cells including the path from S to G. The object-independent decomposition genera lly uses more cells. However, the calculation is less specifically for nontrivial objects. Quadtree [ 15 ] is used for a 2-D

PAGE 33

20 motion planning problem, or octree for a 3-D motion planning problem. The motion planning by quadtree is shown in Figure 1.12. Figure 1.11 Object-dependent cell decomposition. [ 14 ] Figure 1.12 Quadtree motion planning. [ 15 ] Potential Field The potential field constructs the environm ent by using the scalar function called potential. The goal configuration is set to be minimum, and a high value at the occupied space. The function is sloping down anywhere el se toward the goal configuration. By this potential setting, the robot could therefore reach the goal configuration by following the negative gradient. The example of th e potential field is in Figure 1.13.

PAGE 34

21 (a) (b) (c) Figure 1.13 Function of the potential field. [ 14 ] (a) Obstacles has high potentials, (b) The minimal potential locates at goal, and (c) The path to the goal from start could be found along the negative gradient. It could be observed from above discussi ons of various MP methods that a global understanding to the environment is a requisi te. This can be solved by either using a global positioning system or a cooperative data sensing network, which will be mentioned in the next section. 1.3.2 Cooperative Operation and Communication The cooperative operations for the MRS syst em are diverse. Different cooperative behaviors are developed, for example, forma tion, sensor fusion, cooperative localization and control. The disciplines involved to so lve these problems are also very different. Three most common issues are mentioned here. Formation control The formation problem is the most freque ntly discussed problem in the mobile MRS. It is the fundamental problem to cont rol mobile vehicles. Ma ny applications like

PAGE 35

22 search and rescue operations robot soccer, formation flyi ng of Unmanned Air Vehicles (UAV) required the cooperation between vehi cles. Reza OlfatiSaber and Richard Murray [ 16 ] uses a set of parameters to present the formation graph. Then by using a structural potential function, the local collision free stab ilization of formation of multiple vehicles can be obtained. Figure 1.14 (a) and (b) show s the formation for 3 and 6 vehicles respectively. Data fusion The MRS application usually requires the system to be operated in an uncertain environment. Therefore, the capability to sense the environment becomes an important function. Sensing can include simple measur ements from temperature and range to obstacles, to sophisticated multimedia data. However, the sensing information from different robots needs to be further fused in order to gain a better understanding of the environment. The methodology to exchange, process, or integrate the information becomes another issue. A distributed mobile sensor system is mentioned in [ 12 ]. The dynamic localization of all devices is estim ated by Kalman filter, Markov method, and Monte Carlo. The location estimation of unknown tags by Kalman filter is shown in Figure1.15.

PAGE 36

23 Figure 1.14 The stabilization of formation control. [ 16 ] (a) formation for 3 vehicles and (b) formation for 6 vehicles Figure 1.15 Estimation of sensor positions using Kalman filter. ( o is the true location, + is the estimation ) The temperature gradient map is genera ted then by the ad-hoc network of the Mote sensors, as in Figure 1.16. Therefore, the temperature insi de a scene can be dynamically monitored and the better d ecision to rescue lifes can be made. Figure 1.16 Temperature gradients inside the targ et building. Task schedule A distributed layered architecture for mob ile robot coordination is proposed in [ 9 ]. The autonomous high level task decomposition and subtask dispatch service, especially

PAGE 37

24 for time stringent cases and cooperative beha vior, are mentioned. However, for non-time stringent mission, that the delay of operation will not cause catastrophi c effects, a market based auction mechanism is proposed by Brian P. Gerkey and Maja J. Matari [ 17 ]. The mission is constructed into a hier archical structure first. The child task must be performed before the parent task is performed. The parent task is responsible for assigning child task to the robots and monitoring the status of th e tasks. The auction is processed in the following steps: 1. Task announcement 2. Metric evaluation 3. Bid submission 4. Close of auction 5. Progress monitoring/contract renewal The purpose of the auction mechanism is choosing the most appropriate agent to execute the task. The assumption here is the auction is always won by the most appropriate agent. This concept will late r be used for a network self-configuration procedure. 1.4 Motivation and Scope of the Research A common component for all level problem s for cooperative behavior mentioned above is the communication. Just as with hum an beings, communication is a requirement for a successful cooperation. Also, for the mobile robot, wireless communication is needed instead of the communication through wi res. The wireless Ethernet protocol IEEE 802.11 provides a well developed solution for wireless communication network. However, for some physical restrictions, like signal coverage, reliability and hardware complexity, IEEE802.11 could not always occur in all of the MRS, es pecially for deep

PAGE 38

25 space applications. The cooperative control sche me for the MRS could be mainly divided into two types: decentralized and centrali zed. It also relates to the communication architecture and many other factors. Table 1.3 compares the differences of the centralized and decentralized control. More details a bout the differences for communication will be discussed in the next chapter. Table 1.3 Comparison of centralized control and decentralized Centralized Decentralized Hardware requirement Low High Autonomy Low High Communication architecture type Infrastructure Ad hoc Communication complexity Simple Complicate Communication direction Unidirectional/Bidirectional Bidirectional The research interest in this thesis is mainly focused on the hardware implementation and the communication network solution. Fo r centralized control, some simple architecture, for example, channelization is often used for comm unication network to share the medium. However, a central stati on must be used and constantly update the status in order to keep all the devices f unctioning. A service like packet switching or virtual circuit switching is often required to be provided for a decentralized control scheme. As the potential issues discussed above a generic wireless Ethernet solution can not be adapted by all the MRS design. Proper modification must be made in order to customize the communication system to fit the specific scenario and other hardware designs. Chapter 2 will offer a technical review on the network developments before the presentation of the wireless network design. Several network consid erations like hidden terminal problem and ARQ will be implemented on this network. The effort provided in

PAGE 39

26 the thesis is a minimal wireless network in frastructure for a wireless MRS testbed. It offers a wireless networking solution for a small size MRS for the distributed control purposes.

PAGE 40

CHAPTER 2 NETWORK COMMUNICATIONS The importance of communications has been laid out with the discussion of chapter 1. Network communication, as defined here, is a service which could provide the capability to transfer the information betw een three or more different hosts. Network communication is one of the essential features in MRS since the need for different robots to work cooperatively and the information ex change between different robots. In other words, the network services f acilitate different robots to establish the in terconnections instead of to be operated individually. A very fundamental problem for network communications is the medium sharing issu e. A couple of techniques have been developed, for example, channelization, Ti me Division Multiple Access (TDMA) and Code Division Multiple Access (CDMA). For a distributed control scheme, a static medium access control tec hnique like channelization is not enough. The communication of information has to be more dynamic. A p acket switching or virtual circuit method is usually considered. This chap ter introduces the concepts and developments for the network communications. An understanding of the evolution of ne twork developments could highlight the possible issues and functiona lities during the developments of a network communication environment. Some existing works will also be referenced in this chapter. For instance, the layered network concepts and necessary features for the Ethernet like requests of retransmission could also be properly a pplied to any specifi c MRS communication though the existing protocols may not alwa ys proper under all the circumstances. 27

PAGE 41

28 Therefore, an introduction to the history of the network developments will be helpful to identify the network communication problems. Moreover, in order to structure and simplify the various problems, the layered arch itecture for the network is an effective manner to organize the functi onalities for network. Therefore, a brief introduction of Open Systems Interconnection (OSI) model fo r discussing different functionalities of network communications will be presente d in this chapter. Along the network developments, an assessment of the networ k performance would also be analyzed. Some measurements to properly addr ess the performance of the netw ork are included in the last portion of this chapter. 2.1 Evolution of Network Communications Classical network communications, includi ng the telegraph and telephone services, are developed to transfer th e information in text and voice. However, with the rapid growth of semiconductor technologies, info rmation processed by computers increase exponentially. Modern network communication re garding to the computer data transfer, is then developed on the basis of the clas sical network communication techniques. The research objectives of this thesis are to implement a simplified and optimized network design for the computer network communicatio ns. So the evolutions of the network communication are important references fo r such purposes. Alberto Leon-Garcia and Indra Widjaja [ 18 ] separate the network evolutions into three phases: message switching, circuit switching, and packet switchi ng. A brief overview for the classical communications is mentioned here. 2.1.1 Message Switching The most original network communica tion is implemented by the message switching method. Long distan ce communication is relied on the messenger travels

PAGE 42

29 through different locations by feet, animals, or other manners. The communication is fairly slow and unreliable for the manua l message switching. In 1837, the telegraph service was first demonstrated by Samuel B. Morse as a practical communication of text messages over long distance. The message was encoded by the Morse code (see Table 2.1) with a combination of lines and dots. The transmission is made by sending electrical current over a copper wire. The messages could therefore be transmitted almost instantaneously from node to node. A well tr ained human operator can transmit 25 ~ 30 words/minute. Meanwhile, a message routing is still made by human decision according to the destination of messages. Table 2.1 Morse code. A J S 2 B K T 3 C L U 4 D M V 5 E N W 6 F O X 7 G P Y 8 H Q Z 9 I R 1 0 In order to increase the transmission rate for the telegraph, a technique called multiplexing was then introduced. Multiplexing is the attempt to combine the transmitted information over a single telegraph wire. It is also an initial attempt to share medium on the electrical signal transmission. The Baudot system is developed and adopted in 1874 as the first multiplexing system. Baudot system used five binary symbols as a character. The encoding system further evolves a m odern alphanumeric expression ASCII code (American Standard Code for Information In terchange). The table of ASCII code is shown in Table 2.2.

PAGE 43

30 The multiplexing can be realized by si gnal modulation. The modulated signal carries different sinusoidal sign als at different frequencies. The binary information could be transmitted at a pair of frequencies, for example, 0 f as and 1 f as . So by using different pairs of frequencies multiple signals could be transmitted simultaneously. This technique is also known as Frequency Shif t Keying (FSK) as a modern communication modulation method. Table 2.2 ASCII table. 0 1 2 3 4 5 6 7 0 NUL SOH STX ETX EOT ENQ ACK BEL 1 DLE DC1 DC2 DC3 DC4 NAK SYN ETB 2 SP # $ % & 3 0 1 2 3 4 5 6 7 4 @ A B C D E F G 5 P Q R S T U V W 6 ` a b c d e f g 7 p q r s t u v w 8 9 A B C D E F 0 BS HT LF VT FF CR SO SI 1 CAN EM SUB ESC FS GS RS US 2 ( ) + / 3 8 9 : < = > ? 4 H I J K L M N O 5 X Y Z [ \ ] ^ 6 h i j k l m n o 7 x y z { | } ~ DEL 2.1.2 Circuit Switching The telegraph service successfully solv es the huge propagation delay of the traditional communications. However, the manual routing of information still limits the network communications because of the efficiency and reliability. In 1876 Alexander Graham Bell developed a device to transmit th e voice signal. A bidire ctional, real-time transmission of voice signal is called telephone service. The telephone service is an analog transmission system compares to the telegraph service as a digital transmission

PAGE 44

31 system. The telephone service is convenient and could be ope rated by the users with little training. This important charac teristic quickly leads to th e exponential growth of the telephone service. Nevertheless, a problem was recognized very soon after the number of users grew. The telephone service requires a dedicated line between two users. A network of the telephone service with n users needs (1)/ nn 2 dedicated lines, which makes the cost significantly increases. Therefore, in 1878, the telephone switches was introduced to reduce the number of dedicated lines by having operators manually establish the connections based on users demands. Figures 2.1 (a) and (b) s how the differences between the networks conn ections mentioned above. (a) (b) Figure 2.1 Telephone network connections. (a) without circuit switching, and (b) with circuit switching. With the development of electrical switc hes, a hierarchical decimal telephone numbering system was developed later fo r the dialing connec tion in the telephone network for the compatibility of the large numbe r of users. One of the important features with the telephone network is called connectionoriented because a proper setup is required before the information can be excha nged. This communicati on procedure is also called circuit-switching. The connection can be separated into three phases within a phone call. In the first phase a connection reque st sent by the user a nd proper set up needs

PAGE 45

32 to be accomplished. The second phase is the actual transfer of voice from end to end. Finally the third phase is the release for the connection. However, for the circuit switching, the routing decision is made while establishing the services. So the routing information is not needed after the connecti on has been set up in this situation. 2.1.3 Packet Switching The message and circuit switching is frequently used by the classical communication systems. However, they both have their deficiencies. The circuit switching improves the efficiency of traditional manual message routing, but it is initially designed for analog information transmission. The occupation of the dedicated lines will also decrease the efficiency of transmission in a network. Meanwhile, the development of computer technology after 1940s largely increas es the capability of the information process. So the need for a discrete data tran sfer in a more dynamic system is requested. The modern communication network was developed on the basis of such a communication manner. The developments of computer networks were initiated for the military purposes. The first wide geographical area network, Advanced Research Projects Agency NETwork (ARPANET), was developed by Defens e Advanced Research Project Agency (DARPA). ARPANET is the ancestor of the Internet. A discussion on the history of Internet can be found in [ 20 ]. A critical concept before th e implementation of computer networking is packet switching. It is the fundamental principle of the Internet. The packet switching is somewhat similar to the message switching, however information transmitted is cut into separate short segments There are two types of packet switching: virtual circuit and datagram. The virtual ci rcuit requires the conn ection of two end nodes to be set up before the actual transmission, and the route of the transmission is fixed

PAGE 46

33 during the session. This can guarantee the fram es received in order and less overhead is required in the frame. The most common exam ples are the ATM and X.25. Nevertheless, for datagram transmission, each packet is treated as an independent entity, and full information for the recipient is included in the header of a frame. The frames received at the end point might be out of sequence because of their various routes. The example for datagram transmission is Transmission Control Protocol (TCP) and User Datagram Protocol (UDP). The transmi ssion methods for message switch ing, circuit switching, and packet switching are shown in Figure 2.2. (a) (b) (c) Figure 2.2 Network Switching. (a) message switching, (b) circuit Switching, and (c) packet switching.

PAGE 47

34 2.2 Layered Architecture With the growing scale of the networks, the increasing need for switching message leads to the further development of layere d networks. Different functionalities are regulated at each layer. The proper effort is made for the definitions and reorganization of network services based on such an environm ent. The network architectures developed by different vendors varied in the early desi gns. However, the compatibility between different networks begin to attract more and more concerns. A regulated reference network model can also help to bridge diffe rent network environments. This section will discuss briefly on the developments of th e layered networks. Two most frequently mentioned models are introduced here: OSI mode l and TCP/IP structure. The similarities and differences between these structures will be highlighted. 2.2.1 OSI Reference Model The OSI reference model is developed to provide the needs discussed above. The effort is made by International Organizati on for Standardization (ISO) to provide a reference model for Open Systems Interconnec tion (OSI). OSI reference model regulates the functions for networking to seven layers in a stack. Each layer only utilizes the services in the lower layers and provides th e higher-level features to the upper layers. Figure 2.3 shows the relative position of each layer in the mode l stack and how each layer can interact with the others. In Figure 2.3, process in each layer on different machine communicates directly with its counterpart by Protocol Data Unit (PDU). A header contains protocol information and user information is encapsu lated with the data provided by the upper layer for each PDU. It is the dialog across the peer processes in the parallel position. However, in order to support such a para llel communication, each lower layer will

PAGE 48

35 received the entities calle d Service Data Unit (SDU) from the upper layer and encapsulate the SDU with its header as th e supporting informati on for the operation in this layer. This is a vertical communication between layers. Figure 2.3 OSI reference model. The OSI reference model is a framework for any further developments on the protocol set for different environments. The de scriptions about the seven layers are listed below: Physical Layers The physical layer is responsible for the low level transmission of each bit of the data. It includes all the electrical and m echanical hardware specifications for the

PAGE 49

36 transmission of signals. The wire materials, connection interfaces, radio frequencies, and signal modulations are all spec ified with in this layer. Data Link Layer The data link layer provides the transfer of frames between two different nodes. It includes the procedure to tran sfer the information by blocks indicates the boundaries of each frame. Usually the information encapsulated into the frame in the data link layer includes the flow control and addressing info rmation, as well as the cyclic redundancy check bytes. The correction of transmissi on error within the frame is upon these additional bytes. The error correction is especially important for a high transmission error environment. The flow control and address provides the functions such as Medium Access Control (MAC) so the shar e of medium could be handled. Network Layer The transmission of information through one host to another host over one or more networks will be handled in this layer. Usua lly a hierarchical addr essing scheme is used to transmit the information over the global netw ork. Appropriate routing services are also provided here in order to deliv er the packets from the origin al network to the destination network. The routing protocol is also res ponsible for the determination of the optimal path for packet transmission in order to mitigate the congestion and obtain better transmission efficiency. The error control is provided in this layer as well. Unlike the address in data link layer, which is a hardware-based identification, the address for this layer is a logical address. The network layer offers a broader coverage over global network. Transport Layer

PAGE 50

37 The transport layer is responsible for end-to-end transfer of data. The segmentation and reassembly of data from upper layers is processed in this layer. The connection between two ends is set up in this layer as well. Besides, the Quality of Service (QoS) is also provided here based on different connection requirements in order to provide the communication in a more reliable or cost effective mean. Session Layer The session layer manages the manner for e nd user to exchange information. For example, a half duplex dialog or full duplex operation can be assigned here. Other functionalities like ch eckpointing, adjournment, termin ation and restart procedure are also offered in this layer. Presentation Layer Presentation layer is in charge of the enc oding methods for data from the application layer. The machine dependent codes can therefore be converted into machine independent codes. For instance, an ASCII c oded file can be converted into an EBCDIC coded file. Different data encryption schemes are offered here as well. Application Layer The codes and protocols used in this la yer provide various co mmunication services. For example, File Transfer Protocol (FTP ) provides the service for file transfer. HyperText Transfer Protocol (HTTP) enab les the access of World Wide Web (WWW) documents. Other services offered in the application layer include virtual terminal (TELNET), name management (DNS), and mail exchange (POP) (See Appendix A).

PAGE 51

38 2.2.2 TCP/IP structure OSI reference model has presented a we ll regulated framework of the layered network structure. The model regulates all the functions a nd procedures of the network communications. However, it is a quite concep tual network model. A set of protocols to allow physical network comm unication under OSI reference model is still required. TCP/IP structure is develope d by DARPA research project to connect the networks from different vendors to provide several fundamental services over a wide area network. It later becomes a very successful network struct ure worldwide. In this section, a brief introduction about TCP/IP struct ure is discussed as an exam ple of network developments under OSI reference model. TCP/IP architecture usually contains 4 layers: application layer, transport layer, internet layer, and network layer. The mapping between OS I model layers and TCP/IP structure layers is in Figure 2.4. Figure 2.4 Comparison of layer definition be tween OSI model and TCP/IP structure. From the figure above, it could be found that the upper 3 layers in OSI model are concluded in the application layer in TCP/ IP structure. The tr ansport layer usually

PAGE 52

39 contains two major protocol s: TCP and UDP. TCP provide a more reliable mean to transfer packets. Also error recovery and flow control is allowe d by TCP. UDP, however, is a connectionless way to send packets. It is a less reliable but co st effective way to transmit data. The Internet layer is the same as network layer in OSI model. It is mainly responsible for packet routing. The networ k interface layer covers the physical data transmission over various hardware interfaces. However, in TCP/IP structure, layers are not as strictly defined as in OSI model so application layer is allowed to bypass the intermediate layers and directly send data units to network interface layer. As mentioned in OSI reference model. When the data units are passing to the lower layers, each of them will be encapsulated w ith some extra information regarding to functions in each layer. Appropriate operati on could therefore be performed during the delivery of packets from one host to anothe r, and from one end to another end. For example, as in Figure 2.5, the application data like HTTP or FTP reque st is first sent to the transport layer. The header depending on the service established (TCP or UDP) will be given in front of the application data uni t. The encapsulated data unit would then be forwarded to the Internet layer, an Internet Protocol (IP) header will again be attached with the data unit in order to assign destina tion of the data unit. Before the frame is actually transmitted, the Ethernet header and error checking code will be combined at the front and end therefore the fr ame could be delivered to the next node correctly. An inverse process will be performed at each of the intermediate node to deliver the information correctly to its destination. Appr opriate packet rearrange ment and request of retransmission of the lost or error packets wi ll also be handled in order to guarantee the correctness of the data transf er. The documents regulates the formats of TCP, UDP, and

PAGE 53

40 IP headers could be found in [ 19 ], [ 21 ], and [ 22 ]. Other protocols re garding to different services like Real Time Protocol (RTP) could also be found in other relevant RFC documents. Moreover, except for the networ k protocols develope d by DARPA, more network protocols have been developed by various vendors. Table 2.3 lists some of frequently used network protocols for each of the OSI layer. Table 2.3 List of network protocols. Layer Protocols Application Layer HTTP, FTP, SMTP, RTSP, POP, TELNET Presentation Layer SMB, XDR Session Layer SSH, NetBIOS, ASP Transport Layer TCP, UDP, RTP, ATP Network Layer IP, IPv6,DHCP, ICMP, X.25 Data Link Layer ARP, RARP, DCAP, IEEE802.11 Physical Layer T1, encoding met hods, signal modulation/demodulation Figure 2.5 Encapsulation of header and error check code into data units.

PAGE 54

41 2.3 Wireless Communications and Issues The early developments of the network communications are mainly developed for the wire-based environments. Not only becau se the wire communication is more friendly for signal transmission but also much simpler and secure. However, the wires significantly restrict the mobility of the ne twork communications and applications for communication. With the advances in computer technology, the need to use the electrical mobile devices increases drastically. Mobile devices like laptops, PDAs and cellular phones, as well as the mobile robots and sate llites, are required to communicate with each other wirelessly. Also, various industrial and military applications have immense needs to send information without using wires. Th erefore, the wireless communication has become a big step in network communications. The media used for wireless communications are typically radio, or sometimes optical signals. By using the modulation/de modulation methods on these signals, binary data could be transmitted over the air. Th e wireless communication could hence be implemented. Morse code has also been used extensively for manual operation using radio transceiver in early times. It played an important role for military communications during World War I & II. However, although both wired and wirele ss communications need to modulate and demodulate the signa ls, there are still some considerable fundamental differences between them fo r reasons mentioned in the following: Wireless signal is susceptible to noises and Electro-Magnetic Interferences (EMI). A higher error rate is expected in wireless communications.

PAGE 55

42 Wireless signal strength varies greatly from both different positions and time instants because of the EMI and the multi-path propagation. Transmission collision is hence difficult to be detected under wireless environment. The spectrum of wireless signal is restri cted comparing to the wired signal. So communication bandwidth and rate are limited and slower. The differences between wired and wireless network communications are mainly in physical layer and data link layer in OSI seve n-layer model (or network interface layer in TCP/IP structure). The upper layer protocols are mostly compatible in both network environments. In the following sections, some discussions will be made about the specific issues related to the wireless network communica tion. It also plays a very important role in the wireless network design for multi-robot testbed later. 2.3.1 Medium Access Control Protocol One and maybe the most significant influe nce of the wireless communication is the medium sharing technique. When two or more devices need to share the same medium for communication, a proper procedure needs to be followed in order to transmit and receive the data correctly in each session. The medium sharing techniques are defined in the network layer in OSI model as MAC pr otocols. The wired network communications uses collision detection to avoid the collis ion of transmission between multiple machines. However, since the transmission of wireless signal is susceptible to EMI and multi-path propagation, collision detect ion becomes impractical to implement. Therefore, the collision of packet transmission needs to be avoided in advance of the actual data frame. This technique is called collision avoidance.

PAGE 56

43 A widely used medium sharing protoc ol for wired network communication is Carrier Sense Multiple Access with Collis ion Detection (CSMA-CD). As shown in Figure 2.6, when host A starts to send a frame to host B while it detects no activities over the medium. After a short period of time host B also transmit a frame to host A. The transmission frames from both nodes collide at 0 t 1t 2tt Because of the abnormal voltage from the frame collision hasnt reach host B, the medium status for host B is still available. After the propagation time the transmitted frame from A reaches host B therefore collision has been detected by host B. Also, when another transmitted frame also reaches host A so both hosts can detect that collision has occurred. After the collision has been detected, appropria te strategies will be applied in order for frame retransmission. Either the persistent scanning for the availability of the communication channel or the back-off algor ithm for rescheduling a retransmission will be performed. propt 12propttt However, the detection of the signal a bnormality is not practical for wireless signals due to the significantly variant in magnitude and low Signal to Noise Ratio (SNR). Moreover, the radio transmission has the uneven signal coverage issue. It is called hidden terminal problem as shown in Figure 2.7. Both of node A and C have only the limited signal coverage to the intermediate node B. Therefore, the transmission from either node A or C cant be detected by each other hence the collision cant be avoided by the CSMA-CD method.

PAGE 57

44 Figure 2.6 CSMA-CD. Figure 2.7 Hidden terminal problem for wireless network. Therefore, instead of sensing the collision of transmitted frame, a more conservative strategy must be applied for an effective medium sharing. In IEEE802.11, a modification is made for CSMA-CD. In order to prevent the data frame collision, a short

PAGE 58

45 signal for flow control is sent in advance. Wh en the collision happened to such a request, no response will be answered for the transmissi on request therefore the collision for data frame could be avoided. This technique is called Carrier Sense Multiple Access with Collision Avoidance (CSMA-CA). It is shown in Figure 2.8. Figure 2.8 CSMA-CA. Figure 2.8 shows the occurrence of the collision and how data frame is transmitted. In the beginning both channels attempt to transmit Request-To-Send (RTS) signal. The collision happens and therefore both hosts A and B will not response to the request. A back-off algorithm is applied on both hosts. After the waiting for a shorter back-off period, host A attempts to re-transmit the RTS frame and seize the control of the channel. Host B responses to host A with a frame Clear-To-Send (CTS) after RTS frame is received. The data frame is hence sent from host A to host B after RTS is recognized. With the time elapse during the data transmission, all of the other host will remain silent

PAGE 59

46 so the period is collision free. The channel will become available again after the transmitted data frame is confirmed by sendi ng the acknowledge frame ACK from host B. 2.3.2 Ad hoc and Infrastructure Topology Another important aspect for wireless networks is the network topology. The wireless network has great mobility compari ng to the wired network. Therefore it has been vastly used on mobile devices. As a re sult, the dynamic reallocation of the network nodes leads to the requirement for more dynami c connections in the network. Traditional network topologies for wired network communica tion are somewhat restrictive in such an environment. In this section, the network interconnections and developments for the dynamic topologies will be briefly introduced. The interconnections betweens network nodes in a network could be divided into two categories: physical connect ion and logical connection. Physical connections means two nodes are locally connected to each ot her by physical hardware such as Network Interface Card (NIC), cables, sw itch or router. It is a hard ware based connection. Logical connection means two network nodes are linked to each other by various network protocols as they are in the same network (L AN). It is a software oriented manner. Two or more nodes at different geological locati ons could logically be located in the same network. Protocol such as Virtual Private Network (VPN) is an example to provide a logical connection. Various wi red network topologies could be found in Figure 2.9. The links could be either physical or logical.

PAGE 60

47 (a) (b) (c) (d) (e) Figure 2.9 Network topologies. (a) mesh topology, (b) star, (c) bus topology, (d) ring topology, and (e) tree topology. The traditional network topologies are basi cally the centralized architectures. All the packets usually wont be transferred dir ectly between two nodes. The transmission is centralized so all the packets will be sent to a leading node as the network hub first. However, for the wireless mobile environm ent, the communications sometimes need to be performed in a decentralized mode when the global signal coverage for the hub is not possible. Instead of using any fixed network topology as either one in Figure 2.9, the network topology is required to be more dynamic. Therefore, the network topology ad hoc mode has been used in the wireless m obile communication environment. In ad hoc mode, transmission could be performed between different nodes directly without using an intermediate access point. It is also known as peer-to-peer network. Moreover, this peerto-peer network topology is mostly used for mobile wireless network. Therefore,

PAGE 61

48 sometimes it is known as the Mobile Ad hoc NETwork (MANET). The official charter for MANET could be found in [ 23 ]. The MANET is usually implemented under the environment with following characteristics: The network topologies could cha nge rapidly and unpredictably. The network is decentralized and la ck of prefix allocation agency. The devices running in MANET are usua lly power limited and equipped with lower processing capability. The MANET provides the network some advantag es such as the scal ability, mobility, and dynamic communication environment. It could so lve the issues when the reallocation for network nodes occurred frequently. Howeve r, the MANET is still under development. Only a few experimental network standard s have been proposed for MANET. These standards mostly provide p acket routing services. Some reference documents for experimental packet routing protocols can be found in [ 24 ] ~ [ 27 ]. There are still a number of technical challenges that needs to be overcome such as addressing, packet routing, network security, and QoS for MANET. Also, characteristics for different types of network links such as standalone ad hoc network or hybrid ad hoc network could vary greatly. Nevertheless, MANET could provide a quite attractive communication solution for the wireless multi robot systems. In addi tion, the communication network proposed in this thesis is a simple experimental network for a semi-MANET communication environment. 2.4 Communication Performance Another important aspect in order to understand the network protocols is the analysis for the communication performance. Performing such an an alysis could provide

PAGE 62

49 a better understanding for the network capabili ty and reliability. Any revision for the existing protocols in order to improve the ne twork performance could also be verified theoretically due to su ch analysis. In this section, some of the fundamental network measurements will be briefly discussed. More terminologies about some concepts of the network performance factors could be found in [ 28 ]. 2.4.1 Bandwidth For the communication terminologies, this term is obscure. It has two different meanings. The first definition is the width of the band of frequency used to carry data. This will usually affect the communication rate since the carrier with higher frequency has larger density for data. For example, the bandpass communicati on is usually faster than the baseband communication. However, th e concept addressed here is the second definition, which is the communication rate for data transmission over physical carrier. This is determined in the physical layer. Different modulation/demodulation techniques, hardware specifications and si gnal carriers used decide th e maximum capability of the transmission. For example, the maximum ba ndwidth for the tradi tional dialup network using MODEM over the phone line is usually 65536bps. The bandwidth for IEEE802.3u is 100Mbps. The bandwidth for IEEE802.3ae is 10Gbps. The protocols adopted in the upper layer will not have any influe nce on the communication bandwidth. 2.4.2 Transmission Loss The transmission loss will deteriorate the actual transmission rate and quality. It occurred for various reasons from hardware ba sed factors to protocol based factors. For instance, a low SNR environment could lead to a higher transmission error. This usually results from either longer transmission distance between relay nodes or low quality transmission medium (higher EMI or worse cab le quality). Also, the protocols adopted in

PAGE 63

50 the network design could also have impact on the transmi ssion error. The more bits contained in each individual frame, the more possibility for a transmission error could occurred within a frame. Too much routing co uld also make the transmitted packet to be obsolete. Two different circumstances will happen as the transmission loss: frame loss or frame error. When a frame is lost during the transmission, no response for the frame acknowledgement will be received therefore it could be detected. On the other hand, the frame being received incorrectly could be found by the Cyclic Redundancy Check (CRC) code attached at the end of each frame or packet. When either of the communication abnormalities happened, an appropriate arrange ment for re-transmission will be made so the problem could be corrected The protocol used for frame re-transmission is called Automatic Repeat reQuest (ARQ) protocol. Th ree different types of ARQ protocols are used for frame re-transmission: stop and wait ARQ, go-back-n ARQ, and selective repeat ARQ. Different efficiencies and computational requirements are needed for different ARQ protocols. The details for ARQ protocols used will be further discussed in chapter 4. However, the efforts to guarantee the data is being transmitted and received correctly usually need more overhead, which cause a lower effective transmission rate. Some tradeoff must be made between the quality and quantity for communications. 2.4.3 Throughput The communication bandwidth couldnt present the actual communication rate for user data. The transmission loss, repeat request, and the wait for establishing the connection could all reduce th e rate. Therefore, the ne twork throughput makes more sense to present the real communication perf ormance. According to the definition in [ 28 ], the network throughput is define d as the maximum rate at which none of the offered frames are dropped by the device. The throughput of a network is mainly determined at

PAGE 64

51 lower layers (physical layer, data link layer, and network layer). The factors which have influences on the throughput include ARQ protocols, MAC pr otocols, network congestion and packet routing protocols. 2.4.4 Latency One of the very important issues for ne twork performance is the latency. It is defined [ 28 ] as the time interval starting when the last bit of the input frame reaches the input port and ending when the first bit of the output frame is seen on the output port. The network latency also represent the time it takes from the time data is requested to the time it is arrived. Sometimes, it is also used to define how much control we have for the network. Not only the latency but also the jitter ing, which is the variation of the latency, are important when we want to measure the perf ormance. It is a fairly critical issue for the distributed control problem or multi-agent systems. One of the methods to mitigate such problem is QoS, which normally reserv e certain amount of bandwidth for specific services. Also, a real-time system to guara ntee a minimum delay of system operation for MRS is another solution. In the other hand, the communication perf ormance is also highly dependent on the hardware specifications. The hardware design is another critical point in this thesis. Different hardware architecture has influen tial impact to the performance on not only communication but also processing and mane uvers capabilities. The next chapter will therefore focus on the hardware specifications and their restrictions.

PAGE 65

CHAPTER 3 WIRELESS MULTI-ROBOT TESTBED As the discussion in the previous chapter, it could be found that the development of the network communications is based on th e hardware specifications and mission objectives. Protocols used in the lower layers are highly dependent on the physical signal types and modulation/demodulation techniques adopted by the hardware. Therefore it has become fairly important to understand the phys ical hardware used for the experimental testbed before the discussion of th e dedicated network environment. Meanwhile, the understanding for the concept of system ope ration is also critical before designing the network. To appropriate ly identify the requirements for the system operation is the essential to support the operatio n of the testbed in a cost effective way. The design of the network protocols for upper la yers is mainly determined by the mission requirements. The services provided by the ne twork and their performances are the key factors which lead to a satis factory network development. Therefore, all the hardware facilities and various sub-systems, for example, positioning system and communication system, will be mentioned in this chapter befo re the discussion of network protocols. 3.1 Hardware Architecture of the testbed The wireless multi-robot testbed is com posed of several sub-components including mobile robots, positioning system, the opera tional area which provide the region for its operation. The hardware architecture of the testbed is shown in Figure 3.1. The mobile robot is controlled by the independent contro ller on each of them. The positioning system determines 3 dimensional positions of the LEDs and orientations of the LED-mounted 52

PAGE 66

53 objects by using eight cameras mounted as show n in the figure. A central computer is responsible to provide inform ation for localizing all the obs tacles and moving objects via the communication system. Also any user i nputs to command the system can also be transmitted though the central computer. The de tail description on each sub-system will be provided in the rest se ctions of this chapter. Figure 3.1 Hardware architecture of the testbed.

PAGE 67

54 3.2 Wireless Mobile Robot The core components for the testbed are the mobile robots, which are considered as intellectual (and autonomous) agents in the system. Due to the physical application requirements, the robots in the testbe d must have the following features: Mobility A mobile robot system can be used to va lidate the concepts and algorithms for path planning, vision servo control, leader-follower problem. It is also widely applicable for many autonomous vehicle control applications So the appropriate maneuvers to move the robot itself are required. Wireless communication The control of the robot n eeds the communications. Hence an adequate transceiver to send/receive control commands, transmit a ll the required information back and forth between different vehicles is needed. Ho wever, for a mobile robot, wires could enormously restrict the mobility. Therefore, the wireless communication is the essentials to remove this constraint. Decentralized control A multi-agent system needs multiple processing threads/controllers to control different physical agents. Under some restricted circumstances, a centralized control of multiple robots is not feasible. Therefore, the design of the testbed requires a controller or processing facility on each of the individual robot. Scalability Mobile robot can dynamically change its loca tion, therefore, the fluctuation of the number of robots during the operation is assumed. For many multi-agent systems, the

PAGE 68

55 scalability is considered during the operati on. So the functional robots in the testbed could be added or removed any time in order to keep the flexibility of the operation. The robot WALKER is developed here for the MAS research based on the considerations discussed above. The WALKER acronym stands for Wireless Autonomous Linux-based Kinematic Expert Robot. It is shown in Figure 3.2. Figure 3.2 WALKER for multi-robot testbed. The dimension for the WALKER is 8.75 "6"7" ( L WH ). It is a modulized design comprising various modules so the main tenance could be easier. Also it could be more flexible to any further development si nce the upgrade of the hardware can be done by simply replacing individual modules. The WALKER has four different modules to support its operation: power module, pro cessing unit, communication module, and hardware control unit. The inter-connections between each module is shown in Figure 3.3. The hardware used for each module will be mentioned in the following subsections.

PAGE 69

56 Figure 3.3 Block diagram for modules on WALKER. 3.2.1 Power Module The power module is composed of a lithiu m-polymer rechargeable battery and the DC/DC power supply for PC/104. The output vo ltage for the lithium-polymer battery is 14.8V, the power capacity is 4400mAh. The battery connects to the power supply and converts the power to the regulated 12V and 5V outputs to support th e requisite operation on the robot. The pictures for the power module are in Figure 3.4. The power module could support the system operation for roughl y 8 hours when the system is idle under Linux. (a) (b) Figure 3.4 Hardware pictures for power module. (a) lithium-polymer rechargeable battery, and (b) PC104 DC power supply.

PAGE 70

57 3.2.2 Communication Module There are two different channels for the wireless communications. The differences for two channels are listed in Table 3.1. However, this thesis only discusses for the low bandwidth communication channel. The reason fo r this choice will be mentioned in the next chapter. Pictures for the ha rdware are shown in Figure 3.5. Table 3.1 Comparison of different communication channels. Wireless modem IEEE802.11b PCMCIA card Modulation method FSK BPSK/QPSK/CCK Radio Frequency 900 MHz 2.4 GHz Bandwidth 38400 bps 11 Mbps Signal range (outdoor/indoor) 300 / 1000 300 / 1000 Power Consumption 0.225 Watt 1.235 Watt (a) (b) Figure 3.5 Hardware pictures for communi cation module. (a) Low bandwidth channel (wireless modem), and (b) High band width channel (IEEE802.11b PCMCIA card). 3.2.3 Hardware Control Unit The hardware control for the WALKER cont ains the controller for motors, input D/A (Digital to Analog) and output A/D channels for vari ous sensors. It is the intermediate interface between processing unit and the phys ical hardware. The hardware control unit uses a Motorola 68HC11 based board with 32 K SRAM as the controller. The detail reference for the board can be found in [ 29 ]. Also, the WALKER currently uses two servo motors to control its motion and a pair of infrared sensors for obstacle

PAGE 71

58 avoidance. The servo motor is controller by Pulse Width Modulation (PWM) signals, and the IR sensors needs to be converted by An alog to Digital Converter (ADC). However, more optional facilities like camera and power management system could also be controlled by this module. The pictures fo r the hardware control unit can be found in Figure 3.6. (a) (b) (c) Figure 3.6 Hardware and hardware control unit. (a) servo motor, (b) infrared sensor, and (c) MC68HC11 controller. 3.2.4 Processing Unit The processing unit is the kernel of the r obot. It greatly enhan ces the capabilities and features for the robot. The processing un it turns the robot into an intellectual and autonomous agent instead of simply a mobile robot. The processing unit is responsible

PAGE 72

59 for performing the required computations for control actions and the cooperative behaviors, as well as the requisite network communicati on with the other agents. The processing unit uses PC/104 board for the robot. The specifications for the processing unit are in Table 3.2. Figure 3.7 shows the picture of PC/104 module. Table 3.2 Specifications for processing unit. Specifications Manufacturer / module number Kontron / MOPlcd6 Processor Pentium MMX 266MHz Memory 64Mbytes Disk 512 Mbytes CF card Operating System White Dwarf Linux 2.0 Figure 3.7 PC/104 processing unit. 3.3 Positioning System The localizations of the mobile robots a nd all the possible obstacles for many of the multi-robot testbeds are fairly computationa lly expensive. The onboard camera with the simple image processing technique like feature points extraction is usually required with the sharing of the information between each robots for localization. However, for the wireless multi-robot testbed discussed in this thesis, a computationally cost effective

PAGE 73

60 positioning system is used for the localization. The positions for the objects are provided in the system by the PhaseSpace position measurement system. Eight cameras are mounted on the outer aluminum frame in orde r to cover the whole operational area. The LEDs are required to be attached on the targets so that the ca meras can track the positions and orientations of the LEDs and hence the ta rgets. The block diagram of the positioning system can be found in Figure 3.8. Figure 3.8 Block diagram of the PhaseSpace positioning system. In Figure 3.8, the cameras are connected to the hub by an USB connection. The PC is running under Linux operating system. Also, the hub has the connections to the LEDs via proper adaptor interfaces. Different LED s could be recognized by the system by sending the flashing signals at specific freque ncies so each LED could be identified with its specific number. The positioning system could generate the 3-D data for all the tracking points at a rate of 220 fps. It is, however, enough for detecting the robot motions.

PAGE 74

61 The pictures of positioning system are shown in Figure 3.9. The detail discussion about the determination of 3-D position of LED under the PhaseSpace system could be found in [ 15 ]. (a) (b) (c) (d) Figure 3.9 Pictures of hardware for PhaseSpace positioning system. (a) hub, (b) camera, (c) LED board, and (d) LEDs. 3.4 Operational Area By using the positioning system, the object s could be properly located within certain range. However, due to the physical restrictions of the positioning system and available room, the operation can only valid with in a limited area. Therefore, in order to enhance the operation of multiple robots and ensure the vehicle operations are valid, a platform is made for restraining the region of the motions and placing the obstacles for different research scenarios. The floor and several blocks used for the platform and obstacles are made of foam. The figures in Figure 3.10 include the geometry of the testbed for the platform, obstacles, and the positioning system.

PAGE 75

62 (a) (b) Figure 3.10 Geometry of the testbed. (a) positions of the opera tional area, and (b) projections of the visible ra nge for the positioning system. From the description in Figure 3.1 above, it could be observed that a communication network is required to suppor t the operation of the wireless MRS testbed by using the communication module mentioned in section 3.2.2. The next chapter will describe a dedicated protocol suite using th e low bandwidth channel to enable the intercooperation between various robots as different individual agents.

PAGE 76

CHAPTER 4 PROTOCOL SUITE This chapter describes the protocol desi gn for the self-configurable network used for the wireless multi-robot testbed. The netw ork interfaces and protocol suite mentioned in this chapter are the communication implementations on the experimental multi-robot testbed mentioned in the previous chapter. A complete network environment is outlined in this chapter. Some considerations and hard ware specific issues are addressed in section 4.1. In order to simplify the communicat ion scenarios and problems, only LAN is considered here. Section 4.2 introduces a de dicated LAN for the testbed. A few of the protocols have been specified for this laye red network to perform the operation of the network communications. This will be di scussed in detail in section 4.3. 4.1 Limitations and Requirements Chapter 3 has describes the equipments us ed in the wireless multi-robot testbed. Before the attention is focused on the design of network environment, explicit statements of the network restri ction are provided. The most important principle for the network design for the testbed in this thesis is the minimal realization for a required ne twork communication with better hardware compatibility. This means the optimization of the use of the onboard resources including the power and computational needs, for example, memory and minimal processor required. In other words, to support the syst em operation in a more efficient manner in terms of the power and computational us es is the highest design principal. 63

PAGE 77

64 The first step for the design problem is th e understanding of all the constraints and requirements. The limitations of the wireless multi-robot testbed are hence listed below: 1. Power consumption on the mobile robot needs to be minimized. So a transceiver with lower power consump tion used on the robo t is preferred. 2. Lower bandwidth channel (wireless modem) is the preferred device for the high level data transmission such as a control command because the lower required bandwidth, lower power consum ption and better signal sensibility ( and reliability). 3. Higher bandwidth channel (PCMCIA wireless Ethernet card) is the preferred device for a low level data transmission such as the image data. A mixed use of both low and high bandwidth channels can make the communication to be more robust and e fficient. However, the transmission of low level data is not in the scope of this thesis. The requirements to support the system operation are specified below: 1. Command refresh rate for mobile robot control could be less 5 Hz since the transient response for a mechanical mechanism is relatively slow. Therefore the data rate of 38400 bps is enough for robot control 2. The operation requires the inter-opera tion between agents. Therefore the application layer Agent Communication Language (ACL) is also needed to be defined for the network communication to handle the conversation in lieu of simply low level data transmission.

PAGE 78

65 3. The scale of the fleet in the multi-robo t testbed is restrictive. Hence in order to simplify the network, only a LAN is discussed here. The packet routing protocol is not under consideration in this thesis. 4. The network topology is comprised of one leader and multiple followers. 5. In order to enhance the autonomous ope ration and decentralized control, an automatic configuration of the topology is required by the system so the robot doesnt definitely need the comma nd offered from the leader to make such a change. This procedure will be covered in the next chapter. 4.2 Local Area Network Architecture A very important assumption made here is the scale of network. As the most frequently used topology, the tree topology needs a hierarchical ad dressing technique and ability to forward the information to its de stination. This also includes sophisticated algorithms and protocols for data routing. Ther efore, a significant simplification could be made in the design problem by confining the network to only LAN. Moreover, some of the intermediate layers could also be skipped since only a single LAN has been considered here. The difference of the LAN and Wide Area Network (WAN) can be found in Figure 4.1. (a) (b) Figure 4.1 Comparison of the interconnections of different networks (a) LAN, and (b) WAN.

PAGE 79

66 As it is shown in Figure 4.1(a), all of the nodes in LAN 1 could be accessed using the local addresses in the LAN. For example, the local address for node A and node B in Figure 4.1(a) are x and y respectively. The two nodes c ould therefore reach each other by using their local addresses within the same network. The information routing under such case is not going to be required excep t in the infrastructure mode. The frame transmission from node A to B will go though the intermediate node C as the local network hub under infrastructure mode. Otherwise, no forwarding of information will be required in ad hoc mode under local network. However, as shown in Figure 4.1(b), the addresses within each LAN are only valid for the other nodes in the same LAN. A direct connection cant be made across different LANs. The only exception is th e specific bridges used for interconnection between different LANs. Therefore, in order to ha ve the access for each node in the WAN, a global address needs to be given to each node. Usually, a hierarchical addressing method is considered in this case to improve the scalability of network. For example, a local address for node A in LAN 1 is x and the address for node B in LAN 2 is y However, since node A and B are not in the same netw ork, they cant reach each other by using their local addresses. The numbering on each of the networks is required to address a global identification. Therefore, as the topology in Figure 4.1(b), a given global address for the node A is 1. x and for node B. When the transmission occurred between these two nodes, the information will first be tran smitted from node A to the local network leader B, then the received packet will be forward to the other ne twork leader node D in LAN2, another information routing will happene d then from node D to node B in LAN 2. As it could be observed here, the addressing is used not only for the medium sharing 2. y

PAGE 80

67 purpose but also for the information routi ng. Nevertheless, as it has already been discussed in chapter 2, the da ta link layer and network layer (in ISO model) are used to handle the problems. Another critical packet r outing issue also happened when signal cant globally cover the physical range of a network. When the communication needs to be made from one node to another node in the network betw een two far ends, the determination of the forwarding route becomes sophisticated, especi ally when the network is under ad hoc mode or the hybrid of ad hoc and infrastruct ure mode. Although the routing protocols for wired network environment are already we ll developed, the networking in a mixed network environment of different communicatio n modes is still somewhat limited. The proper routing protocol used highly depends on the type of the network environment. The wireless multi-robot testbed discussed in the previous chapter is comprised of several mobile robots. The number of robots is however limited. Since the objective for the network communication is only for the co llaborations between a limited numbers of small robots, a more cost effective manne r to implement the network communication here will be to consider the environment to be only a single LAN. Iin the next section, the network and transport layers in the dedi cated network are ignored under such an assumption. As a result, both the computationa l cost and complexity of the network can be significantly reduced. 4.3 Protocol Specifications The above discussion provides an ex plicit explanation of the network communication considerations and infrastructure. The deta il specifications of the dedicated wireless network are described in this section.

PAGE 81

68 First, the network layers of the wireless communication ar e shown in Figure 4.2. In Figure 4.2, the protocols in clude the pre-programmed prot ocols on the wireless modem and specified protocols. The pre-programmed protocols are hard coded on the wireless modem and the header for the raw transmission frame cannot be accessed. The preprogrammed protocols covers two ISO layers by using Modbus protocol [ 30 ] as the data link layer protocols, a nd a specific protocol [ 31 ] for network layer. The network layer protocol using vendor identific ation number, channel, destin ation address, and address mask to provide the networking features and filtration. However, the hard coded network layer protocol is basically a Point-to-Point Protocol (PPP) which is designed for a centralized basis communication scenario. As the decentralized communication need, the provided protocols cannot satisfy the requi rements for the operation. Moreover, the header for each frame cannot be captured or modified in a software manner. Hard coding the protocols into the firmware on wirele ss modem provides great convenience for the end user to control various devi ces using serial port communi cation. However, the lack of the flexibility with such design method signifi cantly restricts the possibilities for further network programming and control. Therefore, in order to obtain the full control on the robot network communications, the pre-pr ogrammed protocols are ignored, and the additional protocols are specified in the thesis instead. The specified network communi cation structure contains tw o layers, data link layer and application layer. As discussed in the ch apter 2 for the layering in TCP/IP structure, it is feasible to skip over the intermediate la yers and have the network services directly encapsulate the application layer SDU into th e data link layer frame. Hence, due to the

PAGE 82

69 consideration mentioned in prev ious section, the global addre ssing is ignored and the data link layer could have the direct access to the application layer. Figure 4.2 Dedicated wireless network layers 4.3.1 Data Link Layer Protocol The specified data link SDU contains th ree fields: source address, destination address, control information, and its payloa d. The purpose of this layer and the frame header is for the medium access and flow c ontrol purposes. The design of this frame is referenced from High-level Data Link Cont rol (HDLC). However, the HDLC is also a PPP. Therefore the destination address is a dded to the frame for the wireless network

PAGE 83

70 medium access. The detail explanations for the meaning of each field are defined as below: Source address It is the address or ID for the frame transmitter. This is an 8 bits long field. 255 devices can be addressed. Destination address It is the address or ID for the frame recei ver. This is an 8 bits long field. 255 devices can be addressed. Control The control field is responsible for esta blishing or releasing the connection for data transmission, as well as the f unctions of the frame retransmission and acknowledgement for the reception of pr eviously transmitted frame. The ARQ protocol is also implemented by the control fiel d. It is an 8 bits long field. The format of this frame is defined in Figure 4.3. The control field used here is similar to the HDLC procedure [ 32 ]. Three types of frame are used here: information frame, supervisory frame, and unnumbered frame. Information frame are used for the transmission of datagram from upper layer. NS represents the send sequence number for the frame, and NR represents the response sequence number. The bit indicates the direction for the data tr ansmission. The supervisory frame is for the flow and error control. The ackno wledgement (ACK) frame and negative acknowledgement (NAK) frames could be us ed to confirm the correctness of the transmission. It comes with only response sequence number at the end since no data is contained in this frame. The unnumbered fr ame is used to esta blish or release the / PF

PAGE 84

71 connection. Three different modes of conn ection could be established by unnumbered frame: Normal Response Mode (NRM), Asynchronous Balanced Mode (ABM), and Asynchronous Response Mode (ARM). Figure 4.3 Bit-wise format of the control field 4.3.1.1 Link management The normal response mode is the only mode used here for the network communication. It is a synchronous transmi ssion mode. So the secondary node can only transmit when it is instructed by the primar y mode. It is used under the half duplex environment. The transmission mode is less efficient. However, it is computational inexpensive since less buffer is needed and no data rearrangements needs to be made. The transmission procedure for NRM is in Figure 4.4. In Figure 4.4, the first two entities in th e square bracket are source and destination address. The third entity is the type of the frame. I is the information frame. The fourth and fifth entities are and NS NR The acknowledgement of the received frame can be piggybacked with the transmitted frame sequence number. Also, when the transmission error occurred during the transmission, no acknowledgement will be sent. The error will be detected if no acknowledgement is

PAGE 85

72 received for the transmission for a period. Th e error control mechanisms in the wireless network will be described in the following sub-sections. Figure 4.4 Normal response mode 4.3.1.2 Forward error correction Error control is important to guarantee the correctness of signal transmission. Usually in a communication cha nnel, two error control methods will be used in order to improve the reliability of transmission: Fo rward Error Correction (FEC) and feedback error correction (ARQ). The FEC is, by usi ng redundant information attached in the transmitted frame, the transmission error can be detected and corrected. However, the FEC cant guarantee the corre ctness of transmissions. In the wireless network environment discussed here, the FEC is automatically provided by the CRC field in pr e-programmed Modbus protocol. No further FEC is made

PAGE 86

73 in the protocol described in section 4.3.1 However, the FEC provided by the firmware will only drop the received frame once error is detected. No negative acknowledgement will be sent in the network channel. Therefore, a feedback error correction is also offered in this network. 4.3.1.3 Feedback error correction The feedback error correcti on in data link layer is cal led ARQ. As described in chapter 2, it is a mechanism to improve th e correctness of transmissions. For the ARQ protocols, there are three frequently used t ypes: stop-and-wait ARQ, go-back-n ARQ, and selective reject ARQ. For the latter two ARQ methods, the transmission efficiency comparing to stop-and-wait ARQ are better. However, the cost is their relatively expensive computational needs since more buffers and data rearrangements are required. So the stop-and-wait ARQ is selected for th e dedicated network here. The ARQ methods are shown found in Figure 4.5. (a) (b) (c) Figure 4.5 ARQ methods. (a) stop-and-wait ARQ, (b) go-back-n ARQ, and (c) selective reject ARQ

PAGE 87

74 4.3.2 Agent Communication Language Another appeared problem after two agen ts could have communicated with each other: How could the agent inte rpret the binary data into meaningful context? A common language for two homogenous/ heterogeneous agents is required in order to translate low level data into useful high level informati on or control command for different agents. The syntactic and semantic definitions for such a language hence have to be regulated. This task is defined in the presentation layer and application layer in ISO model. On the other hand, since the dedicated network environm ent has skipped over some intermediate layers, the compensation for some of these functions could have been made in the application layer. The Agent Communicati on Language (ACL) has therefore become another critical protocol for the multi-agent environments. A few of the agent communication languages have been developed to satisfy the interoperability need between agents. It in cludes both semantic and syntactic definitions for the language. Two major ACLs have been developed and widely discussed: Knowledge Query and Manipul ation Language (KQML) [ 33 ]~[ 34 ] and FIPA-ACL [ 35 ]. Usually the ACLs are comprised of different performatives with va rious lengths of the arguments and contents, depends on their reque sted actions. The information contained in each ACL is often encoded in the ASCII format, while the binary message encoding method is also used once in a while. Howeve r, the introduction of the ACL is quite lengthy and beyond the scope of this thesis. Therefore, a detail introduction for the developments and principles of ACL will be skipped over. The proposals and specifications for various ACLs could be found in [ 33 ] ~ [ 35 ]. This section will only focus the attention on th e dedicated ACL proposal.

PAGE 88

75 The ACL proposed here is composed of a couple of different components. Table 4.1 lists the essential compone nts as the syntax of the la nguage used in this thesis. Comparing to the other generic ACLs, this ACL proposal is restricted to a specific use for some fundamental operations for the multi-r obot testbed mentioned in the previous chapter. For example: the query for the IR se nsor reading, sending a command to request a motion on another robot, or monitoring the e ligibility of a robot in the network. The detail semantics of the la nguage are explained below: Table 4.1 Proposed agent communication language arguments Bits value semantics meanings length 8 0x01 ~ 0xFF n/a indicate the length of the ACL message sender 8 0x01 ~ 0xFF n/a message sender ID receiver 8 0x01 ~ 0xFF n/a message receiver ID 0x01 ask ask for specific information 0x02 tell send the requested information performative 5 0x09 unregister inform the other agent of leaving the network 0x01 eligibility value for eligibility 0x02 motor speed value for motor speed 0x03 IR reading value for proximity detection 0x04 signal strength values indicate wireless signal strength 0x05 power measurement for the remaining power 0x06 camera signal to turn on/off the embedded camera ontology 3 0x07 connection information about the connection 0x02 data out service-request 2 0x03 data in the information flow of the message in-reply-to 3 0x00 ~ 0x07 n/a ID of a specific process for the message response regarding to reply with 3 0x00 ~ 0x07 n/a message ID regarding to the message itself Length

PAGE 89

76 The total length of each ACL message is not fixed. So this field indicates the total length of the transmitted ACL message in order for the receiver to receive the message correctly. This is an 8 bits field. Sender This field specifies the source of the info rmation. Comparing to the source address in the data link layer, this ID is more fl exible. It could be determined by the physical address or another specific ID due to the change of the network topology. This identification could be either the source addre ss in data link layer pr otocol or a logical address. The range of the address is from 0x01 ~ 0xFF. Receiver This field is similar to the sender mentioned above. The value indicates to receiver of the information. The range is also from 0x01 ~ 0xFF. Performative The performative assigns the type of communicative act [ 35 ]. This field is 5 bits long, so the value can vary from 0x00 ~ 0x1F. The values which have already been assigned could be found in Table 4.1. This field indicates all the control command and information exchange for the conversion. Ontology This term is used in conjunction with performative as an auxiliary statement to express and interpret the message. Here this field is used to specify the type of the information either requested or sent. Service-request

PAGE 90

77 This is used here to indicate the dire ction of the information flow as another auxiliary statement for some implicit performatives. In-reply-to This field indicates the ID of a specific process for the message response regarding to. Reply with This field assigns a specific number for th e identification when any further process and response is being made. According to the above explanation, when we want to acquire the eligibility value from robot 4 to robot 8, we can send an ACL message as shown in Figure 4.6(a). The response for such an information query is in Figure 4.6(b). In Figure 4.6(a), no content is included in the message. In the replied me ssage, only one value is needed for the eligibility query. (a) (b) Figure 4.6 Examples of ACL messages. (a) requ est for eligibility value, and (b) response for eligibility query. According to the explanation of the phys ical meaning for each field and the example, most of the messages transmitted by using the ACL are expected to be short. The overhead of each message then would occ upy a lot of portion for the whole message length. This fact would reduce the performa nce of sending useful measurements across

PAGE 91

78 the robots. In order to improve the effi ciency, instead of sending the ASCII coded message, the message in the ACL used here is encoded in a bit-wise manner. The structure of the ACL message could be found in Figure 4.2. A significant advantage could be made for the bit-wise encoding method. As of the example found in Figure 4.7, the robot of 16 sends a message to the robot of 8 a command of moving forward with the speed 100 on both wheels. The ASCII encoded message is 129 bytes long. The time spent to transmit this message for a 38400 bps wireless modem is 0.0268 sec. However, the same message encoded in a bit-wise method is only 7 bytes long. The time required for transmission is only 0.00146sec. So the bitwise encoding is only 5% long comparing to the ASCII encoding. The similar results could also be found in some relevant efforts under both KQML [ 36 ] and FIPA-ACL [ 37 ]. (a) (b) Figure 4.7 Comparison between different encoding methods for ACL. (a) ASCII encoding, and (b) bit-wise encoding. Figure 4.8 shows the performance comparison between the ASCII encoding and bit-wise encoding. It coul d be observed that the pe rformance has significant improvements for the shorter message since the overhead for a shorter message is a heavy burden comparing to the longer message. Nevert heless, the efficiency for longer message will still be better comparing to the ASCII encoding method.

PAGE 92

79 Figure 4.8 Comparison of the performance on different message encoding methods This chapter provides a network envi ronment for the communication needs between each robot. However, a mobile robot always changes it physical locations and sometimes needs to change the network topol ogy corresponding to different incidents or mission requirements. Chapter 5 will propos ed a procedure for the MRS testbed to configure their logical connections dynamically in order to gain the flexibility during the system operation.

PAGE 93

CHAPTER 5 SELF-CONFIGURABLE TOPOLOGY As in the centralized control scenario, a ll of the controlled devices connected to a single command center. The network topology is fixed and therefore less flexible. Any change of the connection or topology must be made from the leader in the network. However, for each single robot, the lack of the ability to configure it self in response to any event occurred regarding to itself reduces the autonomy on such a robot. As an agent in multi-agent system, the agent generall y should have the following properties: autonomy, flexibility, and to own the thread to control itself. Therefor e, the ability to be autonomous to the environment including netw ork connections promotes a robot to an intelligent agent. a self-configur able network topology can also utilize the resources more efficiently and response to any incident fast er. Moreover, the failures on the members of the MRS will not obsolete the operation of the whole system. Hence the system could also be more robust. Each member in the system is also easier to be replaced and upgraded. In the section 5.1, th e use of an evaluation measurement is proposed to indicate the eligibility of each robot. The eligibility value could be determined by various mission objectives and multiple measurements on each singl e agent. Eligibility List (EL) can then be generated and broadcasted globally based on the collection of eligibility value on each available robot. Therefore, a dynamic task allocation method could be developed based on the awareness of eligibility on each robot The related discussion about the dynamic task allocation can be found in [ 9 ] and [ 17 ]. Section 5.2 explains the procedure of the self-configuration using the EL. Two differe nt operational modes are suggested to 80

PAGE 94

81 indicate the robot status as either leader or follower. The physical operations are demonstrated under various scenarios in the re st sections as the examples of the selfconfiguration procedure. 5.1 Eligibility List The topology for the network is determined by the eligibility of each agent. The MRS is designed for homogeneous/heterogen eous robots interoperation. For each different single robot, different capabilities and resources are assumed. For example, the remaining power, wireless signal strength, and the location of itself could all affect the eligibility to perfor m the allocated mission tasks. The modules and actuators embedded on each robot might vary as well. For instance, a robot with only IR sensors might need to be guided by another robot with a camera so it could become location awareness. Also, the robot with higher processing performance might take leadership of the whole robot team. Therefore, the election of the most elig ible robot to be the leading agent in the network can optimize the system performance. During the operation of MRS, an EL is generated by broadcasting the query to each robot. The EL contains two rows. The first row contains the ranking of th e eligibility for each robot, a nd the second row contains the eligibility value for each robot. After the EL is generated, the list will again be broadcasted to each of the robot. While the control of the network topology is still seized by the leading robot, the most eligible robot could be determined and configuration on the topology based on the avai lable eligibility information could be made. The selfconfiguration mechanism could also justify the unpredictable fault when the leading robot fails and replaces it w ith the next eligible robot. Th is self-configurable mechanism is discussed in the next section.

PAGE 95

82 5.2 Self-configuration A self-configuration mechanism is a critical request to take the system communication to an autonomous level. Th e network functions by scanning the network frequently from the leading robot, and th e sensing of a connection timeout by the follower robot. For each of the robot in the network, the capability to adjust itself individually can be added by such a mechanism. Therefore, the biggest difference for this network is the control of the network topology is decentralized. The flowchart of the selfconfiguration mechanism is shown in Figure 5.1. As shown in Figure 5.1, the operation of robot communication can be separated into two different modes: passive mode and active mode. When the robot works under passive mode, it waits for the connection re quest from the other robots. During this operation mode, the robot works as a followe r under the command of the leading robot which connects to it. However, when the connection request isnt received over a period of time, or the leading robot un-register fr om the network, a follower robot will assume the leading robot becomes invalid and eliminat e its existence from the EL. Then for each follower robot, an inspection fo r the eligibility list to determine the most eligible robot will be performed. The most eligible robot (w ith the highest eligibil ity value) will take over the lead of the network and enter the active mode. During the active mode operation, the leading robot will constantly scan the network to find the other available robots at a given rate. Once a more eligible robot is found, the leading robot will relinquish its privilege to control the other robots, retu rn to its passive mode and wait for the connection request from th e most eligible robot. The self-configurable network contains four different scenarios: network initialization and new robot joining, follower fa ilure, leader failure, and control privilege

PAGE 96

83 transfer. The execution of the procedure will be explained in the later sections as the simulation results for the network self-configuration mechanism. Before the demonstrations of these procedures, the test configurati on for this self-configurable network is explained in the next section first. Figure 5.1 Flowchart of self -configuration mechanism.

PAGE 97

84 5.3 Test Configuration The WALKER is a prototype for wireless multi-robot testbed. With the proposed hardware architecture described in chapter 3, multiple WALKERs are capable to work cooperative within the testbed. However, due to the research progress, only one WALKER is manufactured currently for the hardware assessment. Therefore, tests of the communication network in the following sec tions are performed by similar computer equipments. Detail test configuration incl uding wireless modem setting, used operating system, and software development t ool chains are described below. Table 5.1 lists all computers used for th e tests. The computer one is the PC/104 processing module embedded on WALKER. Othe r two computers are Intel PC and SUN workstation respectively. The purpose for using di fferent computers in the tests is also to validate the compatibility of the communica tion network between heterogeneous robots. However, all the tests are performed under Li nux environments. The kernels used in all three computers are under the family of Li nux kernel 2.4. (See Table 5.1) It is, nevertheless, not a real time kernel. The di splay during the executi on is shown in Figure 5.2. In Figure 5.2, the program shows the value 1 to present the empty fields in the eligibility list. Table 5.1 Computer confi gurations for the test. Computer 1 Computer 2 Computer 3 Processor INTEL Pentium 266 MHz (PC/104 board) INTEL Pentium 4 1.8GHz SUN UltraSPARCIIi 440 MHz Memory 64MB 128MB 640MB Operating System White Dwarf Linux 2.0 Debian 3.0r4 Debian3.0r4 Linux Kernel 2.4.29 2.4.27 2.4.27 Hard Drive/ FLASH Volume 512MB 10GB 9.4GB Compiler Version Gcc 3.2 Gcc 3.3.5 Gcc 3.3.5

PAGE 98

85 Figure 5.2 Display during the test. 5.4 Network Initialization The network initialization process is show n in Figure 5.3. In the beginning both robots are in the passive mode. However, when the first robot ge ts the timeout from waiting, it will enter the active mode and scan the network for other available robots. The eligibility value will then be queried and transmitted between all the existing robots. The leading robot is responsible fo r ranking all the robot s in the list and then broadcast the list to the robots which connects to it. The connect ion will be closed af ter the negotiation is complete in the last step and waiting for another scanning after a short period of time. Also, the same process will also work when more robots attempt to join the network. The difference is when the leading robot scans over the network, more available robots will be found. The connections from the leading robot to all av ailable robots will therefore be established and closed after it is finished.

PAGE 99

86 (a) (b) (c) (d) (e) Figure 5.3 Network initializ ation process. (a)waiting for timeout, (b)connection established, (c)eligibility query, (d)EL broadcast, and (e)close connection. 5.5 Follower Failure Another possible scenario is the failure of the follower. This incident can be detected and updated from the scanning by the leading robot. Once any available robot becomes invalid, the response for connection re quest will no longer be seen. So the failed

PAGE 100

87 robot will be removed from the EL at the leading robot. The new eligibility list will be broadcasted through the networ k. This procedure could be seen in Figure 5.4. (a) (b) (c) (d) (e) (f) Figure 5.4 Topology configuration when follower fails. (a) a network with 3 robots, (b) robot 7 fails, (c) leader scans the network, (d) failed robot detected and removed from the EL, (e) broadcast the EL, and (f) finish the configuration. 5.6 Leader Failure One of the requirements for the dedicated network is to be robust to the possible robot failures. A decentralized system coul d encounter the exceptional failure on any of its member. Therefore, for the system aut onomy, the communication network must be able to accommodate itself to such an even t, not only for the failures on followers but

PAGE 101

88 also the leader. The occurrence of the incident on any individual robot must not make the system operation obsolete. Therefore we need to consider a leader failure as well. The procedure to configure the ne twork from the leader failure is shown in Figure 5.5. (a) (b) (c) (d) Figure 5.5 Topology configuration when the lead er fails. (a) a network with 3 robots, (b) the leading robot fails, (c) time out form the connection waiting, and (d) network reinitialized by the mo st eligible available robot. In Figure 5.5, when the following robots get time out from the connection request waiting, they will assume the leading robot failure occurs. The leading robot will be removed from the eligibility lis t on each follower. The next eligible robot among the rest will then become the most el igible robot, enter the active mode operation, and seize the control of the network, as shown in Fi gure 5.5(c). The network initialization will therefore be processed by the most eligible robot as explained in section 5.4 5.7 Control Privilege Transfer Another scenario that can occur in this ne twork is the joining of a robot which has a higher eligibility value for the mission objective. A transfer of the le adership needs to be

PAGE 102

89 made in the network to a more eligible robot when found. The configuration is shown in Figure 5.6. (a) (b) (c) (d) (e) (f) Figure 5.6 Topology configuration for the contro l privilege transfer. (a) a network with two working robots, (b) a more eligible r obot attempts to join the network, (c) query for the eligibility value, (d) broadcast the EL, (e) leader unregisters from the network, and (f) a re-initiali zation is made by the new joining robot When a more eligible robot is found by the leading robot, the leading robot will unregister itself from the networ k and be removed from the list However, it will still stay in the network. As a result, the leading of the network will automatically be transfer to the higher eligible robot and take over the ne twork control. When the higher eligible

PAGE 103

90 leader enter the active operation mode and scan the network, the orig inal leader will then be found and join in the eligib ility list again as a follower. The self-configurable network discussed in chapter 4 and 5 is a semi ad hoc network. A hub to group and administrate the av ailable robots is stil l a requisite in the network. However, a dynamic adjustment capability is added here regarding to the eligibility of each robot. This method is somewhat restrictive comparing to a MANET environment. However the network complex ity is lower since the burden of using a network layer routing protocol is avoided. Therefore less hardware requirement is needed. It is especially a more efficient manner to implement a network in a MRS when the scalability has a lower priority.

PAGE 104

CHAPTER 6 CONCLUSIONS AND FUTURE WORKS This research presents the effort to es tablish the wireless multi-robot testbed for MAS research purposes. The revision fr om previous design shows significant improvements on both the capability and perf ormance. A more autonomous multi-robot system with a more dynamic and flexible operation mode is created by using both hardware and software means in chapter 3 ~ 5. The work discussed in the thesis especially focused on a cost effective approach in terms of power and computational uses for the hardware implementation. 6.1 Conclusions The applications for the MAS are widely developed and recognized recently result from the mature of critical technologies, such as theoretical developments, computer technologies, communication enhancements, a nd rapid growth of the embedded system industries. To recognize the pending physical syst ems as well as the theoretical issues and progresses is the initial phase before the de sign problem. It is pr esented in the first chapter. Meanwhile, the network communications is an important prerequisite for the multirobot testbed for any further theoretical validatio ns on MAS. The interconnections must be made to organize homogeneous/heterogen eous robots into a fleet. A brief review on the rudimental knowledge of network communicati ons is provided in this thesis as well as the reference for a network environment design problem. 91

PAGE 105

92 The hardware design could be regarded as an embedded system design problem. Therefore, the discussions about the impl ementation emphasize the concepts of the design philosophies and how th e signals and data flow ove r different modules in the testbed system. It is a valuable aspect for the testbed development so another similar design based on different requirements or further revisions could be made by referencing this implementation. In the experimental te stbed a 32-bit Pentium processor is used on the robot in conjunction with sensors and n ecessary modules to perform an enhanced processing and sensing tasks for an MAS res earch needs. The approach used here is valuable for manufacturing a good cost-performance multi-robot testbed. The lower layer wireless protocols extend th e original protocols from peer-to-peer to network wide. However, the economic desi gn has restrictions on the WAN and routing capabilities. The implementation of a more scalable network is beyond the scope of the thesis. Also, an initial vers ion of ACL is proposed so the fundamental inter-cooperation between robots could be dealt. The robot autonomy could be discussed in many different aspect s from navigation, communication to health manage ment and task allocation. This thesis also provides a self-configurable communication scheme fo r the more robust operation and system management. The mechanism is intuitive yet effective for the operation of the MRS. A more ad hoc environment could be develope d based on a revision of the lower layer SDU overhead. 6.2 Future Works The thesis presents a continuing work of developing a MAS research facility. It promotes the control scheme from centralized co ntrol in the previous job (see Kantor [12]) to a decentralized control scenario from th e hardware and software aspects. However,

PAGE 106

93 there are still some significant restrictions for the system as mentioned eariler. For example, the positioning system still required wired signal transmission. Also, the ACL will need to be discussed further in order to support the operation in either a mission specific manner or a more general comp atibility for robot inter-operations. Also, for mobile ad hoc network, dynami c addressing will significantly improve the scalability of the network, furt her discussion can be found in [ 38 ]. The work discussed in this thesis only validates the implementation of the design. A further performance analysis is still needed to be discussed in the future. The hardware components could also be further optimi zed in both mechanical and electrical architectures as well.

PAGE 107

APPENDIX ACRONYMS A A/D Analog to Digital ABM Asynchronous Balanced Mode ACL Agent Communication Language ADC Analog to Digital Converter ARM Asynchronous Response Mode ARPANET Advanced Research Projects Agency NETwork ARQ Automatic Repeat reQuest ASCII American Standard Code for Information Interchange B BPSK Binary Phase Shift Keying C CCK Complementary Code Keying CDMA Code Division Multiple Access CRC Cyclic Redundancy Check CSMA-CD Carrier Sense Multiple Access with Collision Detection CSMA-CA Carrier Sense Multiple Access with Collision Avoidance CTS Clear To Send D 94

PAGE 108

95 D/A Digital to Analog DARPA Defense Advanced Re search Project Agency DART Demonstration of Autonomous Rendezvous Technology DNS Domain Name Service DTE Data Terminal Equipment E EL Eligibility List EMI Electro-Magnetic Interference F FCS Future Combat Systems FEC Forward Error Correction FIPA-ACL Foundation for Intelligent P hysical Agents Agent Communication Language FTP File Transfer Protocol FSK Frequency Shift Keying G GPS Global Positioning Systems H HDLC High-level Data Link Control HTTP HyperText Transfer Protocol I IP Internet Protocol IR InfRared sensor

PAGE 109

96 ISO International Organi zation for Standardization J JRP Joint Robotics Program K KQML Knowledge Query and Manipulation Language L LAN Local Area Network LED Light Emitting Diode M MAC Medium Access Control MANET Mobile Ad hoc NETwork MAS Multi-Agent Systems MEMS Micro Electro-Mechanical Systems MODEM Modulator/DEModulator MP Motion Planning MRS Multi-Robot Systems N NASA National Aeronautic s & Space Administration NRM Normal Response Mode O OSI Open Systems Interconnection OSD Office of the Secretary of Defense

PAGE 110

97 P PC Personal Computer PDA Personal Digital Assistant PDU Protocol Data Unit POP Post Office Protocol PWM Pulse Width Modulation Q QoS Quality of Service QPSK Quadrative Phase-Shift Keying R RTP Real Time Protocol RTS Request To Send RWC Robot Work Crew S SDU Service Data Unit SNR Signal to Noise Ratio SRAM Static Random Access Memory T TDMA Time Division Multiple Access TCP Transmission Control Protocol TELNET Telecommunications Network U

PAGE 111

98 UAV Unmanned Air Vehicles UDP User Datagram Protocol UMS UnManned Systems UGV Unmanned Ground Vehicle V VPN Virtual Private Network W WALKER Wireless Autonomous Linux-based Kinematic Expert Robot WAN Wide Area Network

PAGE 112

LIST OF REFERENCES 1. Defense Advanced Research Project Ag ency (DARPA) Overview Bridging the gap, http://www.darpa.mil/body/pdf/darpaoverview.pdf Mar. 2004. 2. Joint Robot Program, 2004 Unmanned Ground Vehicle (UGV) Master Plan, Department of Defense, http://www.jointrobotics. com/activities_new/FY2004%20JRP%20Master%20Plan. pdf 1/6/2005. 3. Litt, J. S., Wang, E., Krasowski, M. J., and Greer, L. C., Cooperative Multi-Agent Mobile Sensor Platforms for Jet Engine Inspection-Concept and Implementation, Proceedings of International Conference on Integration of Knowledge Intensive Multi-Agent Systems, 2003 pp.716-721, Cambridge, MA, Oct. 2003. 4. Mackworth, A. K. On seeing robots. In Basu, A. and Li, X., editors, Computer Vision: Systems, Theory, and Applications pages 1-13, World Scientific Press, Singapore, 1993. 5. RoboCup Official Site, http://www.robocup.org 1/6/2005. 6. Jennings, J. S., Whelan, G., and Evans W. F., Cooperative search and rescue with a team of mobile robots, Proceedings of IEEE Int. Conf. on Advanced Robotics, 1997, pp.193-200, Monterey, CA. 7. National Aeronautics & Space Administ ration (NASA)s Mars Exploration Program, http://mars.jpl.nasa.gov/ 07/04/2005. 8. National Aeronautics & Space Admini stration (NASA) DART Mission: Demonstration of Autonomous Rendezvous Technology, http://www.nasa.gov/mission_pages/dart/main/index.html 07/04/2005. 9. Goldberg, D., Cicirello, V., Dias, M. B., Simmons, R., Smith, S., Smith, T., Stentz, A., A Distributed Layered Architect ure for Mobile Robot Coordination: Application to Space Exploration, Presented at the 3 rd International NASA Workshop on Planning and Scheduling for Space, 2002, Houston TX. 10. Jet Propulsion Laboratory (JPL), Nationa l Aeronautics & Space Administration (NASA), Planetary Surface Robot Work Crews, http://prl.jpl.nasa.gov/project s/rwc/technology/ rwc_tech.html 1/11/2005. 99

PAGE 113

100 11. Cao, Y., Fukunaga, A., Kahng, A., and Meng, F., Cooperative mobile robotics: Antecedents and directions, Proceedings of IEEE/RSJ Int. Conf. Intelligent Robots and Systems 1995, pp. 226-234, Pittsburgh, PA. 12. Kantor, G., Singh, S., Peterson, R., Rus, D., Das, A., Kumar, V., Pereira, G., Spletzer, J., Distributed Search and Rescue with Robot and Sensor Teams, Presented at the 4 th International Conference on Field and Service Robotics, Jul. 2003, Lake Yamanakako Japan 13. Kitts, C., Palmintier, B., Stang, P., Sw artwout, M., A Distributed Computing Architecture for Small Satellite and Multi-Spacecraft Mission, http://hubbard.engr.scu.edu/docs/pubs/2002/02A_Distributed_Computing_Architecture.pdf 1/11/2005. 14. Hwang, Y. K., Ahuja, N., Gro ss Motion Planning A Survey, ACM Computing Surveys, Vol. 24, No. 3, Sep. 1992, pp.219-291. 15. Sylvester, A. C., Path Planning and Control of a Nonholonomic Autonomous Robotic System for Docking, Masters th esis, University of Florida, Dec. 2003. 16. Olfati-Saber, R., Murray, R., Distributed cooperative control of multiple vehicle formations using structural potential functions, Presented at IFAC World Congress Jul. 2002, Barcelona, Spain. 17. Gerkey, B. P., Mataric, M. J., S old!: Auction methods for multi-robot coordination, IEEE Transactions on Robotics and Automation special issue on Advances in Multi-Robot Systems, 18(5), Oct. 2002, pp. 758-786. 18. Leon-Garcia, A. and Widjaja, I., Communication Networks: Fundamental Concepts and Key Architectures McGraw-Hill Publications, second edition, New York, NY, May 2001. 19. Postel, J.(ed.), Internet Protocol, RF791, Sep. 1981, http://www.ietf.org/rfc/rfc0791 05/15/2005. 20. Leiner, B., Cerf, V., Clark, D., Kahan, R., Kleinrock, L., Lynch, D., Postel, J., Roberts, L., Wolf, S., Labovitz, C., Malan, G., Jahanian, F., A Brief History of the Internet, http://www.isoc.org/internet/history/brief.shtml 7/20/2005. 21. Postel, J.(ed.), Transmission Cont rol Protocol, RFC793, Sep. 1981, http://www.ietf.org/rfc/rfc0793 05/15/2005. 22. Postel, J.(ed.), User Datagram Protocol, RFC768, Aug. 1980, http://www.ietf.org/rfc/rfc0768 5/15/2005. 23. Mobile Ad hoc Networks Official Charter, http://www.ietf.org/html.ch arters/manet-charter.html 5/21/2005.

PAGE 114

101 24. Corson, S., Macker, J., Mobile Ad hoc Networking (MANET): Routing Protocol Performance Issues and Evaluation C onsiderations, RFC 2501, Jan. 1999, http://www.ietf.org/rfc/rfc2501.txt 5/21/2005. 25. Perkins, C., Belding-Royer, E., and Das ., S., Ad Hoc On Demand Distance Vector (AODV) Routing, RFC3561, Jul. 2003, http://www.ietf.org/rfc/rfc3561.txt 5/21/2005. 26. Clausen T.(ed.), Jacquet, P.(ed.), O ptimized Link State Routing Protocol, RFC3626, Oct. 2003. http://www.ietf.org/rfc/rfc3626.txt 5/21/2005. 27. Ogier, R., Templin, F., and Lewis, M., Topology Dissemination Based on Reverse-Path Forwarding (T BRPF), RFC3684, Feb. 2004, http://www.ietf.org/rfc/rfc3684.txt 5/21/2005. 28. Bradner, S., Benchmarking Terminology for Network Interconnection Devices, RFC1242, Jul. 1991, http://www.ietf.org/rfc/rfc1242.txt 5/22/2005. 29. TJ PRO TM Assembly Manual, http://www.mekatronix.com/manuals/TJPro/tjproam.pdf 6/17/2005. 30. Midicon Modbus Protocol Reference Guide, http://www.eecs.umich.edu/~modbus/documents/PI_MBUS_300.pdf 6/4/2005. 31. XCite Advanced Program ming & Configuration, http://www.maxstream.net/products/x cite/adv-manual_XCite_AdvancedProg&Config.pdf 6/4/2005. 32. ISO 7776, Information Processing System s Data Communication High Level Data Link Control Procedures Descri ption of the X.25 LAPB-compatible DTE Data Link Procedures. 33. Finn, T., Weber, J., Wiederhold, G., Ge nesereth, M., Fritzson, R., McKay, D., McGuire, J., Pelavin, R., Shapiro, S., Beck, C., Specification of the KQML Agent Communication Language, Technical re port, The DARPA Knowledge Sharing Initiative, 1994. 34. Labrou, Y., Finin, T., A Proposal for a New KQML Specification, CSEE Technical Report TR CS, Universi ty of Maryland Baltimore County, Aug. 1997. 35. The Foundation for Intelligent Physical Agent (FIPA)-ACL Specifications, http://www.fipa.org/specs/pesspecs.tar.gz 6/22/05. 36. Berna-Koes, M., Nourbakhsh, I., Sycara, K., Communication Efficiency in Multiagent Systems, Presented at ICRA 2004 New Orleans, LA. Apr. 26-May 1, 2004.

PAGE 115

102 37. The Foundation for Intelligent Physical Agent (FIPA) ACL Message Representation in Bit-Efficient Specifi cation, specification number XC00069, Aug. 2001, http://www.fipa.org/specs/fipa00069/ 06/23/05. 38. Sun, Y., Belding-Royer, E., A Study of Dynamic Addressing Techniques in Mobile Ad hoc Networks, Wireless Communications and Mobile Computing, Vol.4, pp.315-329, Apr. 2004.

PAGE 116

BIOGRAPHICAL SKETCH Chun-Haur Chao was born in Taipei, Taiwan, in 1977. He received his B.S. degree in electrical and control e ngineering from National ChiaoTung University in 1999. He enrolled in the graduate program of aer ospace engineering in the Department of Mechanical and Aerospace Engineering at Univ ersity of Florida in 2002. He later joined the graduate program concurre ntly in electrical and com puter engineering in 2003. He received both his M.E. and M.S. degree in 2005 103


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110319_AAAADY INGEST_TIME 2011-03-19T15:46:25Z PACKAGE UFE0011835_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 595610 DFID F20110319_AAASSH ORIGIN DEPOSITOR PATH chao_c_Page_075.jp2 GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
f8ee0702bff56c904a23aa190364e9a4
SHA-1
8f398387c0f3c84e3655e888d48b1b3d0fa17667
798034 F20110319_AAASRS chao_c_Page_060.jp2
d012518b1c48f9974505c3d515668b0c
966fdb4e89257a1c40f0cca52addf2e85401fc44
88548 F20110319_AAASSI chao_c_Page_076.jp2
ed37c9654b3d3807bdeb45e73186a1c6
5ce764b941f3cfa4f0f0085fb547518e8f147b32
97955 F20110319_AAASRT chao_c_Page_061.jp2
84d9aaa5a563267cb5509d2f7430f653
fc93d7fc301f490561f782ed12b9ad10f83b648d
83876 F20110319_AAASSJ chao_c_Page_077.jp2
d746014b036933b37dbe3ce1f7460160
2c8c7fb040e65d50ac8ea7453360a12e9c93401a
108129 F20110319_AAASRU chao_c_Page_062.jp2
0469f2c788ab3360283852648799b189
f45dbc2f01a3a2f82f3c69e7387200443bd4a9dd
893197 F20110319_AAASSK chao_c_Page_078.jp2
3a687aab7f23daf1bd74d947fb0b1ed2
78f657a4a8f052603cb64279963d698165f669b3
114177 F20110319_AAASRV chao_c_Page_063.jp2
f2e97e2fbfa83512088f387f56971bc1
6674d2fa90e9e57ad57d97d7055035d7d8c48ca8
109212 F20110319_AAASSL chao_c_Page_079.jp2
30071bba7d29acbb603af0037f460735
21baecdde2dd23f6fc6c18257a70d6f0793967e2
88172 F20110319_AAASRW chao_c_Page_064.jp2
0dd334e6e5eed0c1f6b8932f7ef5dd45
cb3394893e7ae4a703c1ee8f3bcc9c3ae61c3ad1
104418 F20110319_AAASTA chao_c_Page_094.jp2
a0788ea62d7f60892fd40553d8171b90
81a270231c55a7547abc45e4b6573c7904a82916
99951 F20110319_AAASSM chao_c_Page_080.jp2
8742750dc2b7c621d8b1006b9ffab1e2
24c7c2e7fbb4afd9de97992a358a6cc09609d133
98699 F20110319_AAASRX chao_c_Page_065.jp2
5290904e22a4914cc906724a251860bc
cafa0538d89732e3e1ce50315dfcec22fa862e78
108128 F20110319_AAASTB chao_c_Page_095.jp2
6890a92384ea142febd8bb2a4f191885
57ecf57f8fc20909a982c8fc1451012ff17f53a4
107170 F20110319_AAASSN chao_c_Page_081.jp2
5696deb3fa1d39d4bcfa3089c54b9438
eb48f563200b4056630807c81f2e98ce5ed3b029
1015578 F20110319_AAASRY chao_c_Page_066.jp2
95e145a67a4dd436273499a5862affc3
a72a80c9b4ae82e8bb290e737bc06d45cbfd1c97
691571 F20110319_AAASTC chao_c_Page_096.jp2
be363dfad3cb019a62ee8cffbb0a2f28
3f81558eca3c53078c575b9dd81c79db2260a09e
809008 F20110319_AAASSO chao_c_Page_082.jp2
d5665f48ecd2342d7c2da95f5d61d90b
73d1b47422f41210cb61b0a0ddc461cf86247864
86780 F20110319_AAASRZ chao_c_Page_067.jp2
a319e7138b89201766d1ba238d0e5788
8f3939e95d77560582b8b2ea017397f98b9a510e
1051926 F20110319_AAASTD chao_c_Page_097.jp2
2d42ea3ee69f2adf8b9039fa3da7208a
68c3dfef04eab5e217bc386b562751d258bda52b
86260 F20110319_AAASSP chao_c_Page_083.jp2
ddbeccb345bf2e94a9d126a6ed404f2c
0df658ceffd1db2991c57a1e13ac1f618070b505
840116 F20110319_AAASTE chao_c_Page_098.jp2
aaf56a79fdeb50f6dbf635e9c007f008
6b588ba1faf246387cfc699a19ee1d60b6107d05
741013 F20110319_AAASTF chao_c_Page_099.jp2
0aedfcc370fd31cce81ac30eec7c0350
18d5ac19e7fb7dfbed3df8f380c0a90cd2108203
776558 F20110319_AAASSQ chao_c_Page_084.jp2
1d6e63bd5a5fbbf495cc469ff115d39d
3d61bdd845d7b9d6b539b332635ac31bee0be239
924260 F20110319_AAASTG chao_c_Page_100.jp2
47301f055764cae73ed85e9f12f9a6e3
1561de6bc2c1ca516eecedf637434f400ae7c5b7
715376 F20110319_AAASSR chao_c_Page_085.jp2
994a923e263e41c242f6c82533e92c9f
8a06131f944d875c8ef68189c463304a995988c4
1030976 F20110319_AAASTH chao_c_Page_101.jp2
13cf3cc0f941ea40748d85e51ed2732c
7702f96b00dc0aa8532393c08cb5ec526b9b9d9d
886130 F20110319_AAASSS chao_c_Page_086.jp2
dbd0960335b4d9596c23c9fcabe13697
1668e1a8f7a0ab8f14edff9a4992cf406ed093b9
1000976 F20110319_AAASTI chao_c_Page_102.jp2
8a7837024c01e1aa0c8cbab6afee702e
ab32163ab9c1b788181ff628471dde8cf4454c14
107009 F20110319_AAASST chao_c_Page_087.jp2
816cafa705075fcad8a3cfdd449fc9fc
76ecbbb867819d0bc54fd5b1695c0496bae8ee52
46711 F20110319_AAASTJ chao_c_Page_103.jp2
552cbc316f1a691f613fbedf0cba23df
d53c958fe50ed97e5218f6ba932a9d940c1506a0
1036283 F20110319_AAASSU chao_c_Page_088.jp2
9c5947648e20bf04da655b6214d68889
137801f2cce5874e56df284854b3bec19f2a9db9
84830 F20110319_AAASTK chao_c_Page_104.jp2
68b1cb36681c1f0a9ebe1b6901ca365b
b89d7f2387a6dcd10b012f9f7401cbe05019dff0
76244 F20110319_AAASSV chao_c_Page_089.jp2
58bc6ed29e37e2f1bab4ff0776b7bb2e
144066dbe4fd030a3b4a9b47f24f36035b7b8405
106437 F20110319_AAASTL chao_c_Page_105.jp2
70ac59f33674d614edc92bcec4d1ff9a
cb74fb6c371f49ae01a2cf22c8d683ec262252dd
849878 F20110319_AAASSW chao_c_Page_090.jp2
fe5f49f1f64f98c971b8ac4b4b0b000d
a2d022556c9d1977601c42899d7ad99dd5405874
45779 F20110319_AAASTM chao_c_Page_106.jp2
be5d42518eb7f641b9439ac7bd5e6076
3987d5d67ac4652365de88c8d39a53b8d6c1e80a
1025109 F20110319_AAASSX chao_c_Page_091.jp2
f572abd2d994e23f3b6fdc9d7a9d14c8
1bf94ee3ff4ae5ef84c37df348b135a63e893a39
3777 F20110319_AAASUA chao_c_Page_004thm.jpg
d8ca42a857a438e0e5f4f0b3646e7742
5545a8818bb324a9be45898e3223d5254076659e
44588 F20110319_AAASTN chao_c_Page_107.jp2
0278a4ebf32718c1a29d5b1fda1957c6
ac8bc00dbcf3bcf6a6bfa82784257f2ecd1b4ce2
661681 F20110319_AAASSY chao_c_Page_092.jp2
906efefd952eeb45591c833e16d0b0e2
99835aa18e647f337857768898c7891fac96dabc
4108 F20110319_AAASUB chao_c_Page_005thm.jpg
fb71e986bc840bf478955e9d266c3db3
b5031a5acf552ee73a01dea4844e8056c4882940
41161 F20110319_AAASTO chao_c_Page_108.jp2
2650e45d79c25f84774e0de651c12392
6bf3f7604996d05f7e0d9c6241c726fee2fc6252
102054 F20110319_AAASSZ chao_c_Page_093.jp2
c7e0dd76ddb45e6776b0577ac018c14a
3f5c3ab2691ed60e4ba23bb17029489235ee2f0c
4812 F20110319_AAASUC chao_c_Page_006thm.jpg
68bb5f1d57ba8002f65c5416808e8650
ab4f9159b1587c48ffc1bff721647ca121d906d2
40419 F20110319_AAASTP chao_c_Page_109.jp2
db2c6d2cd8cb9d6b3b79557819d58382
18d5956ed8c62d502489b0fec502d111cf537b27
836 F20110319_AAASUD chao_c_Page_007thm.jpg
ba5637910cee8e55638d6662bb15e756
abb5b17f7d2ecbcd201528079381edbca62adf27
34571 F20110319_AAASTQ chao_c_Page_110.jp2
6b1cca54d5e43f33ea16d9bcf4f42584
f87dc3754bb6224014cfbaf9af91b5bd9826c741
2613 F20110319_AAASUE chao_c_Page_008thm.jpg
8d5d4901abe8dd0300788877914095bb
1c1e408e310e1c23ba962e00be7e688e2ba8bb65
4580 F20110319_AAASUF chao_c_Page_009thm.jpg
0a7d33790ece219fea4ada0a3fd0e8e7
f6190d5211226c034df5099ba0c92218717934d4
19332 F20110319_AAASTR chao_c_Page_111.jp2
a2bbf0a60e64629af3f82976243e0d07
0a1875f91ebf4db2fb947efb98fd23bea4a019d0
5243 F20110319_AAASUG chao_c_Page_010thm.jpg
568a70b1ae715cb0e5e43ef4453fb2e2
bf4de073c519b0c5c2cee5693589cfcb2eb32144
1441 F20110319_AAASUH chao_c_Page_011thm.jpg
984a9caff524785d1400da3936d540ea
f43f2a9652ec5bd97fb1c75771cc74a248749cd8
1051976 F20110319_AAASTS chao_c_Page_112.jp2
a294f3c98e7ed62fe5df38a5529f75a9
416c348445cbc01cff3ef70d89bf655f587f6b8d
4916 F20110319_AAASUI chao_c_Page_012thm.jpg
33685c3ce3cdb6762c0fd01ca03f9cc7
d3807f8db3235f15729af2a97a14ebbad4407e51
1051919 F20110319_AAASTT chao_c_Page_113.jp2
63dcc32f75d075c13522dd1dff09d48c
76460e20ac90b58d541538aba6589feaf31f53f2
3338 F20110319_AAASUJ chao_c_Page_013thm.jpg
1c16e01da3270c9feb9b1ed03998629f
b2d778c6769c8133936849ddfee200548a139a6d
1051979 F20110319_AAASTU chao_c_Page_114.jp2
2e6263a2f8203c5c7931d3d0c76d63f7
f827f9eddb23f31bf089aa0dca5905a95ac3e5bc
5392 F20110319_AAASUK chao_c_Page_014thm.jpg
092aa4ab4bb60947070527f19234b45b
60aed3fad728c0671d1c082608dedf98601f5a61
285942 F20110319_AAASTV chao_c_Page_115.jp2
690999ceaaf37b76431f6f8517e9f3bf
66306d8358a7efdefae045dc331c5848883cb982
5573 F20110319_AAASUL chao_c_Page_015thm.jpg
04c35a051541b5dd7a11f74e2c456c14
56d4feae42c1294d2d3e58a5d0ae9797b5c5f294
32201 F20110319_AAASTW chao_c_Page_116.jp2
eb24ecd782091d23e4c5b145e08aa4db
a15aab7b1357a6e9bf198d00301020fcc0bf5f6a
5524 F20110319_AAASVA chao_c_Page_030thm.jpg
421a08a9a80a916897e4f0c3e1599122
9968426c2877e742618c0a003040b582d68fae4c
6722 F20110319_AAASUM chao_c_Page_016thm.jpg
596bf10823a11fa9328572f8fe865c57
529a56d56390824ea8622986470a58016941f964
1835 F20110319_AAASTX chao_c_Page_001thm.jpg
a57dde3c42ba88807931c3d6f0a96b16
2045c2534c77da55be083255430aa192975c7d14
6086 F20110319_AAASVB chao_c_Page_031thm.jpg
2d4218871db8c6e9dbf25e50258f8797
de100b4b27735bb1ca8c8e2eca028ca1950ebb21
5484 F20110319_AAASUN chao_c_Page_017thm.jpg
fa0ade6c77bb4a8590494e12b84fd2b7
3e7b4703a7243230318922dfd824e16a5b8e166b
537 F20110319_AAASTY chao_c_Page_002thm.jpg
7777722495f469eec63e4c4af179280c
9f98ee406f2aa34895fba9cb3e32b6e563eb504c
4684 F20110319_AAASVC chao_c_Page_032thm.jpg
13e6b19f89332317b14713fb82551e14
e1160098cdcdbbe207d9844836104cf62351b4bf
6900 F20110319_AAASUO chao_c_Page_018thm.jpg
3503b874e661459add93d5111520ce4b
62079e5cb14c0ce1013eae9f7b629a5d92a136af
502 F20110319_AAASTZ chao_c_Page_003thm.jpg
41f75a812301355254c54226a53c377d
fa8e79f3d2134a8e7afa4251e6392b14d0e77fdf
4169 F20110319_AAASVD chao_c_Page_033thm.jpg
034fe7f5711019c6e63949fbb0c9b022
0478937cbc911ca13762f72bb5aacc9e644f03dd
5540 F20110319_AAASUP chao_c_Page_019thm.jpg
adf71601edc02baa850ef57650230832
7b334b039171ad793a14140baeae0cd2b581be19
4944 F20110319_AAASVE chao_c_Page_034thm.jpg
875daa3ada77dfe9415ec168c8f95e22
d95307b2dda07f5985f40e6c6541c177304963b5
5717 F20110319_AAASUQ chao_c_Page_020thm.jpg
d34154edcbd7a4389c41b248640affc6
1d1f3a07e27850b23abea305f5aada1b4eff92cd
5502 F20110319_AAASVF chao_c_Page_035thm.jpg
caa272bbda2d16affd33a3a9450d88c3
3eafef7837c02f885de905cc90313c9571d619ce
5715 F20110319_AAASUR chao_c_Page_021thm.jpg
55ffcb77add61ba4cd24b542e3c49c85
dc2aa512f3f16c21128af9c09cab9531af5f16cf
4800 F20110319_AAASVG chao_c_Page_036thm.jpg
faad85915e6b1cce1304d6dc9c0adf24
2e3b6fae22ff3a7336d19dc06f804975d6f533fe
5408 F20110319_AAASVH chao_c_Page_037thm.jpg
3453e09870f541e2fec818d59906e4b4
03c611073cff5158d68e13a0644a1cba7a4a5bfd
4976 F20110319_AAASUS chao_c_Page_022thm.jpg
b117c4c09300252e0d11b162da326c46
a4634a04e92ee297a3d8fb2ef29254bd696479e1
6159 F20110319_AAASVI chao_c_Page_038thm.jpg
684d1486dcfe27f9292c2914cb4c1766
1310f9a92b558103234161738dc228da52314437
3912 F20110319_AAASUT chao_c_Page_023thm.jpg
20ed5a4ddc94f965abe0bdc9e08de6d0
dccd08b801a95868ff68426a7ba5728a9ab670c5
1070 F20110319_AAASVJ chao_c_Page_039thm.jpg
0221e9ab4d15adc6e9a71c77c12bfd08
a743e18ada85aa707c4d67bf9b3ccca52d919436
4940 F20110319_AAASUU chao_c_Page_024thm.jpg
2494496dd633438de903134a82b3ef60
c8787f7dbc12cfa7e77e4b7913cd4ea72f6c99bb
5334 F20110319_AAASVK chao_c_Page_040thm.jpg
aac26f6d8cea8062ab205358ea3c5252
8b76f118f5a5d89d8bb6c0eb26bb91096e825527
F20110319_AAASUV chao_c_Page_025thm.jpg
0b14c363426ebc73cb591172d03909b6
1fb37b47df79871c766e147c84bdfb82f634063d
5983 F20110319_AAASVL chao_c_Page_041thm.jpg
0d23f7340b8b62a8cc563e478d4cec91
20837aac3224f062cb62c0e562b8739a40875315
5637 F20110319_AAASUW chao_c_Page_026thm.jpg
17d75bd1b3e116e8ab6ca40f146e1b09
18eeef34e8f6397fa2a7132a41f0b9df8cc68503
5800 F20110319_AAASVM chao_c_Page_042thm.jpg
2aefc97f0fc1452935f54f707ad43fee
ba3886d142ba9fa43e4b7381f357955ac8a54f47
6129 F20110319_AAASUX chao_c_Page_027thm.jpg
166bcb2dfc8392026bc9cdcbbc89bcbc
475cd7ee6367473a9181c04e901d7def964066de
5649 F20110319_AAASWA chao_c_Page_056thm.jpg
6e3a361572442b6459136307507121d8
cfa2d2ff91d20e5b535f4e45c76a57505409f215
6246 F20110319_AAASVN chao_c_Page_043thm.jpg
5eb1309d4fbd34ca034d55ca6507098f
d0e83bd43e5c0fe8f5fa3f6cd00f3ccafb9cacd0
6254 F20110319_AAASUY chao_c_Page_028thm.jpg
ed7a870e568fe4c35dc8278a365cc8de
2687190e4c0b1cbc77d11e2ec253e326b60b0a20
3712 F20110319_AAASWB chao_c_Page_057thm.jpg
24b0fd0d11610b0af12e71b097eb69dd
b0f65cb0e707f77a04cb4957fba4261b09c3f47e
6241 F20110319_AAASVO chao_c_Page_044thm.jpg
4e7e887cf26f4f3a1e5f0f7e0386fa6e
1c39129ea4b838d9cb7d0689fc30881448a5db79
6204 F20110319_AAASUZ chao_c_Page_029thm.jpg
edb251872b1f4e6d661187a2742ac366
96075979ef7496caef2c677b641c11eda814a26b
49300 F20110319_AAARTA chao_c_Page_043.pro
fe1723a2bb4cb0f2b298d8a25f7e0d63
1afe7b096864766b7d9d49957a2db85f4ab8c3ad
6531 F20110319_AAASWC chao_c_Page_058thm.jpg
b2b5f02e6d4f4c9938641b5484f3da31
7eb93d32055adc4a0ee888a766cb337a0e16304b
6125 F20110319_AAASVP chao_c_Page_045thm.jpg
da2b08a052ade85f851679cbeb794fa4
8b7007fee9a5405e27a2841488cae89729438634
25415 F20110319_AAARTB chao_c_Page_062.QC.jpg
1457a560394fa49163fcfc0cc9347df3
11501125b04f6491bac562f6988ca2faff2e02d3
5400 F20110319_AAASWD chao_c_Page_059thm.jpg
20e0071c7433af0cc663b2a08d7ac8af
1aaa8063389a5f596d13d2f456796618a1c21550
4955 F20110319_AAASVQ chao_c_Page_046thm.jpg
e084d9267b5bc5319e764de192db4800
969644c6e79129f3bb7d9ac24ce572616f67b2a6
5911 F20110319_AAARTC chao_c_Page_074thm.jpg
9485953cc91aab3d3d42065f3ed06df8
0b3952c6d4993831d021ab330caf9d6066d2499e
4591 F20110319_AAASWE chao_c_Page_060thm.jpg
166e352b0b6dfe08ffa80b6aba92391f
4030e67387928c0521482676fd6c5163bc09e0fe
5912 F20110319_AAASVR chao_c_Page_047thm.jpg
a8e018f54b567de0b4f3bfd420b0f88c
084a9330cfeafc9d2750c465aedbf1ae97db17d8
185956 F20110319_AAARTD UFE0011835_00001.xml FULL
6f20a3554acd799679e17634ca2f7072
3ffe0b272c2fc1a488a9a13d2d5c729e5a8aa82e
5747 F20110319_AAASWF chao_c_Page_061thm.jpg
9837adfadda770982e300ef2416cd6a0
94d6467d1862459fc1a5cc0deb92173118adbdb4
6930 F20110319_AAASVS chao_c_Page_048thm.jpg
a34c12cb962965b98b7c973b9fb0f5b1
e4354e9aff860e43b88a0c630573e7084eb1ed6e
5989 F20110319_AAASWG chao_c_Page_062thm.jpg
d678d8bf0e675ebdb2414565e62220e5
063538e10fda0e440fec95ec57a4889b0f3bbeb0
6494 F20110319_AAASWH chao_c_Page_063thm.jpg
6bec4c53a9ee42235f4faaee296186ba
4b718b402b0e6bcd6f399fd7cdaa8773b29f4f18
5657 F20110319_AAASVT chao_c_Page_049thm.jpg
5a66190f1ee05a31e6de968b8e267894
4ab638b87cc43ecefba28270614b305de00236e0
1053954 F20110319_AAARTG chao_c_Page_001.tif
3ae0624504051480050e92ea3f727512
594ca6253ed6d1a22f3c6dc202e987563a3b9dca
5311 F20110319_AAASWI chao_c_Page_064thm.jpg
744cb6ca350566affacb8a9cb96b1ab0
3b88aa38ab60ff2c8f0dd54bb640b5e566adafde
5146 F20110319_AAASVU chao_c_Page_050thm.jpg
f2a155e1b7b8bcf9e67b2121a8fe7e88
992181e1c1fb75abe381cbb360d8356ca9221a93
F20110319_AAARTH chao_c_Page_002.tif
77aa8a2e76d1462d81a8b0f90293bccc
0a036700d9a7c1d48c205396fadf7d53981bb3b3
5651 F20110319_AAASWJ chao_c_Page_065thm.jpg
c5444ad243a6281ebdf34a412ca081e9
1009df0bb27449ce14f38226071be746374070c8
F20110319_AAASVV chao_c_Page_051thm.jpg
addee24debb3c5f31d405f136eaae2bc
5e41f5ba12f8fc58f5893248c4d519cf14c20b9e
F20110319_AAARTI chao_c_Page_003.tif
131d99c09d1a9fdb16219cc7e23aa598
7ae4cad820b75abdca1f6648ed042f95f5ac6782
5852 F20110319_AAASWK chao_c_Page_066thm.jpg
5c89523fc019efe524b464197fa1dcf7
728817660f48f4c8ac4258cf9db07968da41db24
6522 F20110319_AAASVW chao_c_Page_052thm.jpg
cd8a679b457848472328b36b01014480
87cc987a20d3666c28f828610236e2b3ca9ed187
F20110319_AAARTJ chao_c_Page_004.tif
13bdc1616a85952877e68cdc226285fd
172c818dea3cc1e237049ee82ae011ef77b35967
5158 F20110319_AAASWL chao_c_Page_067thm.jpg
452e57636f743b6574b1477493022d35
4273c800c7a682c8447231adedf7441257eadcb6
5136 F20110319_AAASVX chao_c_Page_053thm.jpg
6e0b2a910152bda127b3590c6a91bcdc
b2e40be23c6af93a944ce9f286fbb42e30abab07
5468 F20110319_AAASXA chao_c_Page_083thm.jpg
73bc15eb9703d047a6d1fa62ac7a3c3b
5092e73b498c3fb2bc1fe093bccf20035880486b
25271604 F20110319_AAARTK chao_c_Page_005.tif
da3f9dfd15effb798e90f5ed72240065
7ec795fb0a61c9292e14bb8166a9002a7c82ed58
6181 F20110319_AAASWM chao_c_Page_068thm.jpg
4e365310248ef0c2ef0e016191ee79e8
477bf29c7899e75dc0ec13bdac8cdd4573525387
5442 F20110319_AAASVY chao_c_Page_054thm.jpg
d480f24d997b22e4d27cfbe9906fd011
5b05d6d83b08755cdcb9c6af48b8be95f05a0cb5
4794 F20110319_AAASXB chao_c_Page_084thm.jpg
31ddb36c31645c2e9342df96c634d2da
5765a854131028d9083bf3fc2ad485a428a462cc
F20110319_AAARTL chao_c_Page_006.tif
ab5204fcab3ca9156494364248163d73
4270ea0ade5813ed8180ba3710ba3d9a4aeaeec3
5712 F20110319_AAASWN chao_c_Page_069thm.jpg
1e61db7188aa7b960a40931751b0dddd
3563a5c33e37bd4ce80c7cdfaaab415884115ce6
5864 F20110319_AAASVZ chao_c_Page_055thm.jpg
e249c8680a971d53f0cace3ca63bb26e
e7f37da9b79287d3b662f499b3aea70c8238026d
F20110319_AAARUA chao_c_Page_021.tif
8160ec2fadbf970a6ef39ae4b8d1db66
9dca6290e4e856b36019b9c0b0f73b76d2ea9f88
4707 F20110319_AAASXC chao_c_Page_085thm.jpg
54b928e013bd4794d508c4f7470c7bcc
4b14684104b46bdc72bcc1cc00bf2fb858922f0e
F20110319_AAARTM chao_c_Page_007.tif
a4aeedead9a4386113bebc64efe8d146
81506f8622310eaf2bd179c04231315c7b06773f
6107 F20110319_AAASWO chao_c_Page_070thm.jpg
d44a81269eed23f2e8cfbaf333eef3c7
c9a94fb757c44e2010a6b2c06cee17785c3ee3c0
F20110319_AAARUB chao_c_Page_022.tif
edc53eb6896e6fed0be534c200c1185f
b2c1be2ebaf0731b9cb8cd7b1ac9013d0cac5488
5479 F20110319_AAASXD chao_c_Page_086thm.jpg
c9f6badfade82b76fba26852f22a9c7f
5dc329489df9a34440b7c346c9d559aee1944b28
F20110319_AAARTN chao_c_Page_008.tif
1528621513b0102eaf9013b6df81a54f
2da30e9505ae8c752cf1c402926427c22b947e23
6091 F20110319_AAASWP chao_c_Page_071thm.jpg
fe15b7cc6bc988e41d9ce1eef0d9b52e
9f651a95e0e65eba3f892a73a5b5030b9b5b2130
F20110319_AAARSZ chao_c_Page_093.tif
cafc44770d7c4234955841084bf26d95
68326726eb302f0cb79a1b7e40d89799a99beef4
F20110319_AAARUC chao_c_Page_023.tif
c099b081549b73fe09a0954886ea0e55
0f15ffa9ba27e8232a4ab66a457b5a1cd7ddd5c7
6097 F20110319_AAASXE chao_c_Page_087thm.jpg
0a410c4388bfce2ef850ee7424f062a5
cab3ffb0b890dbc166e0afc9697073b084134f2e
F20110319_AAARTO chao_c_Page_009.tif
c8784b9db9aff3056dedf3c6a2669e3e
fa188d6c2ce6ed8904771f40d1728200f1c35ef8
5539 F20110319_AAASWQ chao_c_Page_072thm.jpg
36f75a254466710933596396ff11e7a1
4c0e1578b74bfca214ed46052559ee8a272c64d9
6014 F20110319_AAASXF chao_c_Page_088thm.jpg
99bc91e0c8ee49a91e2ca7de518902ed
ad8820018cdf6b6ee11084f81f24028f1194c387
F20110319_AAARUD chao_c_Page_024.tif
fced1e0c2a26bed4dd3871c07d2dbf57
248e22be6bd23fa7295d074e38bc1d7e4a8a771e
F20110319_AAARTP chao_c_Page_010.tif
69548f0e381f4eff97dc03921e795f4d
347b39413917b241e1fad6216c0213c1e6893974
5841 F20110319_AAASWR chao_c_Page_073thm.jpg
65e92dc04b4ba53eb2097bcc183cd3b6
e6af5fd435783c8934f02dc368b8cbb0da4463e6
4950 F20110319_AAASXG chao_c_Page_089thm.jpg
b39ad0c38470b16ccfc5ab6ed64ba2c4
a1c7e1a691d77714d068ec55aed609b44494aa68
F20110319_AAARUE chao_c_Page_025.tif
14b64c4e14a06c357e3f297d097ad33b
244e436a61e3dc641875fe50bed0fe2aaacabfaa
F20110319_AAARTQ chao_c_Page_011.tif
0615dacd8a7036b75743af39d8b2b2fa
442b07831a5c5e1dca173dcbad121343847f9ccf
4107 F20110319_AAASWS chao_c_Page_075thm.jpg
4a728f3c8071b1c1f0aa548f962c5377
45e60899503f723575c07545852cd7d46530ef91
5347 F20110319_AAASXH chao_c_Page_090thm.jpg
cdc7a0f0d3896effe3fb614c9f427809
9a29d4c48b26097d877b029626400ffaefb26e63
F20110319_AAARUF chao_c_Page_026.tif
7fa56c23c1c6bb05fedf3b67212b5a7a
9598c8e4ff9bfa886a651e3ea619c5e591df225e
F20110319_AAARTR chao_c_Page_012.tif
e5dd878618c33fd858ee5df1304e43de
b399d994dd506158bd95ed8cd802fbd31d9ea81a
5103 F20110319_AAASWT chao_c_Page_076thm.jpg
f8d99c2397479fb7521fc50a437784dc
e11580712a7b8f30655e88f96722a27937764bfb
6034 F20110319_AAASXI chao_c_Page_091thm.jpg
31054ab31179558bea4666deaab20a0a
972a62856bdd43117d2551e7658f4b2e7f4f262d
F20110319_AAARUG chao_c_Page_027.tif
38c6d9a0045f2c5e1b9267f27ebf839f
7829a7f68541abc4d424b7a24d5a1840b5191747
1976 F20110319_AAASAA chao_c_Page_062.txt
9da95e5854b1266e3b6bfdf889b38c5c
75fb989a5f6832db9856f539091de51fd9859025
3436 F20110319_AAASXJ chao_c_Page_092thm.jpg
af2fb948e474d9a15b74c9cd86407683
0a0c82ee1eabd6e3cdf07423940b140d427dae05
F20110319_AAARUH chao_c_Page_028.tif
69e6219786bacf53258904474174d4dc
4cda5c26726a0fd6f3398d83bb29c31449e5297d
F20110319_AAARTS chao_c_Page_013.tif
94dfc11036652f1553a4d18804065d61
930e28af8b0becb8774cd692119a3597d83924db
4887 F20110319_AAASWU chao_c_Page_077thm.jpg
55391580f2bd0596eed5443ba126da9f
4496cb8acbff2302d0cec93d4b4e204d3f2c69f0
2071 F20110319_AAASAB chao_c_Page_063.txt
ef23bbe2a60504f84174d3e63794e702
705ee481c1b73cb6369521277258795da771e678
5806 F20110319_AAASXK chao_c_Page_093thm.jpg
c326db5c591ec50692add0deb3f2517c
52c3bf9055f41aff75342acce137876cae51c7d7
F20110319_AAARUI chao_c_Page_029.tif
c6e760ab5e6adc2fef1942d15e077b09
88bb1234472f94268d8172d9eedf107aa104dc06
F20110319_AAARTT chao_c_Page_014.tif
448c72e50a0c4d8195cd4255c67c03ec
37b04310f060f36b2a9954c73404b891bb3ebbf7
5578 F20110319_AAASWV chao_c_Page_078thm.jpg
cf1eb0e82505ab3bd3cdd7e9f892c246
a104e2bfcd1414f03e9ad1289cee0111d59c6dfc
1625 F20110319_AAASAC chao_c_Page_064.txt
ab814357674344a239e29f0044a174e6
14be9046d16ec2e88fb13284b21139581ce0d367
6188 F20110319_AAASXL chao_c_Page_094thm.jpg
ea315d3e0a17b9d7d770a21d4597896d
45a5ef51e7c08d55706d49e2318eb8b5b40a5954
F20110319_AAARUJ chao_c_Page_030.tif
e400b2d2ba6e992c8f2ab1660c7865c3
70ab47d9aaba7e8b3c3224a965c3bb917306ace8
F20110319_AAARTU chao_c_Page_015.tif
b2e414b592291a2b2b704b540ae3ed85
1dedfa5b16fdf14423b63a433a7f0db586bd9877
6508 F20110319_AAASWW chao_c_Page_079thm.jpg
d149ac67e20f504e331552e9c09630c2
67708c9b57aee5070c1406eb14e5965ee0fb1c4c
1883 F20110319_AAASAD chao_c_Page_065.txt
afa61294ed6ff1f3aa4e37b15f97e259
5c78bc0cee475c677e07b41a0445242fb8e1e78d
3102 F20110319_AAASYA chao_c_Page_109thm.jpg
652e0baafb49b9aa985414e2e6bb6080
ddcbb2869155446227aa272ba120ded6a5232d1d
6170 F20110319_AAASXM chao_c_Page_095thm.jpg
345c85283a537c9745fec714ad90cf4f
276874934840756fa57f89dc5fc2de61b601392b
8423998 F20110319_AAARUK chao_c_Page_031.tif
33f7d1bd01c0da3efddf2ccfb34d3c60
d811e0ff5e1bad2556eba97609c988983bb1253d
F20110319_AAARTV chao_c_Page_016.tif
21cf1b35f1a73a23e3d685cd3345633d
df6e1864a465877f26dd2f1c429766586290952d
F20110319_AAASWX chao_c_Page_080thm.jpg
428844366b3cf68275e2a94874ccdc78
c1de223404fdb461e1b262d385e799bc259da2f3
1174 F20110319_AAASAE chao_c_Page_066.txt
63c6e9a7ea7ca98bfd3a1932c760f1d9
8a554a7ff27f00061939f5be43c266964a957a26
2711 F20110319_AAASYB chao_c_Page_110thm.jpg
8e3915357ecf676a76ead4ff923bb84b
40b773e7d87e17639c61526f51f13ac2f0d1c55a
5972 F20110319_AAASXN chao_c_Page_096thm.jpg
01a40dc9655ee3bd3ca21c297e72cf4a
b3dc7f740104075e522b710bbefdb87f3a6c5a03
F20110319_AAARUL chao_c_Page_032.tif
e4ca8a3e35dfb601ffc8e2728ff3732a
a5fb1294285d07c04a38143ef76a03c645494c57
F20110319_AAARTW chao_c_Page_017.tif
24a8af57a96c0969bbf814826c3bc3d5
1411e27c776a9c6b609b9168913e4bd62145bb33
6116 F20110319_AAASWY chao_c_Page_081thm.jpg
8957a08f394d80eba2347bc0473e19b7
53057540341c043ba16db1940adee1b8e0588b2d
1619 F20110319_AAASAF chao_c_Page_067.txt
dd7effae63f57b58cd912b1aa07dccff
6de74af0e0efd0980cdede14bcfc858c6695afa3
1584 F20110319_AAASYC chao_c_Page_111thm.jpg
1a03880da262057ebd345d5db7a942ea
cc12d8a44520884b3998a587149f583eae048d91
6207 F20110319_AAASXO chao_c_Page_097thm.jpg
c5aa4fe9598cbf4837b6dc2195a81ee2
ebb0d8ca246ce7a260fe2030c10516cbd58ad381
F20110319_AAARVA chao_c_Page_047.tif
c9d838fe7dc1eaab66853c430690f81d
c6e97f874ecdfe3b68460488f582e26d60f7b7ea
F20110319_AAARUM chao_c_Page_033.tif
436a0423be57a381e13824742b7925e1
e474e1936eab6cb2b5deec58c691388ceddb77bf
F20110319_AAARTX chao_c_Page_018.tif
f916a0d120abf4a1bf8bc9c6d61990ed
8f379789ce0acb6fa37ed1acf6cb4f950fa509db
5815 F20110319_AAASWZ chao_c_Page_082thm.jpg
24b58ce49e1b7b928aa03daffe8afec9
d8f52072f17fec7c5a505098e59b88cf46c445f9
1140 F20110319_AAASAG chao_c_Page_068.txt
d543a6383adf329352ed4a2cac66859c
05c113eb9f50d1226485e2b1643fba92aa7a31ba
5939 F20110319_AAASYD chao_c_Page_112thm.jpg
611cf125c66a00b424f1779fd1cd9eb5
25839ec5bb631045a898bffe947dff7a87507f82
5886 F20110319_AAASXP chao_c_Page_098thm.jpg
e8bf8426be692246bc0f7f0dc988c5b7
445e11bda8fec9fb920cdeebbd5305628d4e7511
F20110319_AAARVB chao_c_Page_048.tif
756e239f1f60780997a861299bd0f289
7c68bdc8f67c7b77137888311a8ceed2e7dc4f0a
F20110319_AAARUN chao_c_Page_034.tif
697070c61699bbfbd86c8c328b34a5d8
73181b05e778eacb4d59d0136f925848ebb6f48b
F20110319_AAARTY chao_c_Page_019.tif
6c45cc117b59d3d03a878337b61343d2
b339659b850c0b6c32178cdf41af214863d882b4
892 F20110319_AAASAH chao_c_Page_069.txt
737da516b6c7774c6e4a72c7183c58a5
40c2c00ae383ba8f466ba1f1f1056a598fc7f8db
6676 F20110319_AAASYE chao_c_Page_113thm.jpg
e4a0b465a0c1056aab41c091be8a7f0c
122991d564439803e015e2a96327414a0d62fe0a
4713 F20110319_AAASXQ chao_c_Page_099thm.jpg
18e18f231fa6d4c5ae23084f1778c924
cd5e1431a4499ad9168a55eaa10e7424dd9cf525
F20110319_AAARVC chao_c_Page_049.tif
fc799493c1db7e6d9828fdceba55b4fa
084e3adec9a8220021f7f11655df5b365b15d8b4
F20110319_AAARUO chao_c_Page_035.tif
169e69e212140b1f96d2c8f22c3122eb
d50870a1d422ff51e3a08420351f5898c0c76cc7
F20110319_AAARTZ chao_c_Page_020.tif
f4d193bf803fcb5787b674418eedf5a9
063f4a2dd9cab9159f67d85d404101c2709b62a8
1768 F20110319_AAASAI chao_c_Page_070.txt
8439d576605fdaf61f5299d03a5bcadf
5098c625d54d0fa0298bee7fd44b959d0a9613b8
6673 F20110319_AAASYF chao_c_Page_114thm.jpg
ce178bced043d44071984ebb642e7ec9
ea657642c90b379190a46ad95d71f07a0562f158
5908 F20110319_AAASXR chao_c_Page_100thm.jpg
f0738b661dc171bdc1ee3cf7627842e9
6cea2637187682487ca99cd181b9a7d64a8631e5
F20110319_AAARVD chao_c_Page_050.tif
3268763be9348e649949a827748bac55
4e9072388d89d360593e2643d79ba77203bf5ada
F20110319_AAARUP chao_c_Page_036.tif
56d9b996f3ec49fdf6fa3557b8d49d7a
20f5f42cf01a28570b70c0af02799adc5d9bf945
999 F20110319_AAASAJ chao_c_Page_071.txt
1594d79af56112f4ca3ccc266771ce02
5edd57556a06dd959f7319526a38d7af5abbcda3
1713 F20110319_AAASYG chao_c_Page_115thm.jpg
22accb4b2fa1c3a58939aeb646b0135b
6609862f8ab59e648f00bedcc9967e40b46b89fa
F20110319_AAASXS chao_c_Page_101thm.jpg
a50c36fa34d6e31f46b367923bc6f141
2e20a96caff69d6d52eb114b452ea2a56c1cdf55
F20110319_AAARVE chao_c_Page_051.tif
ab8f3e0c19b6f90a7e51d553a0adc3b6
060bc1b39e4b63c9f4cab01b8860d60b6259aa5b
F20110319_AAARUQ chao_c_Page_037.tif
e3f83ca058e9f60b52b48cef08743c66
8ca6e680c2afd80e368f753f3aeff4079a55d1a6
1301 F20110319_AAASAK chao_c_Page_072.txt
c81916bdfe9011284675cd14fb44c143
249ced929ff32e5780ca3be68691070c150c9eca
2120 F20110319_AAASYH chao_c_Page_116thm.jpg
ed3e40d1bce81b2b74db153db2953607
d119f2578122a67259d06298933c5a801fb73e47
6178 F20110319_AAASXT chao_c_Page_102thm.jpg
8a775d6e6581b30ba2743a363db86415
a1f2fa62233771d69cbdf519d276642a9812f425
F20110319_AAARUR chao_c_Page_038.tif
62a53bbd9b452a7a1245d2d062df37d0
84cb5e5f628be1a08d919eaa3ee16cb074482fee
1379 F20110319_AAASAL chao_c_Page_073.txt
841a3758cf23f0eda54df9907ab668c1
4b61a5648aa33800f2e81c13360a07fffe5de428
F20110319_AAARVF chao_c_Page_052.tif
1894c4f713dd818324a717a98c908f12
e5ba9f858992d240d3025fba4f09023b47d784bc
3435613 F20110319_AAASYI chao_c.pdf
d8fee6c731f9ab922b56362b1594ccdc
06b2536c5153bcb44bc9d44f36e089445a9bbc18
2904 F20110319_AAASXU chao_c_Page_103thm.jpg
876ae09a29b6aa8d237de4a16672593f
84ceac5e4b0eedc5687be1f50a793b8f294908a7
F20110319_AAARUS chao_c_Page_039.tif
0b3e4ca504c7aeeb52467b35f5d09229
a790d59edef89962c2fe1be4122f703ab221321f
2779 F20110319_AAASBA chao_c_Page_088.txt
cb8c0548fecf8f1c5aca35cdbeacf0e1
74aa31cfb9fef4764a78c6716f68e048ed482c1f
1212 F20110319_AAASAM chao_c_Page_074.txt
e5d778745f165205b0097e05565fc28f
0688a54be1a6e4bef062a0fb0ae8783686df646a
F20110319_AAARVG chao_c_Page_053.tif
1fbe643b1c5cb4ab669d12fbfcbc3d76
c745d28e31ee814d37e627dea96bf4178fce622b
135406 F20110319_AAASYJ UFE0011835_00001.mets
9354d3c479e5aa813fc2a9066928b08e
fe1798e90d4b8601dd2d2a84bedd7aeb508bb8e4
1415 F20110319_AAASBB chao_c_Page_089.txt
a4fc3b1068e15aefdf0b3fb705539c18
7bd846ee25c7ca0b5aa1c74f376fe95c5f74e144
702 F20110319_AAASAN chao_c_Page_075.txt
4412b04a6bdf5e4712bca9e5ab554e16
0dfd248901dd201df9c44b189c47e5e0d991482d
F20110319_AAARVH chao_c_Page_054.tif
ee8fda4bddb0c10bf7d753d605bb3a59
d25f61186716144d0893cb5990fab8f24284a3c6
4776 F20110319_AAASXV chao_c_Page_104thm.jpg
4a6122c0c58914338b56291ed761f6d2
ce84cb00a41896e86b8bf94368516c923165ef61
F20110319_AAARUT chao_c_Page_040.tif
df2c94120c91d6a00f59b96e2a8b6da3
2f49760e50c6f9b96534afa6f75564b1c40ea11b
1809 F20110319_AAASBC chao_c_Page_090.txt
79a55c5443991d1d43a25f93535c7122
82ba675a590f8520794002a2c1aa25f610940e5c
1721 F20110319_AAASAO chao_c_Page_076.txt
0ea939f6a12f2a4bb3cd9fe76e85fab0
d72c894eb3b2ee9feb6132bbeb1bb9ac1368adc7
F20110319_AAARVI chao_c_Page_055.tif
5b08103288e10ae60c6914a065579843
2738103ab48a6b30a75436918f4795ae6c59a0e7
5825 F20110319_AAASXW chao_c_Page_105thm.jpg
9eb8a280d84d52810be09acfcfa350b9
2749da12c9046aaea59158bf67407c167dcc4a6d
F20110319_AAARUU chao_c_Page_041.tif
9e250185cf5e1db3506d689744063d04
0917feac6732f1501549dc377df5f135980f8acb
1955 F20110319_AAASBD chao_c_Page_091.txt
85889b629fd3824c0299679cc3194f7f
6cfa4356ab3e0b9cddbff7c764df4a98165086c3
1743 F20110319_AAASAP chao_c_Page_077.txt
fa8ae3cb870c244d9ad9549f35289b99
554e43360c4a4329028172f2416d820ce4638305
F20110319_AAARVJ chao_c_Page_056.tif
a69df4e2ee69762f60a0ff3673ef3fde
6373bfb0ffbbf3218d9285a3fe4615135becd02a
2844 F20110319_AAASXX chao_c_Page_106thm.jpg
554165bfd8a6bafaa4a8b3b87b4697d3
f11dabeb3b0e374a27ae326866e07d877c3f9dac
F20110319_AAARUV chao_c_Page_042.tif
1b21af99e0b9e525849f0294c7973a4c
ba409cb1a4ab15df49f9f0f342299706111a0dfa
1562 F20110319_AAASBE chao_c_Page_092.txt
2dcd1a6d256194ac96b03631cae0c2f9
29f37bca22fc14b7e22af9d43ddce9f0675cda40
1850 F20110319_AAASAQ chao_c_Page_078.txt
13426ea68725e5d655534c2448f938da
caf40d86a1bc2b3921ecdc3199a8175fd875f4b9
F20110319_AAARVK chao_c_Page_057.tif
840630a3316af1259824cc127c99440d
8c7419ca995ff5adae29da8dd74bf159c9224e86
3054 F20110319_AAASXY chao_c_Page_107thm.jpg
51aa60e367293f746d709a1f6784d613
e22a77ad771402e0e2d14fbac2a73bb85223b903
F20110319_AAARUW chao_c_Page_043.tif
9b70d1a8df63a95014f6b98b00792877
e97ee025dc87ea1fa5757319512a78724a4f304b
1934 F20110319_AAASBF chao_c_Page_093.txt
0d9141eafc7c2302d74d164d6cf217a4
b0db1ef9c79ae04ab78be3cc369715fc88b8df43
2024 F20110319_AAASAR chao_c_Page_079.txt
25e7746b8992a4ee966ad9ff61787de4
a5163d9e6dddfa21b0891f7d2fce2a15242942b0
F20110319_AAARVL chao_c_Page_058.tif
8a66b346f4e37b9775c32aac1e97d9d0
8884f0f9ff4f4380360750881a9d9bb62194ab52
3053 F20110319_AAASXZ chao_c_Page_108thm.jpg
5ca00ddca338706b83a14ff952b6b38a
483e1b6d0122f3618747a72d1579374c31d4a300
F20110319_AAARUX chao_c_Page_044.tif
fe4704ccd6730db08e8d9ae4fc64b901
54a59918ef5f65dcc027b052f409237bd4d34a2b
1937 F20110319_AAASBG chao_c_Page_094.txt
e26b7447f0695f8681e054dba63a8978
c408543556e7e63285ef3b41ef274c1caef28531
F20110319_AAARWA chao_c_Page_073.tif
307e7a2b7b85d984bbfad7bdc75627b2
250e3df5bbe06451b36145e75eb74bcf81cf5f4a
1854 F20110319_AAASAS chao_c_Page_080.txt
d545941f4c0d468bd064e0d913c97bd2
94a1a2008164283e117286acdaf8b269f819c1e7
F20110319_AAARVM chao_c_Page_059.tif
2235b9758835ed3cc21464a6573394b3
37ff7d93ba73a20bbe896a17554d06f0d21db0cb
F20110319_AAARUY chao_c_Page_045.tif
1e7c6c8eaaaef3bfa0f06015f3c437f7
82c247bb248d976a9f35bea35c615a7ca883d62f
2057 F20110319_AAASBH chao_c_Page_095.txt
7d21418883c5e22ac74f9799dcb4881f
e90f744c3ff53b4b8436b5e6cbf287c5210ef88f
F20110319_AAARWB chao_c_Page_074.tif
274ecb21e8c7733455b829e364d018e4
3ef65aeb8bfd90b87a25904941e54df04f645f9a
1963 F20110319_AAASAT chao_c_Page_081.txt
eafa2158f29333bd8591d12aa6ba4fcc
aa59e65bc5020dd2c351445aa960f59e8a566a70
F20110319_AAARVN chao_c_Page_060.tif
4716aee44fcd93808e899fba84b7c72a
37f74e698c0081558430e55c28970ba223403fde
F20110319_AAARUZ chao_c_Page_046.tif
13b42ee945841078af55ffb7c5c8e406
d26037c1cec7c2cbfe0a8e19142a9d0455b6fe7c
1735 F20110319_AAASBI chao_c_Page_096.txt
14352fc1abd6e417ce22706d10a26581
6191e4a78ceee50320c01aee93c5989e1c0d29eb
F20110319_AAARWC chao_c_Page_075.tif
22e6f6bb088ba1b2a0fae0ac8eec1152
3c0aefbc07281dc1adcba9d87c716869cf5e4505
811 F20110319_AAASAU chao_c_Page_082.txt
d1bc97dce3d9bfeabb2a71f55b0635fa
8948691f78230ec835682798cec23c257dcf8a3b
F20110319_AAARVO chao_c_Page_061.tif
f4c7c57a586b2d4cda7452ecc7be7a81
4c4bba821fa4a054d106bdc9d6c08d9ba56439fe
2109 F20110319_AAASBJ chao_c_Page_097.txt
d92d7cdfa72f95af8cb09c455a6bfe82
fb0a0c1a056af1dd7fdbf6646a2fb4445f3e3611
F20110319_AAARWD chao_c_Page_076.tif
41969047dc5c150803417f3dc0816afb
5c5431c94db0c5ec4012360b9a8c4e23d25e0c65
1651 F20110319_AAASAV chao_c_Page_083.txt
5e71374b05e5ec64cc869d69656b636f
9e1b64c39718c0e2194653c8b0f93494676e8d13
F20110319_AAARVP chao_c_Page_062.tif
7e193047672e69cf285c1e6d41fb3616
b9eebf101f7515682f970fc2e6106d8891bb3630
1166 F20110319_AAASBK chao_c_Page_098.txt
7a8d9f927939e0d774d817c133e8e299
c875e3e53de8316153c0b870fcb3565c9146c7c9
F20110319_AAARWE chao_c_Page_077.tif
160f88b6959bf8ca6c0eb05e0cddf27a
d0858be9397d99d6052466b2889329601db41186
1627 F20110319_AAASAW chao_c_Page_084.txt
28a6fb12f8a6997b90bcd6ec3d2f1337
ab5f83fabb51454b2e00b5e03e88cda617765cb1
F20110319_AAARVQ chao_c_Page_063.tif
b3d4f1ced2e0662b9d970959b8eccd03
69d917ef6808ce079dbdf57cb6c6019f7abb008d
1372 F20110319_AAASBL chao_c_Page_099.txt
12cc32f3b43781efab816b2e26569b03
8044ba63403d4c1bbc7f82d53b4fd82fd61eae64
F20110319_AAARWF chao_c_Page_078.tif
bda5dec6f66f51efeca855510088b716
118efaa5a0e9779f4572c24587b23f72bc475d7c
1645 F20110319_AAASAX chao_c_Page_085.txt
86b3ae7c5feadd3420ea16f45dc72e5d
8a597c48fed7f08ec7e0b7a5229787203d8923c1
F20110319_AAARVR chao_c_Page_064.tif
826c23a27d26e0693607e310ce41415c
4dedfdbda1e3acb71510f493f71a58e24f1d00f8
2303 F20110319_AAASCA chao_c_Page_114.txt
c57779d1d84974b8cb0c7397afb8688d
652c90839c848c123537ce872c471b39f16c3dce
1873 F20110319_AAASBM chao_c_Page_100.txt
b4c66fe6a4122361d83c44b34ca7b98a
bc377f8b76c33bfcc4551d7d3ab6accc6e1e6329
F20110319_AAARWG chao_c_Page_079.tif
d1cee3228a41f97b3ccdab1b637e800a
ff7cd8eb1cb67197f7f54ae53e0626caf44e51e8
1460 F20110319_AAASAY chao_c_Page_086.txt
5f8082fdf94724f58e51fa3f6efd2165
f2a41e0c64f23b17eeebc3546521d89b90246c59
F20110319_AAARVS chao_c_Page_065.tif
2277fa367d5b082777a645154ee1c6f6
3d6521685c406a94c18aea0ea61c63153dc813c0
478 F20110319_AAASCB chao_c_Page_115.txt
ee4bf0a29df6f61c5719c5dcd020a280
2b0e001c739b6fb1ccdb024fd893182f6435960e
1479 F20110319_AAASBN chao_c_Page_101.txt
e268a0f18b107c770558c2fba1e01d6c
2a15cd1e0ad316b2794688f248e2e806e799b05e
F20110319_AAARWH chao_c_Page_080.tif
f6d3ec03460ddd84cb7965f7a8059e1c
235e8612c48e92c3dde797c94d40161b3c14e232
1907 F20110319_AAASAZ chao_c_Page_087.txt
46d8163c0fe6dfa6358e24fe177b1c39
9869289e671e46eceed9b39b70fff38479986dc5
F20110319_AAARVT chao_c_Page_066.tif
40d030879ae764af4bbff3d0648aa88a
2fed2d56b69b612761c502daeef6e016dea690c4
1860 F20110319_AAASBO chao_c_Page_102.txt
bb3835a6c05e1b6c79f67f8ef639e221
f445c308881763e0531bef9bc2dbb2336ef41785
F20110319_AAARWI chao_c_Page_081.tif
c465b7e6c26e5c76b4542b9ad7229317
d2a450976ea5c4d26af06d9d9a471f738ee0c6b5
564 F20110319_AAASCC chao_c_Page_116.txt
504cbf4065d2f65c5d5612ce6c7c0805
83aae93c92d1abca3084af3f363981e0658b253e
821 F20110319_AAASBP chao_c_Page_103.txt
a77085f7acf9685ab6ea7b0e8ed90947
62452a676917befa933c90efd4a0a36acf6e44a3
F20110319_AAARWJ chao_c_Page_082.tif
fab87612cef63f195f3e26da9724a2a0
a19b3d4069178aa2e87fba621aadadf07c3b3ce9
F20110319_AAARVU chao_c_Page_067.tif
d9c0c4cfaa59d661636294bba47db95c
ad4c396d26c3d2441a952bfdc73558a0c28fc7a7
7654 F20110319_AAASCD chao_c_Page_001.pro
4ec821898abe2319586df5539a377bec
1d38bf5dcd0a51604cf6cc9f80de661f4dc0255e
1603 F20110319_AAASBQ chao_c_Page_104.txt
4c9c80d37d60017e9180973a1a9e9081
a84089d40aff31ad5dd58d780f8c3e052929d711
F20110319_AAARWK chao_c_Page_083.tif
58beea8aed206a91d7e4d51cd4c4f9a4
6a19942fbfd29ca6a03441b3f46b6dd1dae0461b
F20110319_AAARVV chao_c_Page_068.tif
40ca9e56efeee82587213e3a0dbded0e
a6afd827feaec9b5d7b4d010490583d577a392eb
1147 F20110319_AAASCE chao_c_Page_002.pro
a9806fe047969546bd9537da2714369b
c83a64b1208ec881d5bfdaf846de4e07db14d4c4
1959 F20110319_AAASBR chao_c_Page_105.txt
3566cb63cfe12d723e2c4662a3161ba2
5f63890c8c38d3e688c1ded022b8c3589e800a9e
F20110319_AAARWL chao_c_Page_084.tif
08d1f3d952fb20db30acc7a0cf76e6ff
30a9598b1ff9494e0faf37c0c0b1fa1393d96e00
F20110319_AAARVW chao_c_Page_069.tif
5aeab6901ee05865e5f7287c0ba71e09
712320473bc35e89b00a198a8e1c0786ddfabdb1
1507 F20110319_AAASCF chao_c_Page_003.pro
621e2f1a7cffcf01bd027de3d22e9f64
8e9ab00e9ce28312f5d46b7288a2db4eac0753d0
F20110319_AAARXA chao_c_Page_100.tif
f35712683b049c3ba3454c00ff256ad0
ca0370a81b91b19044c321b5460bfb2b717f5874
820 F20110319_AAASBS chao_c_Page_106.txt
aa3a657173d56913d8e7e647fae416fe
b0d25f92fc2589625c27fa2b1ee8b81a82efc269
F20110319_AAARWM chao_c_Page_085.tif
96fdad49f61d47197e4e370f61754d44
2688196c219bf6e561d508fe70ae0b92c03b6ba4
F20110319_AAARVX chao_c_Page_070.tif
27f6e32a1586f81270135ce6c0307c48
67b0a1fe4c619f9f6077a595553aa9c58244997f
27989 F20110319_AAASCG chao_c_Page_004.pro
96712fb7f4c9cc136939ba106a484724
361b72d61461a7468d138509d6ba07796c597c9d
F20110319_AAARXB chao_c_Page_101.tif
706ee8067387bf01fc620b0767c98908
501ec3f814a32572a201a39ea3bfcd760426845b
724 F20110319_AAASBT chao_c_Page_107.txt
9c385337668270fdccf787177a273d32
c887daf8fcfdc20e1193103f0a927471b2075979
F20110319_AAARWN chao_c_Page_086.tif
abcaa0aac5ed4d0a80a89bde006d8dc4
30565b8a5851512c4160b287981bdf7d1fd3d4c3
F20110319_AAARVY chao_c_Page_071.tif
c243778b525d690f873f6fdb598b1ff9
41b428c4718f274dcd88dac8d4592cd6c56f4b83
80021 F20110319_AAASCH chao_c_Page_005.pro
f907492c1210d8aba6dc5834b25efa63
27dd0fcf34628c24d4667be7a29dc44c33452ac5
F20110319_AAARXC chao_c_Page_102.tif
66fa6af54d1c874d8f188667865065ca
45d09c2570a92bf51d7606e94b3ec1f50986e6fd
739 F20110319_AAASBU chao_c_Page_108.txt
3ce6020bb07c2f9b350be6a23fc4b03e
75f4ce3db25a2ccd6bb96abfeeeb56699f7775e4
F20110319_AAARWO chao_c_Page_087.tif
6487ea8389cf0ace85d6506b25810d43
311f02ce960ccb3818a5292b480a607a488b9c90
F20110319_AAARVZ chao_c_Page_072.tif
9ed9b6ec46c892c9338dd40c4bc79dc3
b00e86b08828d6216b7af3ed111c07bcc558cad1
103095 F20110319_AAASCI chao_c_Page_006.pro
7e39abf1d7417386b6406bbe7f8e8372
3cbe1e3ecf224d137d292f344817de420566ea36
F20110319_AAARXD chao_c_Page_103.tif
35153d892060ed0570035f0bf8102758
b2cc58bbf856dfb861d8c955ec3487508b645a9f
719 F20110319_AAASBV chao_c_Page_109.txt
7887fc1a2bba052316d4eb04465101e2
d2e5d704ee2d066affc7fa5b78d07d0bd4fba91d
F20110319_AAARWP chao_c_Page_088.tif
0135f0206599eb8ee9c69f92289bbd9e
79c1ed3512b2004b9de75ae7ee499dca35459d06
3151 F20110319_AAASCJ chao_c_Page_007.pro
cec75dc8cbd203d6719a5e6eade47ba5
c181a72c76d04d0eec0541ea2a30dbe4c9baa88f
F20110319_AAARXE chao_c_Page_104.tif
7b1517ab8f08c5c459c2e75d7a395f65
e13c265aa035e221919329af2b8608bbb603c5e9
595 F20110319_AAASBW chao_c_Page_110.txt
dd1a759401bede32d5a5a89a4876f7c4
5242e2566ca29cb6fa623257b0c75a9e2be019d5
F20110319_AAARWQ chao_c_Page_089.tif
4a3f26744b1f9e35bce35f8f9616b4ab
643cd966e96de4f5722c8310128f57cfe6f1adf6
31235 F20110319_AAASCK chao_c_Page_008.pro
09033c65e32cf3a85d74145febfc65ac
e356414e170ce519c2ee81bbe5ba672833bb507f
F20110319_AAARXF chao_c_Page_105.tif
e66f14bbbc9de648d8907ef8438b6e1d
8f7aef95ffdefff9deaf7915a87b22ba990d4573
291 F20110319_AAASBX chao_c_Page_111.txt
3be819d2cf82cdf07518dc2c5706a1fe
bbf980a2a6a749b41b352ca1c1162cc91c5241ea
F20110319_AAARWR chao_c_Page_090.tif
2e079305b78c1f3f3e3b99d37693ee25
66117fe5fbb423c9a9807cdd9f9a99d40788553a
19339 F20110319_AAASDA chao_c_Page_024.pro
6fffe95ccee8d10143318fb5b1a34769
cb23e6f035e54e4a4605a50913173d416e425f76
59530 F20110319_AAASCL chao_c_Page_009.pro
ce4bcaee515d31fc6243eb2c92ccab11
a5dc0036271a209df1428775659420e4a3d7e0c1
F20110319_AAARXG chao_c_Page_106.tif
2d5ca8e463438c3fa36269181dfcca18
62b331458b5f3e5d126926f3ad933cc1c3254f52
2119 F20110319_AAASBY chao_c_Page_112.txt
ef7adc3ee231129f220884f4e638692a
3acc6b880c253e61799fcfbdedd85daddf003d08
F20110319_AAARWS chao_c_Page_091.tif
b4baec45460b147f2021085bb42168f6
d1be89845b96d2845a9fa4677ffcdcacc9a4e286
71022 F20110319_AAASCM chao_c_Page_010.pro
b84bef1f4d0890d5cd677a1948aee209
85bee472b396ebdc47437bc1c9eefdfea4473f86
F20110319_AAARXH chao_c_Page_107.tif
eca819059c3c5e843a95925bfa2953cd
0ccd0d840154c355f325cc5f6a3bbb3fff1d16df
2552 F20110319_AAASBZ chao_c_Page_113.txt
4f8d45c52913d7a8fe8227b30411d6ad
d937ef8c40706b2b871b48cce593cc04c2adab2c
F20110319_AAARWT chao_c_Page_092.tif
1a1a5725ec40326f680248fcda60f74b
fb7938d9177953fece460d9a29dfd666c6be5318
38459 F20110319_AAASDB chao_c_Page_025.pro
c17764113a814507517967fb05e7481b
26caf147e4da8221f15c4219f95c942067d4b4dd
11448 F20110319_AAASCN chao_c_Page_011.pro
605c02279af6875e507123b90c703fc7
4b1c65434fcd2cc5209f63fa6301985017d96fd5
F20110319_AAARXI chao_c_Page_108.tif
22099873686b5bf1789da7cf939b558b
8256910debfd101c64302691e05708e38ab46b99
F20110319_AAARWU chao_c_Page_094.tif
15f3b29e3a20c1ed3635d8cfd0480e1f
b492922e29c29475b524d70fdb257f6b9e9f61dd
46099 F20110319_AAASDC chao_c_Page_026.pro
c3e9aca304cd7efa89902d22b04141ef
54382fdeb63409c79df3f2f8ab5067c6e4140248
36700 F20110319_AAASCO chao_c_Page_012.pro
0ec5c91a9fe51d7f8719de04572ef0ad
f00a5f6279ec65aa84adb4ebf3ce32b3df1ede60
F20110319_AAARXJ chao_c_Page_109.tif
9218808c6a514b04764ccd9e1e8dbacd
92813cc5b6fe2365b8e3889f88205e36861a171a
49963 F20110319_AAASDD chao_c_Page_027.pro
4c7ebce09c103cd6b371b74c5c0d51d1
3da15ae97b4664e087c719de25dd5cf2f8473400
23331 F20110319_AAASCP chao_c_Page_013.pro
4d6207cedf5d6f06c64fdab9eefa87cd
5558dff72b472c4e28aeac845200707656c6d08c
F20110319_AAARXK chao_c_Page_110.tif
53c6c0e804c75e7426bb1150a4d9a536
210d7c8801aea9172909f5db754664570fdc78e3
F20110319_AAARWV chao_c_Page_095.tif
5c9e38d39da5da292911340c7a13d8f0
0d2005758cf4475f6174270ab62726c88efcb35c
52532 F20110319_AAASDE chao_c_Page_028.pro
2f83030c974db9ac8e51e8ae4b49844a
32203a1fcf67daf8627b8a4759ce2fdd9e302581
43493 F20110319_AAASCQ chao_c_Page_014.pro
da09fca9f599d5784c0679e17ea43107
6d59a3c688ab0454971f3d2c1673f96d4012d20e
F20110319_AAARXL chao_c_Page_111.tif
e93f448bb3c66361c340e7b8c1288b3b
d1f22d0a2fa6d627cf9bd20fd61921e947555233
F20110319_AAARWW chao_c_Page_096.tif
de2363301291be3102d6e91489030365
854b0fb931b70512a2db9529272044f5ccc7d57c
26468 F20110319_AAASDF chao_c_Page_029.pro
eb6e42385eb8055a2bd29a9294a23c20
8165206c4fd05cfd9a3921f619e6411ef86de8e8
43379 F20110319_AAASCR chao_c_Page_015.pro
94a3662e11586a0e6425e6c50baadf4a
56dd14d852a603f7acd0996b82ac8ca975b431dd
F20110319_AAARXM chao_c_Page_112.tif
609a15291c371dfc59051fab647b6011
85fddee3017775d682fd138280444f1e29bfbf04
F20110319_AAARWX chao_c_Page_097.tif
81beed2bed1444a149dcdff2a99d4954
a0cfb8e718854e220daf17eca01da8eee3ed6e08
43579 F20110319_AAASDG chao_c_Page_030.pro
907110de5a956307898b52e3ba5257e3
2515931c0ba776064563db3dc7e6ef1ed2909c18
2816 F20110319_AAARYA chao_c_Page_010.txt
65954df717825a902eb62b4b8d5db1fa
5b9492c55aeb99ae407e091f2cd72d2cad95a0db
30931 F20110319_AAASCS chao_c_Page_016.pro
a18611716ebee03d6a713be49e3bafe9
1eeb3eee421ed8e3e020f4077f7fc131dc6d5f3b
F20110319_AAARXN chao_c_Page_113.tif
736dbcc0dd851bee09529137875553b5
3187dcedb21409f0dbec9dcc3e7167da5b9f4e6b
F20110319_AAARWY chao_c_Page_098.tif
916b95c3fee1d18016b4f8fbbb7206ca
3eb6c643838a1f2ab40ad2f559720a97d3b6afb8
47591 F20110319_AAASDH chao_c_Page_031.pro
eebaf9d92ead25785a2006d627dccaa7
162524a37e4c891e996c0e71450a4fdc1b406657
453 F20110319_AAARYB chao_c_Page_011.txt
1496f619e3c1e9de118ed28491aaadd7
a5502094e24532fbdd6e8f4aec226f8830c9a329
42585 F20110319_AAASCT chao_c_Page_017.pro
9c01f79af951c778fb5bfc1413e16dcc
099851b30e3047df205b010b7c808ccdb47deae7
F20110319_AAARXO chao_c_Page_114.tif
34f9d26fd9819865dedee8e6997d337f
bd9e13d5f0c06d1f4711ef3ed1e5f99a6fd2f37a
F20110319_AAARWZ chao_c_Page_099.tif
0e35b86ea263a784df707413d44aa62e
1368f3c5deb11477f50de6f4c9728f8abfc7672b
28778 F20110319_AAASDI chao_c_Page_032.pro
823afc7f08db5fa01bd779de391fc646
b8714e7ece668c255c2f357f45f4dd5221acf57f
F20110319_AAARYC chao_c_Page_012.txt
5b4b2fce4b3bb95de98c295a602bb203
7176c15f03ae54669bc10920bd3fed82f7a2cca3
21658 F20110319_AAASCU chao_c_Page_018.pro
d959befb18fbbe1f20d28d9471826110
9760788c3aaefa2bd016a43eb572d377a951ff5d
F20110319_AAARXP chao_c_Page_115.tif
3f55adbe572ec5a9cb7a78eab8455839
d4088872e8c406eaad8275cd71a6e671dab41140
19269 F20110319_AAASDJ chao_c_Page_033.pro
019a36fbcaa7af7e6770cda807ef90bc
9d76d160dfde443c6bd4e4e0ef7ba91e6da85ec7
942 F20110319_AAARYD chao_c_Page_013.txt
de61fc8ea97bf5645c0b3fe54fbfe0b0
3e9f060704b5a05d8c1c735e639bac9cc8d00ccd
26194 F20110319_AAASCV chao_c_Page_019.pro
fdf653b6918b46d61eee1fbb59644cfd
9048b16403c882cdc9ba21fc2f5db71a861874ed
F20110319_AAARXQ chao_c_Page_116.tif
1f1644a319195d136219116041b9e232
6a94171aab2d745a30b3c9e1d9965c80997be941
28408 F20110319_AAASDK chao_c_Page_034.pro
7e6e773dc38c491e1713c60e81e71ce1
deaf206f56f9e45d347ec3ab8969cb5714f2443e
1821 F20110319_AAARYE chao_c_Page_014.txt
d6e51b0476a97b6e0bacca167f9ce889
d67770591bd7d8d092954b8d152d7bfbb911fa5b
30224 F20110319_AAASCW chao_c_Page_020.pro
5ac0ef7f4fffd56db8828035a68f419c
ae48b6c99013d11f156791377590f5fea6a4a8ec
460 F20110319_AAARXR chao_c_Page_001.txt
0b437bc2843af66ac67cb0914a7381c3
7dcd7f1c287cd91095114f0294cd630fb7cd22cc
33827 F20110319_AAASEA chao_c_Page_051.pro
dc87eec69c82236d1d68de9496db0990
66c08db9154b6f8f3b1e9d4ccbfee9270abbce61
39312 F20110319_AAASDL chao_c_Page_035.pro
e544d275312440483bffa57fb3090054
0ad0bc4c381a4ffdd48cebf1136727e5a74b86ea
F20110319_AAARYF chao_c_Page_015.txt
471faa0bd8c294f24688483d2628a804
6ea46d89f0f9be40c8871f559e2567e0dac905c1
34847 F20110319_AAASCX chao_c_Page_021.pro
9ab8adf2e237cab62bbcc7a00ddffa68
826c86365202a81e8bf21b2cc03b8b3dd6893ef5
109 F20110319_AAARXS chao_c_Page_002.txt
29ef22c327ec20de72638f67b6f60a25
8563bb54da08d526e8fa1007cb566044f89194c5
52853 F20110319_AAASEB chao_c_Page_052.pro
791d1c26171b6fa3657927f0a28083c0
69d6da525830c87f6e8044e6f2324cafa606f126
27648 F20110319_AAASDM chao_c_Page_036.pro
d0dc8bf8a1c30c0612834ce21debf80b
c0e36b07e30d0153726736e8393001bd2565b773
1499 F20110319_AAARYG chao_c_Page_016.txt
dff21cb1246b2a663da42f9c3c74c896
2046a0b87ddbd7643ea7f33c6e10f462c6528624
25751 F20110319_AAASCY chao_c_Page_022.pro
9b8a7a3951b8806aeb21752a3ebe7cc4
602d67774cb9e77d4563b3513da25261bcc2f536
112 F20110319_AAARXT chao_c_Page_003.txt
61be5f126e9213ded9583a613b9c59d8
1bf9a47a190e6774483ed377ae19c88f2531d2ec
41806 F20110319_AAASDN chao_c_Page_037.pro
c1825fd81d506203fae69b3d2ab14988
988ddd0126165cfbcc265de6089298df17559aba
1698 F20110319_AAARYH chao_c_Page_017.txt
d729f45c091523b2b6646e74726a1278
a149be02dc5f51ebdc94c11eeb3e62de7051b1b1
29106 F20110319_AAASCZ chao_c_Page_023.pro
bf4520f9922ca11a643ec645fb770b24
bdedac54af36605375fe76924a7700b153bbb7cc
1173 F20110319_AAARXU chao_c_Page_004.txt
45adb13a4e6d7322c95235bbe38f567c
0068788f5785e4ef09d8fa534aa3c1c2865d858d
22642 F20110319_AAASEC chao_c_Page_053.pro
541379bd95599b103dbf586455fe22de
3f78890f10b800f8f3b55e7d3fbd9728218cce5f
48982 F20110319_AAASDO chao_c_Page_038.pro
4e86404bee077353497d3811c42e76e1
905538679552ccbe825ffdfd725fdb4c76bef2bc
906 F20110319_AAARYI chao_c_Page_018.txt
50998510a08e1e06c6ee277bcd489c43
674be122e584821f7041e7547e98179c6100d147
3308 F20110319_AAARXV chao_c_Page_005.txt
a2026d983b3f19a8fd70d103933ca5dc
481a4be4b25af733c91b38bfce1865b0e576d2dd
43669 F20110319_AAASED chao_c_Page_054.pro
41e768b918dabe5accc9aa6da553f00c
36058dff5baf77342e870b1f7df24f034569932e
5159 F20110319_AAASDP chao_c_Page_039.pro
2e563d16e878b87952082bdecb99b41f
8ed0ef1c8c4308516d7c35fa8a5de7987b9a5874
1040 F20110319_AAARYJ chao_c_Page_019.txt
622b2a8eb9d14950a0c1a320e804b131
46a8b51e415e4d067383a1b0c184349b7d895261
45532 F20110319_AAASEE chao_c_Page_055.pro
fb7644742a61c3b150f08f0582dc5f47
59958f8288c4ff0affb42a287e3d1609f780d4bb
44052 F20110319_AAASDQ chao_c_Page_040.pro
d3c9fe6348462b87268cc98d1b6fda4d
7ede16c5c9dc6560b816e149e247ae002608ad2a
1238 F20110319_AAARYK chao_c_Page_020.txt
ac5e109144556d84d84506e5784ec9ec
cec7aded4d29cf9d06cdf07dedde905ab861db7e
4146 F20110319_AAARXW chao_c_Page_006.txt
a3ae64a310d452c7ba873ed3f8c28416
7646ec21e2a0ffef3dc4888c4af8697f5b8dc860
41632 F20110319_AAASEF chao_c_Page_056.pro
b87a53791dafdf02c810897909292d55
a8c20fb6d94b959d6d550f3cddf6167720f3d4f1
47630 F20110319_AAASDR chao_c_Page_041.pro
3f12f1383c859b526de27c2c1cd3200a
42a62d5a18e09b202d3d9dbafeb5a57ed1b4b262
1411 F20110319_AAARYL chao_c_Page_021.txt
cfd0ff5b91318c0f30ba70400b584358
701a39444ab102f6500036a70bd89491156cea2c
130 F20110319_AAARXX chao_c_Page_007.txt
e8d176445247e81e1bf945270da8bc3c
e5e12f2b6b99199f6f642d0aeebe7122af9f12e4
14604 F20110319_AAASEG chao_c_Page_057.pro
4f17ac62ed53517db3a7a072e61550a4
5d1ae5fe60a02d9357c64c87089eaa3960c77b43
1438 F20110319_AAARZA chao_c_Page_036.txt
fd1abfca13fae396882e9bbe895a0aee
928bcbf099acb3c1658a0d5db370ab40024963bf
47257 F20110319_AAASDS chao_c_Page_042.pro
2859213334d2111673cc550819ead71f
465fdf26e59368bf0eb72851c3d00655a0617048
1246 F20110319_AAARYM chao_c_Page_022.txt
7b4ae13e0432dde1982e5f74df48b7d8
4c062fdf1a100e9ef73c74825e29438905cb0c7a
1324 F20110319_AAARXY chao_c_Page_008.txt
baf9fa0d36a26b1acf77e5ea80ffbb43
b97b643ebe639dd0a284463c6c755bb4f2d726f1
33167 F20110319_AAASEH chao_c_Page_058.pro
21e8ae0cd0f1e663be25b83689c6efa0
a547107f4bb3dba7c5f74758432dacf262e17822
1686 F20110319_AAARZB chao_c_Page_037.txt
fde65cab1f7f9480266e2307389e4398
b990f44ad3328bf00606b3d2624308dc15da94e2
40193 F20110319_AAASDT chao_c_Page_044.pro
37444069630729b832df8a88a6096a71
492adc7b340b2a79e7076ad0d5b1be6f67097d14
1264 F20110319_AAARYN chao_c_Page_023.txt
2609ca1ea3beb2d5f30a275abfdc36fa
8a708f25bc3128c34f2b0270a08d40cb1a316087
2380 F20110319_AAARXZ chao_c_Page_009.txt
06e8946270f2754b4050c56ed8e26e92
825897102c763155e392af6d471c3e8a9ffd860d
41461 F20110319_AAASEI chao_c_Page_059.pro
566f623299b845f59f22d2f0f133e414
122e9c09c1a65fe73e86af1a225ad1d9e8ef13ea
2101 F20110319_AAARZC chao_c_Page_038.txt
8e8452bc7bfa385bd0f40e1bc6d92b07
c16e673b975d5fdb387854e315b32095e1c55d65
50505 F20110319_AAASDU chao_c_Page_045.pro
854ec5e8282efb397980ac00285d9f77
4ffeea1392b7e586ea8ba4aafc01877645757718
838 F20110319_AAARYO chao_c_Page_024.txt
4d7309909b1d070123baa436f76814a4
448906feceaabec9233e77e1def450e893f64bca
30342 F20110319_AAASEJ chao_c_Page_060.pro
bb512072a7d46ed07f42559308a312f4
5f6a50ad731399c164c5053b58dce5c6381f061d
249 F20110319_AAARZD chao_c_Page_039.txt
0c24a674b9f44b76265d9c7a94163be6
3aaa5fe3948686d27173874eeca7d2436069abfb
23104 F20110319_AAASDV chao_c_Page_046.pro
2664166463bfc2f53f901f3ef39f7bbd
aff68ec47edd7657ec524ce5d60246f1c5d68182
1725 F20110319_AAARYP chao_c_Page_025.txt
590d28d3573756b47c198b2003ad0541
afeb75295ca1d1fdb9210394f132df0dcb09f7d0
43955 F20110319_AAASEK chao_c_Page_061.pro
5461e01543cd7b420c1fc548e73e8efd
81ec9f2afae9a99fd10681cedb10013f970e8cc8
1803 F20110319_AAARZE chao_c_Page_040.txt
e4c9eea840e9df896e4f71228db011f3
55b99365cdfdde6176d897647300359b68be6542
49092 F20110319_AAASDW chao_c_Page_047.pro
629f2e4db268ab1040eb396a91685f85
7719d5b0cf32398b1d1a7749b81ebef9407c25ca
1848 F20110319_AAARYQ chao_c_Page_026.txt
316376af6ce0d1f9c98485df6035867b
f23508ba96c0556d9754109f4d7f6683acb2f081
50243 F20110319_AAASEL chao_c_Page_062.pro
06beff3565b5fe7dc920a3c2634f839d
0da25af91ff70cdf93658f9f3243dfe8be744396
1905 F20110319_AAARZF chao_c_Page_041.txt
2938229c4b2b9b1d9162161905581427
34560c598a1cc14779c86c9e044c3e20dd820d8c
16915 F20110319_AAASDX chao_c_Page_048.pro
a27d0b40718632f18cfc4a4d8aa96d32
6676dc1a8958298bc88b110cfaa98ce00f5cab4b
1968 F20110319_AAARYR chao_c_Page_027.txt
f6f894d934181cab7b9bb803db076eb7
184c788feb2985e1a2e0e389c6ecdbc075b9b1ce
38389 F20110319_AAASFA chao_c_Page_077.pro
a42279f24c326ce62d6a4f5f6d972cc2
04ea4b0eabbb079ad91bd4346d66fc85f0e89672
52785 F20110319_AAASEM chao_c_Page_063.pro
b6c14ca4ea3d99ccfd6878436259d302
04ff5bf038c43cd9fc3fc1db3bf20e34603507bc
2122 F20110319_AAARZG chao_c_Page_042.txt
92413a717fdad55c6ec5d4ba10259bbb
9a3a9d1cabdb36d3ba10572f459aa915ef13bab9
44169 F20110319_AAASDY chao_c_Page_049.pro
edf4da8cc9946a3f91ef2703a536a62d
964ad1a3bffee7118be3482f08587589ede19f03
2062 F20110319_AAARYS chao_c_Page_028.txt
81355b6af44f0251aa8b623b07786f6b
7dbaf6320cd0d219fa0b62f9f4a3fe7b169bc69b
37577 F20110319_AAASFB chao_c_Page_078.pro
602de69379c3b8ebff0ed35670ff4667
86595318960ccac33e27345658a52b26e216a6ba
40706 F20110319_AAASEN chao_c_Page_064.pro
4d2c8cabb8da806cb17f8e2f346c31e1
0968bba36326d4135d930ea6e30f6a45f9b8481e
2471 F20110319_AAARZH chao_c_Page_043.txt
ba159eaad55163080f6429da660f0c55
1f474b51b729139690545ff3e24d354f6727f1dc
38761 F20110319_AAASDZ chao_c_Page_050.pro
f39b3fbd97b267ebdd5c301f03bda57c
496908009fd0d47eb5b865ecced8ef6c160b11cb
1094 F20110319_AAARYT chao_c_Page_029.txt
bcaf175a62a9628fec08d55099907d50
9b0020c28e6a596d89659471e2d028242765f57c
51487 F20110319_AAASFC chao_c_Page_079.pro
458553e0df20dfe586cbaea096eca61e
e3814d4427feda4334c98caa20a80bef0658eda1
45477 F20110319_AAASEO chao_c_Page_065.pro
f22a3179bffa526622a682bbfcff9557
580d3a0a9d8c7061b4ceb1d7cb52671d38fb3e08
1661 F20110319_AAARZI chao_c_Page_044.txt
3edacace1c637fc92cc0ac7a24c04deb
a32201685418a5af7d25d05cf922503a352f2561
1752 F20110319_AAARYU chao_c_Page_030.txt
0d1a53cb5fdae1faa968b05c8d220298
9a1ff39eff012b63f7fe88a6e971d0157f78a115
20364 F20110319_AAASEP chao_c_Page_066.pro
f48d9ac9524af346ec8f6b44885bf8ed
952cc0846fd710c6b59e3f3144f4f84c6e95bf79
1984 F20110319_AAARZJ chao_c_Page_045.txt
ffa5a1ddaebffbcdf7d9a72d26a4fa91
4be54a52b5717acd8eaaef587dbcd92496a0a4c6
2082 F20110319_AAARYV chao_c_Page_031.txt
d5029bea7e650aaaa2f8662227e16853
1ec3c4cc812b3a3c7dfb8869080ea78384e38e48
45904 F20110319_AAASFD chao_c_Page_080.pro
06824bc473cf795d8499c25171c05cfd
7cbcc186021400fa260b2dcca6233e697475f679
39421 F20110319_AAASEQ chao_c_Page_067.pro
f357fbf9c11e3ae822ddee7abfebd8b9
876d3851547fc15cf3e183825b43395176ef1ef5
1290 F20110319_AAARZK chao_c_Page_046.txt
b461e7cd08788b490a794da063f70572
de791067fbb2502f7623c82e7c97ed2b2fb39d20
F20110319_AAARYW chao_c_Page_032.txt
45c3cabcd67b5e5405a029f869815c9d
d0e36e94c63375d4f61cad45e6b65f7cd4f1a950
49760 F20110319_AAASFE chao_c_Page_081.pro
9fb2b983c7794db6f96e30d9735cfa55
accabc5e6512701ad6ac6a3dd15da58e97269968
27335 F20110319_AAASER chao_c_Page_068.pro
4ee7611a70aceecc54db3185e010dc24
05c0a69d37a41becec9a68f7540403dd9df9cdce
F20110319_AAARZL chao_c_Page_047.txt
ca40994a9710cc836793fae5564c8f4c
33ed948044f95fc1b360c911a1fd5a0b8226014d
18893 F20110319_AAASFF chao_c_Page_082.pro
58a70fa591ddf57dfdd4a3dd8bf26a52
a6fd3a5793d00a6c750e0494b641fc0437c22b11
19603 F20110319_AAASES chao_c_Page_069.pro
7f2b6064370ae5a339ce04bf8fd3f8fe
1e7af44b1126d662d9d8d1830e5fc53768b73f2c
F20110319_AAARZM chao_c_Page_048.txt
3b71a7b4c7d577bcc8d11c344ec3b348
5fa8a7b28eefc6bf8778e7f717e79e23c2414630
924 F20110319_AAARYX chao_c_Page_033.txt
2434d9538df9130f1edd22c24f459ae6
e4e28d7885fc0d28cc1148e349730e4abe4a08b4
39735 F20110319_AAASFG chao_c_Page_083.pro
00abbd35b0136888628e2a7d6877b92e
41c77b223a1eed7de05be497d25c3f1a30b25192
39025 F20110319_AAASET chao_c_Page_070.pro
e810561609d80ee4b8b6162a3d6e53a2
9b303ce00f14ad27bc6b4e3bf454dfc350744919
1754 F20110319_AAARZN chao_c_Page_049.txt
8d4528bb2e3694b5c73f9935cc94e4dd
89b6068c6771874df783356377f18703f2c72517
1400 F20110319_AAARYY chao_c_Page_034.txt
d5e144dfab72b797544954ab1ba0f657
78d219bf4b0a1063ce9b3ed448942a8e38f7ed16
35859 F20110319_AAASFH chao_c_Page_084.pro
5d33227cae29987132dd023f937f9ad6
a2b41c425f3fcda2696f4bb84e467542d046bed1
21273 F20110319_AAASEU chao_c_Page_071.pro
078059188bc4cc8cab3743f5915db76b
266d30a27f374c38db350306e39d6baedc6b26df
1553 F20110319_AAARZO chao_c_Page_050.txt
3b5a7f46960fe687d0f7553c0ad4f256
be185268226807beef53708fe312f0199f8c8fca
1884 F20110319_AAARYZ chao_c_Page_035.txt
f2e72e1a34dcacc1bea38e587f021f56
0e317a3d7994c9c91aeb3c9e482e287a431ddc54
27130 F20110319_AAASFI chao_c_Page_085.pro
654799535731b9bad92cd1dc873c4bbd
873f01c68c99a716181fa674ef6c721f9afbfee2
29400 F20110319_AAASEV chao_c_Page_072.pro
ed234d1fbafa0378f820086138ffd92a
4c99b43858d251bace46380a00250c2a026beb28
1476 F20110319_AAARZP chao_c_Page_051.txt
a10e2230f82753bcc0b076203c19918e
d10aca299464eb0d086474180fe88680c317abf5
31420 F20110319_AAASFJ chao_c_Page_086.pro
fa23cafb0aa2313359c1c71e0fc5d0b3
2d0d8926bc1b8390335bc5f68a4449b1d3464544
32719 F20110319_AAASEW chao_c_Page_073.pro
c1fbce65a41b18330d42834128931b48
2f1b893ae1e2c0be6450d9e5e71ddae83fc30b7d
2072 F20110319_AAARZQ chao_c_Page_052.txt
0cdf4f41cec40f240679c9b18b7cf59d
2de6190839771bf2bef6cb337e932fe8cdcde6f7
48338 F20110319_AAASFK chao_c_Page_087.pro
2f8701181be3cb020585e5584f151b44
89180adfac3dd69338140b3a496f781663f7437a
27446 F20110319_AAASEX chao_c_Page_074.pro
f32a8eb16b3e39b5e15efe65a3a46733
e7dec2110790f8660aed428f41a02afee0aa1931
935 F20110319_AAARZR chao_c_Page_053.txt
d2b8a265c1ec46083642bab96c1c8e7e
e1c2e07151e795c3b90e2ffcbdd36a418afbcbe6
20467 F20110319_AAASGA chao_c_Page_103.pro
e1e0805e67569173d975e4b74dc3110d
98cb07f20d95cee266772a262ccca0e16070bcb3
49758 F20110319_AAASFL chao_c_Page_088.pro
4622ab26688c4743f867cf5354d10407
95e423a2be63416c40b2e3bf8c1b5523ad881008
15390 F20110319_AAASEY chao_c_Page_075.pro
e62516752527c4fa02193cc9c5cc3dc6
db0e4297f03431e1b26945766105c5b5bfeb5419
1760 F20110319_AAARZS chao_c_Page_054.txt
5a1ada35920b69d72b6044adaf17f1c1
f21f6f296f69ed6c7449faf733189dceceb886e1
37813 F20110319_AAASGB chao_c_Page_104.pro
917e6a028aebe619dbc23ff6ce73ed3c
94731e2957d72e590b613a9da142d4bef78fbc51
34816 F20110319_AAASFM chao_c_Page_089.pro
32282deb03e1720db2a8495f289ef1c0
8d73bceded294e6d24b2e298ba6b50dffa841666
40747 F20110319_AAASEZ chao_c_Page_076.pro
a8b786ff0b66c09d1959ffc897b8f2cc
3e20e1aec7d8c305c1deb99a37a6cce02c5a7573
1814 F20110319_AAARZT chao_c_Page_055.txt
804fef5760af4c663857d4f4aa750f1d
c487cde69205612d814ca4fceaa80f4cca4f8d18
48412 F20110319_AAASGC chao_c_Page_105.pro
6574148ea504c5457305311cf6a32452
a914b34d66aec5a2bc6e06f9343bb4534e9ce9b9
41201 F20110319_AAASFN chao_c_Page_090.pro
5f7943b0b8f88682a94e1d8ff0913991
f35753560adf5316b9e22315c52bc8911bc8e81c
1664 F20110319_AAARZU chao_c_Page_056.txt
b8b3ae33c5ce50b02544cfda43347e03
da1b1d6d747183bbc84da6df82efd8e631f61ae0
20285 F20110319_AAASGD chao_c_Page_106.pro
dee1733a1c7903b8822d3ef5e79d04eb
f173cfbf5fc4ddf653ea72c03d5ff972cff1c67c
46446 F20110319_AAASFO chao_c_Page_091.pro
2b4d57d58fd8886fa239e1c9851cf5ad
3ccfeeedc38ed6231a6d81df5b63a8e43875de1e
1192 F20110319_AAARZV chao_c_Page_057.txt
a51235e5373457aa71e100d8a78d462e
d7a742abd9abbaad9bc28298b835ad03e75d9507
26851 F20110319_AAASFP chao_c_Page_092.pro
5bda7ca9d0defd5e27bc6ae09a837c56
7c21f1689023a2d1414c163f4757c258837e84df
1662 F20110319_AAARZW chao_c_Page_058.txt
b520a55fc9a40fbe8c3b0fe4a8134ad8
1fb7868412ad27df286dca52209c646adfb7aa29
16260 F20110319_AAASGE chao_c_Page_107.pro
6e38004aebb50bec46eec3b35d6d4366
37d151fce5bf9e92cb7498c1ee55ff48e256ce68
47688 F20110319_AAASFQ chao_c_Page_093.pro
bf12617532086fe95235b786068dfbfa
43957336cf7701321aff391a8d9aadad409b9fcf
1656 F20110319_AAARZX chao_c_Page_059.txt
65add9a8469423fa054de760ca23b4a7
97e98c53fe8ad12a4e5a3d97d1440fe3d0cce105
15884 F20110319_AAASGF chao_c_Page_108.pro
8f96f754a19966eb74defee2717fc7c0
16b268dfe6aa5548b8b411b43050cffe456ba8de
48328 F20110319_AAASFR chao_c_Page_094.pro
a403ad3196b1a2b6cc8ca422b1e681e9
000b733f8fe31832c7444e57cba822cf70c3b059
16150 F20110319_AAASGG chao_c_Page_109.pro
7b9cece0e8e71ff9f9a73b12e99fd291
b25dcabfdcf559e358d427a4049d3cecb753fb68
51336 F20110319_AAASFS chao_c_Page_095.pro
393dbbe5feb6273c1ac768e5c5534cdf
fc3bb42ef159740cb7175597edc86df4f9198a86
1309 F20110319_AAARZY chao_c_Page_060.txt
f3ed2846cb9e6912b17a714b00d27836
a138becd908bdf4d39bc0a72d9ef58be97c14aed
12606 F20110319_AAASGH chao_c_Page_110.pro
ac64ef668abb8a4f6edf5e0ec441d8c0
72eeaf551c1999c607d303d94763ccd7b3981245
29627 F20110319_AAASFT chao_c_Page_096.pro
e3eaffc485fa14f48b40a6ce5684beae
d59027cc7ca403c4128f08213805984c9bf1d9c5
1779 F20110319_AAARZZ chao_c_Page_061.txt
2c2e664817b4532146e18446d8d6e338
86cbc719a617a8bfc71f4587d525bc38446eec40
6343 F20110319_AAASGI chao_c_Page_111.pro
8b10b26a33c5483317864e0198b3e5e4
85874e857da602bce33ced1cfef8459d685c7a22
47116 F20110319_AAASFU chao_c_Page_097.pro
a3e508421725b04b99f64a9875ccd691
164353e446e2d61993e6930a7f60c5ab5da9ca80
51223 F20110319_AAASGJ chao_c_Page_112.pro
d67fc16b4635a52a8e8063e9389904b8
8ec1dd075f72b9c75352f634eb2cc553185f5255
27065 F20110319_AAASFV chao_c_Page_098.pro
3809ba28c25f080dacf5318f6ba15a8a
bfa1976309cff86a6b4d4c75c6310c024f896149
61165 F20110319_AAASGK chao_c_Page_113.pro
0d7085f00a8295ad4fd5a8a81da75aeb
7fdb2e0ecf8523d093316bb9b6083e5a1549a04e
25414 F20110319_AAASFW chao_c_Page_099.pro
081025e2cc7397b78f342c068a28513f
94144ad3ab8b32006d360fb9792bb06d48349d1e
55740 F20110319_AAASGL chao_c_Page_114.pro
8c5f39ccaa5b83b5dcae8e05e1be228d
2ff1623ed7b7139675a833f4c28853b676ff1524
38042 F20110319_AAASFX chao_c_Page_100.pro
52ec1b9dc93cd0fc30ce59a6fe6c3617
58ecbbabfef6e01e7bf3b70647d6b9b4989f43b5
6931 F20110319_AAASHA chao_c_Page_007.jpg
3ca27c16c22039ea038de723ed6dd274
23b66e4351ff96feea2cbfb4b595714e6856b0b5
10510 F20110319_AAASGM chao_c_Page_115.pro
341e0d5094a37d81aee1f1b79563278a
0a6d60596a68be7f000979b21c54a71cb6592f36
33846 F20110319_AAASFY chao_c_Page_101.pro
10f382df5d6c397745442f3a9bfc53c6
cd4aed480e41a732fa1e9b23f88c1a2d5d0f6e43
2162 F20110319_AAASHB chao_c_Page_007.QC.jpg
81313a5dc069b08cbdd0d201ae3743e8
8a88f06e95a07663595c0afb4d65fd367a422df7
12933 F20110319_AAASGN chao_c_Page_116.pro
781cd6c1573072add73eadef62072e36
a394ef5c73bbae58b4465c698e5422063bae778b
37346 F20110319_AAASFZ chao_c_Page_102.pro
969ac2eb4ef72fbf03ae637fc4f5795e
89e55132b4cd1a1cc8a106d9746dae1377ad2d0a
36158 F20110319_AAASHC chao_c_Page_008.jpg
ad35dde9944fa8fd8678d0bf3022cfbc
371441905f60275b2b8908274dcedbbe2f78f157
20901 F20110319_AAASGO chao_c_Page_001.jpg
0ea370c9205877217b548fb4ea305783
33376fcde966164dd4f01c0b068a1547b8a6fa47
9594 F20110319_AAASHD chao_c_Page_008.QC.jpg
6e7e6eeb71b57dbf40f2384ce3d140cf
fdf009eb6d2bf465c48347b1d74a2d69dc96d485
6479 F20110319_AAASGP chao_c_Page_001.QC.jpg
bbc2f2493622268c9e721a12b008d1b9
c20e521ae5b1c8da841472e511c072017260189c
68711 F20110319_AAASHE chao_c_Page_009.jpg
8ed2ba747cf81d8f854fcb41bdf430e5
e2746e6dbd73efbb8b808a56d9d5bf4a984eee07
4241 F20110319_AAASGQ chao_c_Page_002.jpg
92dd6e20b8f39e89fd604ba1ccbc4734
f64ec6567cdec97a33bfb493074fa19c8bcc3382
1383 F20110319_AAASGR chao_c_Page_002.QC.jpg
9864ac6048aafcfaa6f6a27a4c1705b9
6a8181591d3f3855a033c2332221414c6212d05b
18165 F20110319_AAASHF chao_c_Page_009.QC.jpg
9ac07a189aa2eca19e3bd2025ef794b5
0929792d5867d8c5a1bb9d3ae5273e465475f76d
4772 F20110319_AAASGS chao_c_Page_003.jpg
48957e54dbd37d34cf053c834ba8146b
8264df34d9d1826e9e3fa0b61a4289fa32af8af2
81879 F20110319_AAASHG chao_c_Page_010.jpg
f1c73ecd0956624a80eb8aca79579ece
638b0afaa0bb84ddfa27e44396030ceb83138552
1242 F20110319_AAASGT chao_c_Page_003.QC.jpg
2ac419412b187f707185399f6bd39ae0
9f55a6533d4cd4cdb28d49c95016d059f8fb7b93
20886 F20110319_AAASHH chao_c_Page_010.QC.jpg
f013214918db2f9d49232f239d59f24c
d963383c7b954dd9f26f2010e6469927096f7d86
49618 F20110319_AAASGU chao_c_Page_004.jpg
26af759260bed304a409b0a15255d7fe
e7e81abdade64de9fee743490761b1afbc532e33
18289 F20110319_AAASHI chao_c_Page_011.jpg
c4cb3ebc1e94321d83151e3be7093427
dd3b79c5740d319658a867abf563322ccaa4657f
15216 F20110319_AAASGV chao_c_Page_004.QC.jpg
32dfde5a9d5dc7a7162393e374895fab
7c2fd6148ce71221b0531f02b935af1530133240
5216 F20110319_AAASHJ chao_c_Page_011.QC.jpg
c151d1d1c40b8ee50af6df844c106fa9
4768310e612a601834976f3ae1693496510e7ac9
79093 F20110319_AAASGW chao_c_Page_005.jpg
36e542a305a84d7df7f5fd7ff9e12db6
a56858b37c4f89fc404dabe6e86d0dd9631beb54
64386 F20110319_AAASHK chao_c_Page_012.jpg
52aa2424b29c81bda44fbd633d9b87f4
d869b33b86ae0942673b0f2e33fc88e183091423
18051 F20110319_AAASGX chao_c_Page_005.QC.jpg
c5d6fb5fd6e65307768d2ec05aab8e11
8cecbe71c29c299a5c70edf53e11bc63f3e31f14
64011 F20110319_AAASIA chao_c_Page_020.jpg
056ee97742c1559c689a573ffa3fb85b
33644e3dcad58871aa437f28f519e42ee250702b
18824 F20110319_AAASHL chao_c_Page_012.QC.jpg
5dc9e448f2e0d79325a005dd0a48a3d8
8debb9f42fc0d3fc2b2699f616b2533c8ce12212
94815 F20110319_AAASGY chao_c_Page_006.jpg
1b1b1cbb479fcd64f4001f5102a6b8fa
9ae4baebe541250dc91c35587b613603abca92d4
20181 F20110319_AAASIB chao_c_Page_020.QC.jpg
dc1bb245cdca1e30eeddf321a799ae5b
5bf0299215ec340c9c57e60fdaa30c1ed379ca15
41109 F20110319_AAASHM chao_c_Page_013.jpg
d137025d88a7c4fbf4c1c857f6fc8744
eb773ba225f5dcc19f605b7472d1d837f45c6972
21301 F20110319_AAASGZ chao_c_Page_006.QC.jpg
6494af6ae3ab1e3a1d916d8762e42a76
cdd416ff4ee8bade81defbf1b8d12e3057af9ae2
68064 F20110319_AAASIC chao_c_Page_021.jpg
6ed482bd4028f5c7acbe1539c1bf8fd2
a8208f575cb3597668c66092a7fee5d5da48cd56
13380 F20110319_AAASHN chao_c_Page_013.QC.jpg
27f27999475a977a9614210f67ebefc9
cee6aa335e1bcf4609fabeedd29646572e99f0bb
21952 F20110319_AAASID chao_c_Page_021.QC.jpg
d9b442d55e10720ced86073bc0266c4a
e94b50efd9d7a4514755a1504c57547c200d4930
73720 F20110319_AAASHO chao_c_Page_014.jpg
f83af43684631326fb9503e2e4277e0d
dcfb0c913f0261fb41ec86013ebbfdfd7bb6e69c
54652 F20110319_AAASIE chao_c_Page_022.jpg
96d120f7ab05a330c73b05846ea813cd
859f4996ce84265317198187f27e4cace54d3208
22643 F20110319_AAASHP chao_c_Page_014.QC.jpg
4647c3548417bda9d9bc937a0c78c746
4cb29450951f785cd8b9f9e8d86f257c0eec2c34
17857 F20110319_AAASIF chao_c_Page_022.QC.jpg
34ddd164215a1461f5992ffa4f64a6ee
b07f675c4bced059ebe846228951ce5e6e0af08f
72921 F20110319_AAASHQ chao_c_Page_015.jpg
6a728f4d991535b76b50f43ece804df0
319a8c3933bb7f1bb7f7a7f85f0ba9f7c3b5aa8d
23092 F20110319_AAASHR chao_c_Page_015.QC.jpg
56393870e7cdf5788671ca561a72f2fd
7d537d865d76183d92b5e52477e9dd80432c20a6
49927 F20110319_AAASIG chao_c_Page_023.jpg
85a2f052d3b5501d09bd4dffdf269dfe
f1fd919750a2aef79f00561f06d15d5b3304322a
76476 F20110319_AAASHS chao_c_Page_016.jpg
9ace4a7a1acd77ed02859a71208c0001
2b62ea5d72e318832045060505f37cd3ac5bf8d3
15977 F20110319_AAASIH chao_c_Page_023.QC.jpg
330ac339189925f9900274541ecb7042
e06840bd75f535580e8510cbd117b7bf0bcfca56
24435 F20110319_AAASHT chao_c_Page_016.QC.jpg
344559d5c0eaeb956dbf73442ea2f017
6c670ef0ba33effe5020253f9983af0f5a82dac0
55920 F20110319_AAASII chao_c_Page_024.jpg
ad89c38b49c6a54b554839b475c9ce63
dea80727e9b257ce289affb2571a57deca8ab78d
72063 F20110319_AAASHU chao_c_Page_017.jpg
9fe715338c12576801462a26509ecfdf
9b3af8424df22e9cfd3e45ab59723b31e9f35b67
18118 F20110319_AAASIJ chao_c_Page_024.QC.jpg
063f0a21b95246d58c27b0a5e330fe98
c3cddb943d0412fc847b17753d3f2cf1ad59555f
22737 F20110319_AAASHV chao_c_Page_017.QC.jpg
e21753905a12f7b7a12ccbc573e67d98
394913e80b7ea6988135d922542b45e2b836acbd
68765 F20110319_AAASIK chao_c_Page_025.jpg
d07e74f482e8fea749b5a6469c8b35e1
04b4f92aa2205aae745fe2b6078e3fa919645177
76316 F20110319_AAASHW chao_c_Page_018.jpg
afb4ef0b422cc456a464e35a449aa7e3
e9ae2f2a90327514f1e29a8909b8c825006a4c6e
48809 F20110319_AAASJA chao_c_Page_033.jpg
f25e50589b11c9d7dea58c316d953cd1
0339e5e8c2d8485945fbfbc2825538be03fa1f79
21916 F20110319_AAASIL chao_c_Page_025.QC.jpg
d2da0493f2223e4c303de3c41c02c793
30cbd6ce0ba5e70d9159f2e66c189b31f5f7ef71
24388 F20110319_AAASHX chao_c_Page_018.QC.jpg
780490b5bb3030c58fbebcbaf7f01c63
83194ac7e19c5d655f6035bd855f63879af5946d
15531 F20110319_AAASJB chao_c_Page_033.QC.jpg
3e258bc0d3d9b811814b49d386089b82
e809514eed860e322b3a627a299ef86d813044f7
75558 F20110319_AAASIM chao_c_Page_026.jpg
1a6a2ddac438e7e5263fa73b9c92987a
7718485dfc2388258d0b332af74b3289f14e2200
64458 F20110319_AAASHY chao_c_Page_019.jpg
2d99dd9bd9cd043d99a1a687a3a942d4
c7914cfbfa64ae4a678c6ab7642fbb9d31486407
61242 F20110319_AAASJC chao_c_Page_034.jpg
c936cfa7f3593a215106657edfd0accb
5e385bebdf5f2aec3f8e08f71bd1b9e8edc961ef
24449 F20110319_AAASIN chao_c_Page_026.QC.jpg
782cd641d0d982e85a43321f81888744
b6e620c6d6685cb82f69ad1ba8d93a4ea4d08034
20735 F20110319_AAASHZ chao_c_Page_019.QC.jpg
8a62474a0e78687167377072fd5b7e37
b4e07342cfa1ab950cfaaa3f494afe7275bcba48
19014 F20110319_AAASJD chao_c_Page_034.QC.jpg
eb0ba3b451d956ffe678433a2ae0d733
4a080d874a0458396c695018c5e7a0e4be5595d1
82183 F20110319_AAASIO chao_c_Page_027.jpg
8b988bb6018cf1d674d26a5d0954efd5
7255b9384ba1c002f86e2b86aab1bc7fe4437f5b
67499 F20110319_AAASJE chao_c_Page_035.jpg
15ea6a041da182cdf9a4e4c04541d265
33807d6a8c1a762063c9ab7641ab8b707e5540b0
25229 F20110319_AAASIP chao_c_Page_027.QC.jpg
7f564e9fcc132a622d2f9c300c3c00f6
402a239caf640629589fa819f54c753f649d9358
21160 F20110319_AAASJF chao_c_Page_035.QC.jpg
fa431429ad3fbc684bf8e353572af88a
00664656198232699147beb8eb1a4d6f06cd3372
87442 F20110319_AAASIQ chao_c_Page_028.jpg
44289ada1b2331cf3170fea3f341d5c3
8346f603174fb39a65303813d1019a94125c277b
55659 F20110319_AAASJG chao_c_Page_036.jpg
553897d1399ad40c3f6f9b460b26115a
4613eb09a9748ddab79859729fa315d90cea1df3
27617 F20110319_AAASIR chao_c_Page_028.QC.jpg
54f41a9c85d0c2afb17e75b7e3d7da16
1083bf74c5e0c72313b6a3e1da3ebed61edfb7e8
69975 F20110319_AAASIS chao_c_Page_029.jpg
549a4f5dcad690920535f5897b20c236
7d13c5a1c52c7c52652664b4a76552267a37854b
17194 F20110319_AAASJH chao_c_Page_036.QC.jpg
0327c1e4619a1e645f74159fb1c52642
170927842b0ec81fcae0efcf4df14ac1d13385eb
22134 F20110319_AAASIT chao_c_Page_029.QC.jpg
e0a6810bdeb4280c728f01ef384ce3fa
5df46976315052d2ab8ab81218498deb7b6dc0be
69275 F20110319_AAASJI chao_c_Page_037.jpg
99832fc7b88d6b5fd44e025945873fb7
e457570306d4d5296b03b9d69cfd124868eec92d
73740 F20110319_AAASIU chao_c_Page_030.jpg
0c0aad5c3ddaa7f5f37fcd2c833153a4
8ae34a9e3b6bd99b54e9869b0cc7c13fadc007ef
22101 F20110319_AAASJJ chao_c_Page_037.QC.jpg
d83762ae158a6883c809c36f73e710f7
b7e991235ec3693cfe5268841d86b25724e16ad1
22231 F20110319_AAASIV chao_c_Page_030.QC.jpg
43d5ca137bfa461746e14f3d7b9b7b8a
0244425cae26f83779153d183f5cae5c48c8df54
82338 F20110319_AAASJK chao_c_Page_038.jpg
4daa75a18b6d4436ea78c29567b75b30
def9938590d15520819b14d9e0613aeff9169d11
80545 F20110319_AAASIW chao_c_Page_031.jpg
3c9ebdc422857b4febc42e14e80c9c63
06fc0594d3ef9ad798d59e40f427e49140f90aaa
24869 F20110319_AAASJL chao_c_Page_038.QC.jpg
e914024e79dee1244231cb455721d33f
7446f4049cb989ae40489733e4dca3402baefe79
24978 F20110319_AAASIX chao_c_Page_031.QC.jpg
5132c54ab4b167046472fd51993d8660
4553144f0f12c152e4068165f2d91475cc5059d3
51836 F20110319_AAASKA chao_c_Page_046.jpg
24e565f47c6d8d130cdd6e95ef4395c7
717621e7d1bd8c876467c6b86aedba0f9fbc0adc
11204 F20110319_AAASJM chao_c_Page_039.jpg
db7a6bd9c7684129a6c46e5d9ffc0285
f3c210808a81992c20da265c417e9dbda1e7d87d
58575 F20110319_AAASIY chao_c_Page_032.jpg
ae45efcd7633a27b2a959f6854974db0
f5b2b59b925bd37f9485a0c46e69a7c7d03fa9a3
17204 F20110319_AAASKB chao_c_Page_046.QC.jpg
20dd3dbd2ab3f20a556185469b4606ab
ba0756fcf5a4661b16cd5bb946339ce55711283f
4076 F20110319_AAASJN chao_c_Page_039.QC.jpg
f45be1f326d89a72a9b71974b5147612
0ac9d3b08c41baa1e858d3b77a7fe430901b6fcc
18506 F20110319_AAASIZ chao_c_Page_032.QC.jpg
d361c0e689b23641a881d78384575e29
6382fd94b3aa74156b9522c486f1ca84ee67891a
80370 F20110319_AAASKC chao_c_Page_047.jpg
a8e558b5969e59d181d790ba3e9fc923
f94ba46e30dfc94ca17907ea0706031284424d5d
74580 F20110319_AAASJO chao_c_Page_040.jpg
323a092f6ddcd828a224c5fc80facbcd
6a96d85d624865f40b1ba7b25b5fc62a438636d4
25793 F20110319_AAASKD chao_c_Page_047.QC.jpg
9579e00535bace6406f88cd471c1daf3
fd401ef8fcbefedbbccbfef33882528199207a04
22799 F20110319_AAASJP chao_c_Page_040.QC.jpg
9b184898d26ec650bda206371367e705
1eebaa70df518bc64d7d8e81d395e77997912159
66026 F20110319_AAASKE chao_c_Page_048.jpg
fd0e644349207249ca02c72cfd06c883
a1928c8fbf97c4aa40e0c5b770c85bcc15a149b1
80142 F20110319_AAASJQ chao_c_Page_041.jpg
3d8359ff5c7a3987a68b012934650016
9ac13ba96f2bc3af611937ed1ee153a15a24e08e
22360 F20110319_AAASKF chao_c_Page_048.QC.jpg
fff319867970606d3c2795e33577dd29
24a2ecb4b5c967d078a86c56232ad2f0f9779b8a
25303 F20110319_AAASJR chao_c_Page_041.QC.jpg
582ed361aaafc61a01dee23bb9c43cfa
0d085fbe543d027ad5f49e607e8ce274730df24d
73687 F20110319_AAASKG chao_c_Page_049.jpg
c859d5b9d074d3dedf7951ae7265106f
cf9ea742acc4b904389fd7fd857b6c62717171cc
77744 F20110319_AAASJS chao_c_Page_042.jpg
b1aebdadc65f24d1ff31b6591ceb1b02
b4c8aef54b9fcb07d314e434279c035ddf34a1fc
F20110319_AAASKH chao_c_Page_049.QC.jpg
17141ed1dfa0c376718c1543253b4374
0a095e1e973e00910943012c40090efb9c8b12f2
24512 F20110319_AAASJT chao_c_Page_042.QC.jpg
5bd4801768276def13d54e3dbd09474a
2fc4dd13d6477653a15483f06af8a9151a46e542
80880 F20110319_AAASJU chao_c_Page_043.jpg
fb256aee092a4697e1ab2746538daa54
4517efbe14712fae46fc4b3ed70cfa9f6781a368
66001 F20110319_AAASKI chao_c_Page_050.jpg
25ec3d84ddae0c3e06924d90aeb45546
9c735c061e7610a8d539d89a4de1945119966d48
25703 F20110319_AAASJV chao_c_Page_043.QC.jpg
8e96228390c10a4d77c439dc398faf11
028d3236729e7ab9dea676762d57a3d166359683
20589 F20110319_AAASKJ chao_c_Page_050.QC.jpg
2c64acbd561783061ae0e37e6a5cd15c
fef2e65d632ffe99acc7aa4d15c31f82bcc57813
75379 F20110319_AAASJW chao_c_Page_044.jpg
5d6dd56f08c51c2f9df3f085be45b2b3
dd71acf6136090b57aec943b245fb9329aea4cae
66378 F20110319_AAASKK chao_c_Page_051.jpg
7d4804cbd333ece2f2ecadf15a9e1880
dabc113b1ec1903efd8415bf1d730c9c1db478c1
24269 F20110319_AAASJX chao_c_Page_044.QC.jpg
a415f4032ad9b443b40545189fafa462
0b0e136b6bfea5486b23df423b1e9cb1b58b052f
70923 F20110319_AAASLA chao_c_Page_059.jpg
fc1dbefdbefb1961a746df8c63ab6628
5483f51c1c246f82eb8ced0f9228769c98af17db
20685 F20110319_AAASKL chao_c_Page_051.QC.jpg
a30b6e5b55a0970f7e9ed08ac80b9272
73c09d1481cebadf2639e90e364605d8b028e42c
22444 F20110319_AAASLB chao_c_Page_059.QC.jpg
78605278ec4aa033ab4dccf4dbad271e
92dadfba339fd56033081a0d9874f373fa6cff5d
86331 F20110319_AAASKM chao_c_Page_052.jpg
c16f33dae08a60308308920f1d13793b
b733918c98c83cc60c9463220613ede12d401c0e
83851 F20110319_AAASJY chao_c_Page_045.jpg
1c71a7a0a9211cd56cd2e104356fe731
754032555a721b7510e4775222770a21a8aa970d
58948 F20110319_AAASLC chao_c_Page_060.jpg
2715c179755e112d4b901ebefd6a6b22
53711b35a7f2b65d0d08b1fd6b905dc74e14eb88
26838 F20110319_AAASKN chao_c_Page_052.QC.jpg
142605ce1d26980788918a610657c996
5acf7214f6728d942d6d69c7da157d1660933859
26055 F20110319_AAASJZ chao_c_Page_045.QC.jpg
a8cb9badbfc2498afbe9954351539e29
ed1831167f50ded8baaa271fac15f45939163971
18413 F20110319_AAASLD chao_c_Page_060.QC.jpg
3d9c425f44c29ad89f7bd414bdfc4e50
455dadff506032a53430f1dc9c8f5606fbb2a874
61104 F20110319_AAASKO chao_c_Page_053.jpg
70f3b2dde75880c95dd9094734d15ad6
ee2e3df580610f05c5788588c383e9b2174e42c7
75234 F20110319_AAASLE chao_c_Page_061.jpg
bb87c14dace1a4e52b2e8c6d87e2a1d4
0597f28f5794fc4f9702c184115cc6f21f968c89
19243 F20110319_AAASKP chao_c_Page_053.QC.jpg
2b6f7b3ac41a72bd7434894b1df8d3aa
df52878e6f525ba9b6b8d261614cb27742cdf3b0
23154 F20110319_AAASLF chao_c_Page_061.QC.jpg
f08c274a760cfb6728dad26e3b277359
c14f85b793e9e113a1635e16c4bbb3668b513a28
75083 F20110319_AAASKQ chao_c_Page_054.jpg
5cfe45af3c811995504f9e831bfcf075
bc219b4ad754fe36034bc1079b2b82eeb2f8f897
82420 F20110319_AAASLG chao_c_Page_062.jpg
09e9cb3f17b09e3851927796ff89a390
546c73463d1a0c4b8f3d567d244c46abe86c7c6f
23089 F20110319_AAASKR chao_c_Page_054.QC.jpg
d7539cf9d96b150d1785c48f8fb29d23
5a8065376c64e88201f29e352f0ba5f12f576bac
87542 F20110319_AAASLH chao_c_Page_063.jpg
cd60ac7fa831389be2834b7e47321b6c
7b0d4920caf7616852a13f4af14ebe25b2156e02
75764 F20110319_AAASKS chao_c_Page_055.jpg
f2227074abb6bac51fcae8aedc9e0b86
f28bd5cd617b8a3f6d69beaf6fa60defa44dfc1f
27761 F20110319_AAASLI chao_c_Page_063.QC.jpg
438093a7c700efd478f4550532b21f7e
1dc11275b527044dffffe511e506fa2edb30ad77
23498 F20110319_AAASKT chao_c_Page_055.QC.jpg
102c6411abaa2dfc9e18f686b28ccefc
1491d9586722b861155279e651adcf43516cf996
69685 F20110319_AAASKU chao_c_Page_056.jpg
b639f01fe8fc1d967561e7c8af944670
9058b22946681bcee518ab3fee84ef3983e95095
67341 F20110319_AAASLJ chao_c_Page_064.jpg
18976a51e3bc32748d3ef57aa215f881
cfff4beefc155441a15ab1b260d45cad64c88ee8
21406 F20110319_AAASKV chao_c_Page_056.QC.jpg
1132a096db02f201bba68d2acd5551e4
c5d296c90170d9851c24329f46dc9333ece0b6f2
21690 F20110319_AAASLK chao_c_Page_064.QC.jpg
a96dac437aa65d212af8161cae3e319d
bababb6f2500527df54edb98755fa71b40da4aae
36644 F20110319_AAASKW chao_c_Page_057.jpg
893ae7bf020039f49f2a75af16566745
20fdc282afe7a859778b89050e94f3eb34683cfa
77297 F20110319_AAASLL chao_c_Page_065.jpg
d2014105e624983642a6aad17e95713e
141a40c821104ba94e16f882c32e7f5e5827e1ad
13138 F20110319_AAASKX chao_c_Page_057.QC.jpg
3fd235d1f967f6ec26832fc7925f974a
a7fb1ca3389fc8ec3567e32c0792a4650d5a576a
21864 F20110319_AAASMA chao_c_Page_072.QC.jpg
ec622ab21718dd8c2126b3dc0c033aa2
22ce82275bde9924283d51e3281e13b91d03e72e
23421 F20110319_AAASLM chao_c_Page_065.QC.jpg
34008af4cb7a95568953d50bcd2270fd
05cfc52016219104aa569d94fdbd82bdb8197aa7
64756 F20110319_AAASKY chao_c_Page_058.jpg
f09c5126fd016646cd898a5d0a3a015e
9922e454f4aebd3841305d197b2094b64315d76f
63765 F20110319_AAASMB chao_c_Page_073.jpg
6daf7a2bb2e44f0d50a40cbeb537c14b
6e879e0b7512e7390d910f3eae7b01edf2eaacc9
65160 F20110319_AAASLN chao_c_Page_066.jpg
d7639ea90cc3b72a4a9284be2dc5b214
ae72f371132a183e8306d41bb817621f2b789769
21362 F20110319_AAASKZ chao_c_Page_058.QC.jpg
2abb2eb26bced32aae14269db134c63b
ee212d485fda256c542e88b4da8c4a0cf256504e
21960 F20110319_AAASMC chao_c_Page_073.QC.jpg
131fc20557ee1902d17b4a1fe91714bf
62535e696c1838db5296dcc47ccd77c1813083d8
20716 F20110319_AAASLO chao_c_Page_066.QC.jpg
456a2da5384534c8b7e8470f3666ba01
d71005ef0611ee4dff8fe5c7e92b90a41a5eda20
67628 F20110319_AAASMD chao_c_Page_074.jpg
38447ab7ca421e5bdd269cf6b1fe3739
91d1d3ba400612d1e6d4a7fbd8a18778e187bcde
66717 F20110319_AAASLP chao_c_Page_067.jpg
6302aa133fe7486c8ef256f771edf7e5
087bdc554f52b5d281b4763ebbf99bce98876484
22207 F20110319_AAASME chao_c_Page_074.QC.jpg
765221c4a137d02ecc58fbb498a3679e
bf88726d7d11a9914961498e91fb4f8827f76608
19654 F20110319_AAASLQ chao_c_Page_067.QC.jpg
44add0cbd12f108ef0d389017de17b31
42001d2168f27ab5394643b2d74f6da9a9198dae
42640 F20110319_AAASMF chao_c_Page_075.jpg
52011c69eb1b8a7ad5afac27d9676fd4
502bfe734c6a946fd82ed7fe4b85e11a10a41e59
70654 F20110319_AAASLR chao_c_Page_068.jpg
c62d291129898914080b4020d98b01bf
f2c6c85e548f2e1fd1e83644b24c2390f04c1ee6
14069 F20110319_AAASMG chao_c_Page_075.QC.jpg
a0142f5fd77d1fe25890fe69cf65b3dc
0063f979a666da3a915c5ac75da668adc5170e20
22825 F20110319_AAASLS chao_c_Page_068.QC.jpg
d3f939377ab0b5517310569e6313382a
36fb326c2a3e8e229cfe9d75e44f96b5d73ed2cf
68827 F20110319_AAASMH chao_c_Page_076.jpg
b01c1005d03eef2dc6804c6621f955f3
d30d5024868b48a54adfe1de2d45c2e5a09944b7
60813 F20110319_AAASLT chao_c_Page_069.jpg
562cf8ddfac3ae5ee2e797e972962fe8
f068e833ff416ae547e235e10c35258d5fa7e93e
21063 F20110319_AAASMI chao_c_Page_076.QC.jpg
5004f61fb4caebe13ac348fd2d1202b2
eac24956e894482672f0e3e8a22119f191aa181c
20264 F20110319_AAASLU chao_c_Page_069.QC.jpg
b864c098f8dd16b24ec46a66b383aeee
6576ac222197ba184a6e13455da98d0c0201a281
64729 F20110319_AAASMJ chao_c_Page_077.jpg
54d2b41a23246b5436c7254736be95b2
c562263108368c8550478cd565b706b6612d04bc
76426 F20110319_AAASLV chao_c_Page_070.jpg
873507cec5f711fa376f506cc4d1a720
14e65f83fb13074d517cd60de55192f8e04d5156
22722 F20110319_AAASLW chao_c_Page_070.QC.jpg
c12bf2c403827577fc1bc4566c0f65ac
aedf658d522fbbd0115d19b16f0c3d04b5ed413b
20828 F20110319_AAASMK chao_c_Page_077.QC.jpg
3989a679cd22c60e78d7fe5a7c5fb128
f6567f4891bdcd3dae895c6dc0146c1fde844783
71410 F20110319_AAASLX chao_c_Page_071.jpg
e5e8a3a326dcb0ba7a1235792ce8823f
595245cb4dbbf50d5be7d7ecee608bae4be48362
17368 F20110319_AAASNA chao_c_Page_085.QC.jpg
fdcf67b6cf49f627b14f2c0e1b5cda9d
8068a3078a5188d3cd94930fb965877770d1f3e4
68211 F20110319_AAASML chao_c_Page_078.jpg
b29aa0567e0ed431097ed3302bae0180
7ab4ed645dc6fafed02fe8d7bd2ae7174f938463
21068 F20110319_AAASLY chao_c_Page_071.QC.jpg
770c1853d4637999b8be9872822aa6c4
7f61239e10cc0b44af44775eb9477148dbc0f24e
66951 F20110319_AAASNB chao_c_Page_086.jpg
1111422c2179012af47bb0149a4ee920
2aaf8ce47bc871f450b7a0c7641f71173967043d
21876 F20110319_AAASMM chao_c_Page_078.QC.jpg
41b38c69033b8da025c5a6ef97406202
b15d85117d00eb22f5149891fef0401fad962c15
72639 F20110319_AAASLZ chao_c_Page_072.jpg
385f322d2b1bd7ed458f6a0f3778fbc2
c29824ccf978529de125fe5b39f1ad1d95f5578b
21422 F20110319_AAASNC chao_c_Page_086.QC.jpg
cbd65c544f375a6f8034d40662f2411e
1b785ff98d33de8ecf6e44750d8e5ba66d448a86
85553 F20110319_AAASMN chao_c_Page_079.jpg
0de1c0c752ec783ef02592ea9d7d9227
a58511b2683a19936ffc71e7a6769a0320faf26d
82342 F20110319_AAASND chao_c_Page_087.jpg
067701302e7e71af43ac70a73a81cc79
10b2e1fdf4b75eee5072b085dc99d6bb1b7ae859
26894 F20110319_AAASMO chao_c_Page_079.QC.jpg
ec8ae27680ac2167325fdfb2c8d48145
095da8367a3019a36c54f42a4fb94fc8595a6fad
25338 F20110319_AAASNE chao_c_Page_087.QC.jpg
8b445537b568c6502fa34e2480a91118
e2bce7ab78ba789a684bc892ec2e2988baeb471e
77170 F20110319_AAASMP chao_c_Page_080.jpg
6ba48f32606359004782989d24b81d16
65c23fbf26781a6aaa026da7eb867709d9ac14ba
83826 F20110319_AAASNF chao_c_Page_088.jpg
7fd9d6042c512771aa35eb12f6c9937a
47909b7f100f1aa3d37a95fef4238a3cc69257d0
23911 F20110319_AAASMQ chao_c_Page_080.QC.jpg
5d002d26bc98c337dcaacd441bf54bb6
fb687b3f64805fcc44dc182c3d5ba64511447f7c
25426 F20110319_AAASNG chao_c_Page_088.QC.jpg
7f9d4bd07d6efbab4b25d7a87f99a177
12ac99bb20aafe491efb5ce0a3412487aa4c74a7
82580 F20110319_AAASMR chao_c_Page_081.jpg
24aa952109eb7d20be117deab73178d8
f42a1da7889fb958608f2476ad8b7c3776097e86
58789 F20110319_AAASNH chao_c_Page_089.jpg
b74d92aa2281844dd0d33251a7f2aef9
a10992d0fb103bad90f2c0627d3c014608c892a9
25531 F20110319_AAASMS chao_c_Page_081.QC.jpg
4ba1cc791c58f85c255c9e58b0650807
c7029c7b1a764b375ae31bb24759d5d369d80a52
19303 F20110319_AAASNI chao_c_Page_089.QC.jpg
fb2d58b1a0381896ae7ba09b93e7b75c
0218253deabc115895f6e6ec1aeabb0f526abfd7
64924 F20110319_AAASMT chao_c_Page_082.jpg
f3bd7ccd17111913ba654ec283f2e2a2
def71f04d40d17032e15c1071b3385de56d3c40e
63097 F20110319_AAASNJ chao_c_Page_090.jpg
4e3c4d0f2f6e89e18133cf12dbba63df
c5a3ce5f9b175a7a1974507f26b1b47467c41be1
21360 F20110319_AAASMU chao_c_Page_082.QC.jpg
b5f1f603ca2daa2c0b2e157e67c0e194
ce41c9fd1a228907464e84752af32f536b99144f
20591 F20110319_AAASNK chao_c_Page_090.QC.jpg
3ce4e82f9c2d6cd3a829ca5c1b16a62e
1ea665b69603570c8739203fbaec02f845a8cdac
67174 F20110319_AAASMV chao_c_Page_083.jpg
d1cb069619d5241871f93f34f78a599b
1a51cbb3d699f8221879830573ed66b66bc45c38
21534 F20110319_AAASMW chao_c_Page_083.QC.jpg
9ba45faaa56bfaf59d1a9fbc0ade4c76
8c0d377193b5b34fd397b283ecf6dc1be91bb532
20989 F20110319_AAASOA chao_c_Page_098.QC.jpg
71c41102a048ed4b688dac3f20180be4
9509a54519ac901505b148dbcb14efbe6183aa6e
80183 F20110319_AAASNL chao_c_Page_091.jpg
ddb39025a436c481a9ef5122f7c0d55b
0bf93a9b462dddec620d548eaa01c3c90ad19123
60862 F20110319_AAASMX chao_c_Page_084.jpg
d0afd2692199fe08842a89121f81cca6
897430239de308a4836a2b368d16322dbf9eda33
53684 F20110319_AAASOB chao_c_Page_099.jpg
c8aa0b1158b4dd771ecfa437d887bab7
135a4a30fe9973687ae32892869edbc6ed23e463
25120 F20110319_AAASNM chao_c_Page_091.QC.jpg
0c041a75f835f0f43508524b798df7d4
51062fb818d97ca4907f6d7f0745b60e0c9fe3a0
19541 F20110319_AAASMY chao_c_Page_084.QC.jpg
f5da14c9bdc727e91ee2dfa0c725e8a9
61b9aa2bee628dce349a6ae5b9b408638f344a02
17362 F20110319_AAASOC chao_c_Page_099.QC.jpg
53e3283e53eb4a42e17dea3089b46f96
1cc294831e31dbb1903bbc513e4e869e37dcfd2a
42567 F20110319_AAASNN chao_c_Page_092.jpg
61d6e09cbb48f5f23a8597e489553357
1e2c172e21f3cea262f20ad30bd91339d10c7166
55134 F20110319_AAASMZ chao_c_Page_085.jpg
dbb2a8830444fc5f0c4a1dab33f261cf
0cb38d07f98b08e4dd92d5cf4f7fd988766ca970
69993 F20110319_AAASOD chao_c_Page_100.jpg
fac302b555e5ed989140d3c99fa74ffa
e211f8fdbdfa419a5622fbba1288195d105e33b0
12840 F20110319_AAASNO chao_c_Page_092.QC.jpg
bd2472371cb5f659ce84232e1d4fa9f1
3e472da49c7e8f2a938e4eeffa2f08abd6ec857a
F20110319_AAASOE chao_c_Page_100.QC.jpg
e19f2db8dbfd226b94f2750cc37fd65a
3aa83c4ceaaf32d95499be46b4a2814972e112f0
79144 F20110319_AAASNP chao_c_Page_093.jpg
4dfa705cd3aa5e7d43fd6be0622a8725
f63e2220d531fb784d0d665e68e7949ce9e3331d
75455 F20110319_AAASOF chao_c_Page_101.jpg
6772073a7c580aa89ed3dec075994044
ccd060a010c228e0f3c26bfb11d754efaea3b15c
24707 F20110319_AAASNQ chao_c_Page_093.QC.jpg
e860f8612cbab18ba9fb6251781c4933
b481a9481f58d6e6b7c10e29b6c4c728e5da42d1
23523 F20110319_AAASOG chao_c_Page_101.QC.jpg
22414cc0563e3499a813d88c4fcd7b72
92f156a24d2bc4e8dceba1d7bfb7831113e0b522
79624 F20110319_AAASNR chao_c_Page_094.jpg
74917864ec67d0a3160df5c8306ebf26
c13b7bf575f50c12d8bb882f1c90e37e626b34e4
73263 F20110319_AAASOH chao_c_Page_102.jpg
4d22bc532ef9f1ecab5a5c84d14c1bb2
c103c8e1b53d86c9d11ba9eee7f5d76cff7d52ea
25156 F20110319_AAASNS chao_c_Page_094.QC.jpg
4cb8db7359b21f6bce2e3d0015a8ec7b
d0b8fb50ce099ed5fe306b5a7338aafb433b2d17
22145 F20110319_AAASOI chao_c_Page_102.QC.jpg
95925f5202aabd974d4ae72eda633293
49083e3397ef870cfa766a033f3a38ac2c2ae38b
83912 F20110319_AAASNT chao_c_Page_095.jpg
64fb4cd70e251904ee6ad270db119bc6
829b698e1fa4bf105bd82785e59a71903c420253
36321 F20110319_AAASOJ chao_c_Page_103.jpg
fe6e9c80d1b750995bf5315328ec693e
58f8e02690c6351b8348cc59e282fccf326badc9
26467 F20110319_AAASNU chao_c_Page_095.QC.jpg
4fe10a0b7e0b02435fa30f948dd051bb
7d32f0ce5183efc1174fcb6fec47c5be55d03ec6
11297 F20110319_AAASOK chao_c_Page_103.QC.jpg
de6c1e0f84315ae75a40985d765a4899
fcb74e7a993fa34900a2ef2591e3ad6bc87a39a9
62051 F20110319_AAASNV chao_c_Page_096.jpg
c8b2b6558283a108cd67adfc0ebd3929
0f07ad9b057cff72d903ca86bb61b441417fcda1
64699 F20110319_AAASOL chao_c_Page_104.jpg
ecb72be77ce69ff3fc3a67af3e4ec991
28126c0bf0ef9b1ec3ee611cf81ea819d68b04ce
20705 F20110319_AAASNW chao_c_Page_096.QC.jpg
4665805206df076c796cf1a4b123e388
acf82ab18ec1f300c22e1e1dcbdb135b8c4e776c
83450 F20110319_AAASNX chao_c_Page_097.jpg
482bcce89689bd6754e50f490f700af6
7459509b5bc54ed06a3c147a61b546426931eefd
5245 F20110319_AAASPA chao_c_Page_111.QC.jpg
2c5bc026f831c123aef374e181e140b7
44589abc645cad882886ce3b60b76266fb9f86b4
19931 F20110319_AAASOM chao_c_Page_104.QC.jpg
66f1ee2770c81daae7bd07a09684be87
64a03ae3ebd4cf89d234bd9076491403b6458a32
25762 F20110319_AAASNY chao_c_Page_097.QC.jpg
10f442b4e8d94c58ad91e080998fa16e
509a79f28fd9838909412407f991d9cde449a835
92497 F20110319_AAASPB chao_c_Page_112.jpg
6e7adc0922cc664e89cf1286d62770f3
5053e34c0f6ada9b9fb8c85eb8c19cca4506fa7d
81891 F20110319_AAASON chao_c_Page_105.jpg
f6b1450a84dd9012c498ae7a2747a2da
4c14e2391d029796c8c00bf5fd5089a41404ba0d
66462 F20110319_AAASNZ chao_c_Page_098.jpg
d534564f9b1778de28f28deaf9d852c1
20309edb28b33d65a145d851c4a2cfae062fbf0d
25387 F20110319_AAASPC chao_c_Page_112.QC.jpg
e8cfe7721183f9c67301bcdf347ba33a
91c12a6dba5b7a4a5005919e801082d4bb6a8d8e
25694 F20110319_AAASOO chao_c_Page_105.QC.jpg
36f002866629cd0c4a2544a0c1c260b0
5d0cd4e794d9309efe22b99fd18935fb67e4fa68
103947 F20110319_AAASPD chao_c_Page_113.jpg
5d671855abf7ff04478cce482d3c8cbb
af5dc19b51dd62b8161167f2eedfb9f52a6b320c
35721 F20110319_AAASOP chao_c_Page_106.jpg
dff8fd065b5ce0ad81be38a24ae6cb1b
74ceccac3f81328ee299d1f44e8f0bb2f1a4ae98
28651 F20110319_AAASPE chao_c_Page_113.QC.jpg
6c341cce724eba319db81692ff7d84b1
025ce73a754365f7e996711e5accb89370a1e021
11085 F20110319_AAASOQ chao_c_Page_106.QC.jpg
4f675bb7912d12ce4f56d55e629552a7
2733d26b7fc1fa2d4fb252e8343f3e847651e9e6
102670 F20110319_AAASPF chao_c_Page_114.jpg
5abd6117b01dd8126e5e48fab98312e0
144d8ff231f5df16becd746a9638b6d9366c6444
33849 F20110319_AAASOR chao_c_Page_107.jpg
d4c03137f60a7d43e6c6f95e3e86f4c4
8afdbcf009d6274b254656ef320610a13aaae563
28578 F20110319_AAASPG chao_c_Page_114.QC.jpg
54b5a2075d01b2019330cbbbd1e529aa
268b37dac529085f583c4a4c15c14d63cfb695be
11277 F20110319_AAASOS chao_c_Page_107.QC.jpg
8c7fd59ef75a9bbb8065820a07537870
2283ea0cd7756ef587b44c8313c2db49bc185dc2
23826 F20110319_AAASPH chao_c_Page_115.jpg
1bae79ee67f1013f914c337fa039692d
a9ecf98bbee0f84fde1038a6c6787cc6ccdbc146
31101 F20110319_AAASOT chao_c_Page_108.jpg
4b916c8442673ddaf23599afc9335a02
3e1032129d7f948ac81142e368c1806f9a30c16f
6375 F20110319_AAASPI chao_c_Page_115.QC.jpg
c8d74b5ee01a692da6e92e51e5777194
4eca79dd0f3f744e7fd5ffb6bf0b853391094217
10663 F20110319_AAASOU chao_c_Page_108.QC.jpg
7a8d3616e1f13f9477d09af0899fcf4f
e128381fd76843953c8bca1056b42454d610491c
26134 F20110319_AAASPJ chao_c_Page_116.jpg
04facde4e4d35c1d93136a44d9ac405d
640b1d30b2558237a90cc931e2839bd487f98544
8390 F20110319_AAASPK chao_c_Page_116.QC.jpg
302576ec3661e8bf8cac5f2cb61266d5
da0939a48a62f16f585d67e9c321de123e8c8bcf
30786 F20110319_AAASOV chao_c_Page_109.jpg
40c55c363407ee9a3ed588b193ca1c29
f49dcec1929b4f10f01186b50bf8f3c3e79badfb
23300 F20110319_AAASPL chao_c_Page_001.jp2
ca9e98a799ec666b7bf54fc4e04622af
2e79a760bb1857731ae5123b92167c91c5c05ed1
10471 F20110319_AAASOW chao_c_Page_109.QC.jpg
6362231b305ef1f60e9406fbb36460a2
1a0da942cf94fbb1a0a8cd9ddd1327e31c27c55c
F20110319_AAASQA chao_c_Page_016.jp2
b6d1b02f96243331287bdcdf0c63ecae
bc515a2a4fc844ecb03e981ad7d3b96ddaf55b20
5529 F20110319_AAASPM chao_c_Page_002.jp2
ca34a83f0b3b47de460293bec393e70c
46ec0570ec9240d469606bb07913152bc2adfa2c
26035 F20110319_AAASOX chao_c_Page_110.jpg
bd4e315bbb6d5a8d42b39dee1fd0d5f4
65383439ff9a77ef4597b46d99672faa4d95b038
93941 F20110319_AAASQB chao_c_Page_017.jp2
1c52641981d38b81e9abece1da6c8d91
203e3ef8a07f8fa3ba14ff4cff2e8bdc726362c3
8928 F20110319_AAASOY chao_c_Page_110.QC.jpg
6bcfc09b0212ad34f9ad74d93d253d29
f49ca2cff1cda6af89707b5104300e401a316fe4
1051978 F20110319_AAASQC chao_c_Page_018.jp2
e1b9457cb00c2fd9722f7aaa877a5210
72b3a80e15afbc7308b924cedf5dd0f65299820b
6335 F20110319_AAASPN chao_c_Page_003.jp2
19920282caa4a47a590e2ebe38acf3ed
e3bd23b5ef65c904eb1a93cada53711dc5705b27
15061 F20110319_AAASOZ chao_c_Page_111.jpg
5e235e1d6e61e388272f28597f9699bd
2dcebc6c0fd4ae0422772b57aa3b145fca7edf05
913340 F20110319_AAASQD chao_c_Page_019.jp2
26bea8b02001ca0399a61d2ef3cdbbce
1c01c8d3b5990c1be594e54d91d734f412193ff8
63674 F20110319_AAASPO chao_c_Page_004.jp2
84ae09e5a13094112775139b1c2b2c6c
6b9db498f248dae8af1bb2417ee063a85a638d74
1051959 F20110319_AAASQE chao_c_Page_020.jp2
3d1edd6f65ccbb640b9c9ac9f5c6712b
a99f76d38453a7101b72c1396ab6fdc4e7ed54ab
1051970 F20110319_AAASPP chao_c_Page_005.jp2
e4181e3e0fc94029ce70adb12f917c7a
9c37333d8956c43dda72140e33a5dd41773fd6de
1051986 F20110319_AAASQF chao_c_Page_021.jp2
e377a7293793a5ae1830b8339c3c3f1a
93c8d12898c2e5f544926bd583c8bf12585d7ea2
1051982 F20110319_AAASPQ chao_c_Page_006.jp2
76f3bcf1c58db8c9b96224e20fd4d192
1d1be629ceed5381122af8ae1db298726788a2a0
758551 F20110319_AAASQG chao_c_Page_022.jp2
27c5778d4e1fc650c7e4f00bc0802c3a
d09da46173a05c5a0a339555bf0efbb422b2b0f8
87876 F20110319_AAASPR chao_c_Page_007.jp2
1789fe0f3d96f59e25248cb944ce34a5
262f6894ff3e7475396d8da867e3ddaba1ff00db
64639 F20110319_AAASQH chao_c_Page_023.jp2
68bad7c745c9443cb7df01cdfd6acfc7
25c09fecc8b947f0a3f6d11c69da761d4859bfe4
818068 F20110319_AAASPS chao_c_Page_008.jp2
65fa4192657fc9fdbac5525d30adb274
cbd0502e873e203d929a14005c601bf828408006
721882 F20110319_AAASQI chao_c_Page_024.jp2
31da009243681b749b2c4c4fbdbc1c74
b12fff1036ab1358adf71cfc813708424daedd1a
F20110319_AAASPT chao_c_Page_009.jp2
c9eea8f2cfdc5965df709a084fdb13db
524270b29c4e3019d4b8030f585b60240c71509f
843347 F20110319_AAASQJ chao_c_Page_025.jp2
103fabc82bf9925c045c0ea1c622ae69
dc8721ab64d12c8c27695705bfd718598fa5d5ae
F20110319_AAASPU chao_c_Page_010.jp2
fba2e9d7064789d2421de1b94c17891b
b0381a798be7a4bce535141cec4f69a208ca3414
99465 F20110319_AAASQK chao_c_Page_026.jp2
4e95b7aff9e11ec1ee3a8a128d9bde8c
f377f78177f8476b8a0330a30e3d4b92a87540d3
374067 F20110319_AAASPV chao_c_Page_011.jp2
e3ff9749467a8f7eb681ad97112488cc
88f4edc523d0f268fad0ae42c5898c78c26b90c7
107523 F20110319_AAASQL chao_c_Page_027.jp2
e202eb1d0e45c02a2712df7f320a1f8b
99b69ba83d02eb1d0764534588a165819f3b2f36
82875 F20110319_AAASPW chao_c_Page_012.jp2
6e37bbe1ef1dbc7047a54581412e8357
a0b0963af6f30eeec6aabbb395251a15b63a56b7
113427 F20110319_AAASQM chao_c_Page_028.jp2
6aa61b04b1b7e950b249937cda291246
7e2f145255bffd9d20047c658571b5f8e794012d
53804 F20110319_AAASPX chao_c_Page_013.jp2
8b703a099cfc8bcf5351f4dac231b2bf
e021620e3a28fa8edd079b2828524795d51d1c1c
943882 F20110319_AAASRA chao_c_Page_042.jp2
273ec7df683740a1618b52a728fdfe50
4a893054964f3fde422a322744d0fb9a3dd7bad3
1051958 F20110319_AAASQN chao_c_Page_029.jp2
45da503237486af31531fc65303ca303
3aa490dc0469fe10460d5055eb0eaede41641799
95042 F20110319_AAASPY chao_c_Page_014.jp2
9d5f86e96b7af9acdfba143b8c18d9b2
09fcb8079cc7f4898f871ac8c6aca023f489b2f1
934135 F20110319_AAASRB chao_c_Page_043.jp2
e718a4633ec4421c138419f7fef8808c
3030f4db82bf451710e594b8c85a2768cca7f021
96127 F20110319_AAASPZ chao_c_Page_015.jp2
061bb5b018076250d0fc4ceb99fdfdc9
0a953f1761520d5a6ff799e35e3b4af83f9425b1
996186 F20110319_AAASRC chao_c_Page_044.jp2
eb2dcbccc2322171112d32a51153c5c2
3293ff8e2d4aa8ebbd530b115df3aa7eb4c18254
94071 F20110319_AAASQO chao_c_Page_030.jp2
f6a649b32b159f6da796234d7f2898fc
13872ec281349be2dc48409d8d83cfa31e9226d4
109154 F20110319_AAASRD chao_c_Page_045.jp2
d1f032e1ef7aa6ea92dd1a8e83c92704
e895496b26b0d78c3871fff3d2b0861c0a621086
1047450 F20110319_AAASQP chao_c_Page_031.jp2
da5b2b5a318310c544eea5c39f54318d
0157a3e2349a7b826152e84802a7e0aac780f47a
695821 F20110319_AAASRE chao_c_Page_046.jp2
9d2ab0b9e28b8332be293420e0b4cc97
338b33b89ec8e56cf18ebfd52c79fe5ee7af2a92
756787 F20110319_AAASQQ chao_c_Page_032.jp2
be55ada4fc89941ccbde076fe299dcba
d477a5ea05d09385f1eb0c8570950bb63169c44d
106290 F20110319_AAASRF chao_c_Page_047.jp2
ee33d6c615b0fc02436c6ebfccb52be7
b32614001ac1c9fe26e0cbfe5aef86e39c6b94a4
615498 F20110319_AAASQR chao_c_Page_033.jp2
1bb62baf2c9e9b7fb9476418f56f1e99
904d5f68e16f9a2d07335e4b77b66101757eb014
920762 F20110319_AAASRG chao_c_Page_048.jp2
f2a507f5eda58a3975717b6f0f19b711
7024172a285c5dd897d7f2c1d9b4b532127c6ec0
911119 F20110319_AAASQS chao_c_Page_034.jp2
5cfb60177950d6a266db66e825a4c588
75a1ebb4c41d18b1f65900e3a06abce0f1ef5128
94961 F20110319_AAASRH chao_c_Page_049.jp2
b0cd3c1585fb3e425c4325d47aa2fa95
bd1de47dd935ae141f52ea15242c07f4e6682a95
908119 F20110319_AAASQT chao_c_Page_035.jp2
8e2f4ddc1b94bf88363f252106ac7023
b7542c9dec4407481725f5b34b515d64e444c8f8
86858 F20110319_AAASRI chao_c_Page_050.jp2
2fcb7b416b3502433afc3fe98b5d80a7
72ba4ce6ba7f2b009dfeb2c05d1bb064422fddd1
734874 F20110319_AAASQU chao_c_Page_036.jp2
35e02a7f37d23ddd22d692e8050d6246
f5891e095789a5c94dd28ff96c1f9e822b0fe334
875465 F20110319_AAASRJ chao_c_Page_051.jp2
07f5db27babecb6916851feb75aead48
e2620a5fd90fbacb52d253df5bc53d8e6bf84d78
90553 F20110319_AAASQV chao_c_Page_037.jp2
49aa6d9f4430befc23c25a09b8579944
f34667301dab4e9129153137dad3e623376dd25e
111229 F20110319_AAASRK chao_c_Page_052.jp2
90863e26db88d314321e3fbfe914780d
8900cb0969f21becfa1a66f3e47712a81e688f90
1051848 F20110319_AAASQW chao_c_Page_038.jp2
20f56860b9bdd3636dd1155d381080a3
f8afeed1161c3f366c0d6ced0224de035fb3cf00
719976 F20110319_AAASRL chao_c_Page_053.jp2
54857711c2d812680c3e7207c51115d6
68bb00bfd97d9584a5f3184a88d0d031667aa3b1
14093 F20110319_AAASQX chao_c_Page_039.jp2
880e53707fcbabdd7510878d031de491
968eb158d94e25d072b25eb084e9e54c059cb6d8
1051892 F20110319_AAASSA chao_c_Page_068.jp2
3f39a7b53f95537631905f401eea485f
0e387e643056e27d7a0c1892c58c21219ebd1812
96383 F20110319_AAASRM chao_c_Page_054.jp2
5227cdbf891a2182bb8327591bd9da3f
ed0a972e93d6909bd43ce4a6c8fef4c49cf1df3e
96953 F20110319_AAASQY chao_c_Page_040.jp2
5e524ed63c0610b1793c20ef43451f3e
e241e1b7fe022ca8c724418511629483307fa9ac
917779 F20110319_AAASSB chao_c_Page_069.jp2
17b42895c24786d10ce56db13ce2ada8
eb7195ca4d045bc3c757629a83d2afce17d2d084
99080 F20110319_AAASRN chao_c_Page_055.jp2
8982e302a063bd69fa160e66fe2ce090
4048685e36e3b5d13ef1476cf2f7b3dbe08511f8
104853 F20110319_AAASQZ chao_c_Page_041.jp2
45d8023965c37cbd3888418938407cc2
ed841d896785781103723c7469034306066f2eda
967198 F20110319_AAASSC chao_c_Page_070.jp2
3a0000c3582835cab292052b01cf99ef
045c858332e797a72879ac6fd2bf31ed1858b66c
90452 F20110319_AAASRO chao_c_Page_056.jp2
0ba073072af5c06b3e96bd5a8b4dab2e
52e9dbdcf887402bc529f2d4119f4bddbf8e1653
1051905 F20110319_AAASSD chao_c_Page_071.jp2
4c9f8c9a518a84b5e2b67c7a2ac8bc81
be9f37ee5e1f5028ab96637de9ab57cde3ef7319
1051981 F20110319_AAASSE chao_c_Page_072.jp2
56793f08abb9a8e72432dd5b1adcdf7d
81f0ad40b44b28a742e36edd9771eb146b3faa29
422430 F20110319_AAASRP chao_c_Page_057.jp2
882a6d3e18d37a40ce265ea41c162703
455f779d5e30193c49d9985b052ef213c71b1dfa
799632 F20110319_AAASSF chao_c_Page_073.jp2
c800a21cd089965707a5136358419371
3445285cd38c16e6c37ec687ff3f5c3161c54559
844398 F20110319_AAASRQ chao_c_Page_058.jp2
095090955ceff1cf9abe8772d3c21e15
117bd6595df3fa20af4f009a428a8c6a3a22b6aa
995768 F20110319_AAASSG chao_c_Page_074.jp2
97edc42bfc22c1919165e28be14d6e40
f059c341b9da19c2021a4e86449573ff944dae62
92529 F20110319_AAASRR chao_c_Page_059.jp2
c2387e84f08afdd4761da9b73b39ad21
8f3818c89ae6f090c50d2cffa009d7a2f6f1d4eb


Permanent Link: http://ufdc.ufl.edu/UFE0011835/00001

Material Information

Title: Self-Configurable Communication Network for Wireless Multi-Robot Testbed
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0011835:00001

Permanent Link: http://ufdc.ufl.edu/UFE0011835/00001

Material Information

Title: Self-Configurable Communication Network for Wireless Multi-Robot Testbed
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0011835:00001


This item has the following downloads:


Full Text












SELF-CONFIGURABLE COMMUNICATION NETWORK FOR WIRELESS MULTI-
ROBOT TESTBED















By

CHUN-HAUR CHAO


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE


UNIVERSITY OF FLORIDA


2005

































Copyright 2005

by

Chun-Haur Chao

































This document is dedicated to my loving family.















ACKNOWLEDGMENTS

The author expresses his sincere gratitude to his advisor, Dr. Norman G. Fitz-Coy,

for his exhortation and motivation to drive this research in its attention to detail. The

author also expresses his gratitude to his committee, Dr. Gloria J. Wiens and Dr. Haniph

A. Latchman, for their instruction and guidance. The author acknowledges the University

of Florida's Mechanical and Aerospace Engineering Department for offering the

opportunity and financial support to finish the Master of Science degree.

The author thanks the friendship and selfless knowledge sharing found in the

members of AMAS (Autonomous Multi-Agent System): Andrew Tatsch, Svetlana

Gladun, Daniel Jones, Sharanabasaweshwara Asundi.

The author would like to express his gratitude for the unconditional support from

his parents, especially when his family is in a difficult situation. This research would

never have been completed without their general giving both spiritually and financially.

The author also appreciates his girl friend for encouragement and companionship.
















TABLE OF CONTENTS




A C K N O W L E D G M E N T S ................................................................................................. iv

LIST OF TABLES .................................................... ............ .............. viii

LIST OF FIGURES ......... ............................... ........ ............ ix

CHAPTER

1 IN TR OD U CTION ............................................... .. ......................... ..

1.1 A applications of M R S ................................................................................... 1
1.1.1 M military A applications .................................................... .. .... ......... 2
1.1.2 C civilian A applications ........................................ .......................... 5
1.1.2.1 M R S for jet engine inspection.............................. .....................5
1.1.2.2 R obot soccer com petition ..................................... .....................7
1.1.2.3 M ulti robot search and rescue ........................................ ..............9
1.1.3 Space-Based Applications .............. ............................... ............... 11
1.2 Fundam ental Issues .................. ............................ .... .... .. ............ 13
1.2.1 Autonom ous Behavior ................................. ............... ............... 13
1.2.2 Cooperative Operation and Communications......................................15
1.2.3 H ardw are R restriction ................................... ............................. ....... 16
1.3 M methodology ................................................................ ..... ............ 17
1.3.1 Autonom ous Behavior ................................. ............... ............... 17
1.3.2 Cooperative Operation and Communication ......................................21
1.4 Motivation and Scope of the Research .........................................................24

2 NETWORK COMMUNICATIONS .............................................................. 27

2.1 Evolution of Network Communications .................................. ...............28
2.1.1 M message Sw itching ........................................ .......................... 28
2.1.2 Circuit Sw itching ......................... ...... ................ ........ .. .......... 30
2.1.3 P acket Sw itching .......... ........................................ .......... ......... 32
2.2 L ayered A architecture ............................................... ............................ 34
2.2.1 O SI R reference M odel ........................................ ........................ 34
2.2.2 T C P/IP structure......................................................... ................ 38
2.3 W wireless Communications and Issues.................................. ............... 41
2.3.1 Medium Access Control Protocol........................................................42


v









2.3.2 Ad hoc and Infrastructure Topology........................................................46
2.4 Com m unication Perform ance ........................................ ...................... 48
2.4.1 B andw idth ............................................................................. 49
2 .4 .2 T ran sm mission L oss ........................................ .................................49
2 .4 .3 T hroughput .............................................. ................ ...... 50
2 .4 .4 L ate n cy ............................................................................................. 5 1

3 WIRELESS MULTI-ROBOT TESTBED...........................................................52

3.1 Hardware Architecture of the testbed ............................................................52
3.2 W wireless M obile R obot .......................................................... ............... 54
3.2.1 Pow er M odule ................................................ .. .. .. .. ............ 56
3.2.2 Com m unication M odule ............................................. ............... 57
3.2.3 H ardw are Control U nit ........................................ ....... ............... 57
3.2.4 Processing Unit ....................................................... ... .............. 58
3.3 P ositioning Sy stem ............. .................................................. .. ............ 59
3.4 Operational Area .................. ......................... ........ ................. 61

4 PR O T O C O L SU IT E ............................................................................. .............. 63

4.1 Lim stations and R equirem ents ........................................ ...... ............... 63
4.2 Local Area Network Architecture .......................................................65
4.3 Protocol Specifications .......................................................67
4.3.1 Data Link Layer Protocol .............................................. ...............69
4.3.1.1 L ink m anagem ent ................................... ..................................... 71
4.3.1.2 Forw ard error correction .........................................................72
4.3.1.3 Feedback error correction................................... ............... 73
4.3.2 Agent Com munication Language ....................................... .............74

5 SELF-CONFIGURABLE TOPOLOGY ....................................... ............... 80

5 .1 E lig ib ility L ist ................................................................................. .. 8 1
5.2 Self-configuration .............................................. ....... ............................. 82
5.3 T est C configuration ........................................................................... .... ... 84
5.4 N etw ork Initialization ........................................................................ ...... 85
5.5 Follow er Failure .................. .......................... .... .... ................. 86
5 .6 L ead er F ailu re ................................................................................. 87
5.7 C control Privilege Transfer ........................................ .......... ............... 88

6 CONCLUSIONS AND FUTURE WORKS................................... ...............91

6 .1 C o n clu sio n s ................................................................................... 9 1
6 .2 F future W ork s .............................. .......................... .. ........ .... ..... ...... 92

ACRONYM S ................................... ................................... .......... 94

LIST OF REFEREN CE S .............................................................................. 99









BIOGRAPH ICAL SKETCH .............................................................. ............... 103
















LIST OF TABLES

Table p

1.1 Classifications of m otion planning. .............................................................................18

1.2 Classifications ofM P algorithm ............................................................................ 18

1.3 Comparison of centralized control and decentralized...............................................25

2 .1 M orse code ........................................................................... 29

2 .2 A SC II table. ......................................................... ................. 30

2.3 List of netw ork protocols. .................................................. ........................................... 40

3.1 Comparison of different communication channels ............................................... 57

3.2 Specifications for processing unit ........................................ .......................... 59

4.1 Proposed agent communication language......................................... ............... 75

5.1 Com puter configurations for the test ........................................ ........ ............... 84
















LIST OF FIGURES

Figure page

1.1 Key DARPA accomplishments since 1960s.................. .......................................3

1.2 R obotic evolution ............... ...................................... .... ................ .... 5

1.3 NASA Glenn miniature mobile sensor platform. ........................................ ...............6

1.4 The concept of jet engine inspection. ....... ................... ......... .............. ............... 7

1.5 Sm all size league in RoboCup 2004. ....................................................................... 8

1.6 Control diagram of robot soccer. ............................................ ................... ..... ..9

1.7 Search and rescue operation by MOVER system....................................................11

1.8 Layered m ulti-robot architecture. ........................................ .......................... 12

1.9 Planetary Surface Robot Work Crew (RWC)......................................................16

1.10 The solution path is shown in the bold lines in the visibility graph ........................19

1.11 Obj ect-dependent cell decomposition...... ................................... ....................... 20

1.12 Q uadtree m option planning ............................................................... .....................20

1.13 Function of the potential field ......... ................. .............................. ............... 21

1.14 The stabilization of formation control. ........ ......... .................. ............... 23

1.15 Estimation of sensor positions using Kalman filter. ...............................................23

1.16 Temperature gradients inside the target building...........................................23

2.1 Telephone netw ork connections. ............................................................................ 31

2.2 N etw ork Sw itching. ............................................. ................... .... .. ....33

2.3 OSI reference m odel. ......................... ........ .. .. ..... .. .............35

2.4 Comparison of layer definition between OSI model and TCP/IP structure.................38









2.5 Encapsulation of header and error check code into data units............................... 40

2.6 C SM A -C D ............................................................................44

2.7 Hidden terminal problem for wireless network. .................................. ...............44

2.8 C SM A -C A ...................................................................45

2.9 N etw ork topologies ................. ................................. ........... ..... ...... 47

3.1 Hardware architecture of the testbed. ........................................ ...................... 53

3.2 W ALKER for multi-robot testbed. ........................................ ......................... 55

3.3 Block diagram for modules on WALKER.......................................... ...............56

3.4 Hardware pictures for power module. .............................................. ............... 56

3.5 Hardware pictures for communication module................................. ...............57

3.6 H ardw are and hardw are control unit................................................. ....... ........ 58

3.7 P C /104 processing unit. ...................................................................... ...................59

3.8 Block diagram of the PhaseSpace positioning system...............................................60

3.9 Pictures of hardware for PhaseSpace positioning system................ .............. ....61

3.10 G eom etry of the tested. .............................................. .......... ........................ 62

4.1 Comparison of the interconnections of different networks..................................65

4.2 D dedicated w wireless netw ork layers......................................... .......................... 69

4.3 Bit-w ise form at of the control field ........................................ ........................ 71

4 .4 N orm al response m ode ........................................................................ .................. 72

4 .5 A R Q m eth o d s......................................................................................................... 7 3

4.6 Examples of ACL messages. ............................... ........................ 77

4.7 Comparison between different encoding methods for ACL.......................................78

4.8 Comparison of the performance on different message encoding methods..................79

5.1 Flowchart of self-configuration mechanism ..................................... ............... 83

5.2 D display during the test ........................................................... .. .......... .... 85



x









5.3 N etw ork initialization process. ............................................ ............................ 86

5.4 Topology configuration when follower fails. ................................... ............... 87

5.5 Topology configuration when the leader fails. ................................. .................88

5.6 Topology configuration for the control privilege transfer ........................................89















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

SELF-CONFIGURABLE COMMUNICATION NETWORK FOR WIRELESS MULTI-
ROBOT TESTBED
By

Chun-Haur Chao

August 2005

Chair: Norman G. Fitz-Coy
Major Department: Mechanical and Aerospace Engineering

The Multi-Agent Systems (MAS) have been studied over decades. Various issues

were well discussed conceptually. The implementation of MAS on the physical hardware

is an important phase for the research development. Hardware validation process

promotes the theoretical concepts to realistic problems. Nevertheless, the hardware

implementation is somewhat costly and sophisticated. A multi-robot testbed is a cost

effective solution to implement different concepts for MAS.

This research proposes an architecture for the hardware implementation of a MAS.

A description about the design of a wireless multi-robot testbed for further MAS research

is provided. Relevant research topics including path planning and cooperative control are

also briefly introduced in the thesis.

Meanwhile, the communication is a prerequisite before the validation of MAS. The

work in this thesis will mainly focus on the network architecture for the communication

between robots. The objectives for the network communications are to simplify the









existing framework and maintain the flexibility for any further revision. Also, in order to

enable the inter-cooperation between robots, the agent communication language is used

to provide a standard for the conversation.

Moreover, the MAS provide a decentralized control scheme for the system. It is

more robust for any exceptional incidents and failures. In order to enable a more robust

communication environment, the self-configuration of the network topology is proposed

in this thesis as well. For such a self-configurable network, the failure of the robot for the

communication can be accommodated. Dynamic adjustment to an optimal topology could

therefore be made.

The work presented in this thesis provides a hardware solution for MAS research.

The self-configurable topology offers a flexible network scheme for wireless

communication and decentralized control scenario.














CHAPTER 1
INTRODUCTION

Multi-Robot Systems (MRS) have brought the robotic researches into a new

paradigm. Not only the functionalities but also the cooperative operations between robots

have excited the interests and attentions of the robot communities. The widely applicable

MRS research could be mainly divided into three categories: space-based, military and

civilian uses. Moreover, for some of the issues, for example, cooperative control, path

planning, and communications have become more and more important to the

development of MRS. However, the physical implementations of such systems may be

restricted by the limitations of hardware. The dimension of the robot body, extra payload,

the performance of the sensors and actuators, or the controller's information process

capability could all substantially affect the overall performance. In addition, the

communication capability is one of the restrictions for some implementations like

spacecraft communications or smaller size robots. In this chapter, the applications for

MRS and the corresponding issues will be mentioned. This effort will facilitate a further

integration of the hardware implementations provided in the thesis as well as the network

communication design into a larger framework of the MRS research.

1.1 Applications of MRS

Traditionally the robot communities focused their research interests in the domain

of the single robot applications. However, with the growth of the semiconductor and

communication technologies, for example, MEMS, wireless network and Global

Positioning Systems (GPS), the development of mobile robot technology has been









transferred to the multi-agent level. The studies of the cooperative and collaborated

control for MRS has been extensively discussed and implemented since last decade (see

JRP [2]). The applications for MRS are fairly diverse from military unmanned espionage,

mine sweeper to rescue or space exploration mission. The following section summaries

all the applications into three major application categories military, civilian, and space-

based. Some relevant developments will be mentioned as an example..

1.1.1 Military Applications

The effort to apply the MRS research for military uses has been initiated and

performed by several agencies and programs, for example, Defense Advanced Research

Project Agency (DARPA) [1] and Joint Robotics Program (JRP) [2]. The objectives for

the robot researches within these programs include:

* Increase the autonomous mobility

* Refine the tactical behavior

* Design the innovative platform

* Minimize the robot dimension

DARPA has been successfully been merged the cutting edge technologies into the

robot researches in the past couple of decades, for example, the communications and

artificial intelligence enhance the controllability and autonomy on the robot. Figure 1.1

shows the major accomplishments by DARPA since 1960s. With the current

achievements on the communication systems, for example, GPS technology, the mature

of the network development and the various artificial intelligent algorithms for the

autonomous behavior, the Multi-Agent Systems (MAS) could have been able to promote

the status from software agents to physical agents. Therefore, as for the military










applications, the MRS including various unmanned vehicles would become new thrusts

for the warfare development.






turn J-CAS Future Combat System
Vela Hotel


M-16 Rifle |inhile Robl[h iJ



Ground .I n ,'
Surveillance Radar Llobal Hawk
-Arpanet

Uncoo.ld IR
slealrh Phrel r .1,Predator



Sea Shadow GPS Taurus L-anch BAT
Vehicle


Figure 1.1 Key DARPA accomplishments since 1960s. [1]

Moreover, the integration of separate Unmanned Ground Vehicle (UGV) robotics

developments and projects had also been made into the Joint Robotics Program [2] by the

Office of the Secretary of Defense (OSD) in 1990 at the recommendation of the Senate

Appropriations Committee. This effort also advances the successfully global deployments

of UGV in some of the military operations including Bosnia, Afghanistan, and Iraq in the

past decade.

UnManned Systems (UMS) have been validated widely during the mentioned

ongoing services. UMS as described in the concepts below are envisioned to contribute

the increase of mission effectiveness and are planned for integration into service force

structures [2]:









* Army Future Force: Future Combat Systems (FCS)

* Marines/Navy Gladiator Tactical Unmanned Ground Vehicle: Autonomous

Operations

* Air Force Air Expeditionary Warfare: Robotics for Agile Combat Support and the

Airborne Explosive Ordnance Disposal Concepts

However, the goal of JRP has been to develop a diverse family of UGVs and to

foster service initiatives in ground vehicle robotics to meet requirements for greater

mission diversity and increasingly more autonomous control architectures [2]. As in

Figure 1.2, the maturation and transition of the technology to the robotic systems will

feature the robotic services with more autonomous capabilities. Therefore, the enhanced

object recognition and tactical behavior could enable the use of robotics to a fairly

extensive and effective manner. As it is shown in Figure 1.2, the advanced technologies

such as route planning, mission planning and target recognition will lead the MRS from a

teleoperational service to a more and more autonomous service. As in the current

progress of MRS, the system is undergoing the development of route planning and

heading to a mission planning level, where the robot autonomy will exceed the level of

human intervention. Both the route planning and mission planning have the critical need

to perform the dispatched tasks cooperatively. Such a cooperation behavior requires a

reliable and robust communication system. Moreover, a further advance phase in Figure

1.2 requires higher and higher communication data rate. Therefore, the importance of the

communication system in a MRS will be higher and higher.




































Figure 1.2 Robotic evolution. [2]


1.1.2 Civilian Applications

The efficiency and effectiveness of the MRS can not only facilitate for military

purposes but also benefit civilian uses. Sometimes the tasks are too complicated for a

single robot and we need multi robots to work cooperatively. For example, the search and

rescue mission using multiple vehicles will influentially decrease the time required to

complete the mission. Multiple robots can also be used for moving a larger object. Brief

descriptions on some of the general applications will be discussed in this thesis.

1.1.2.1 MRS for jet engine inspection

The maintenance of the aircraft is critical to the aviation safety. Jet engines undergo

examinations for detecting potential flaws on the surface of their components, such as









cracking and erosion. Usually inspection methods are usually either the invasive

borescopic or a full teardown, both of these methods are time-consuming. Full teardown,

however, is even more time consuming and costly, and is often applied for only the

situations when damage is detected and the replacement of parts is necessary.

NASA Glenn Research Center [3] proposed another approach for the inspection of

the jet engine health. Instead of manual inspection procedures, miniature mobile sensor

platforms could provide another option. The mobile robots equipped with the vision and

communication systems can roam arbitrarily on the surfaces inside the engine. The robots

could hence send the internal image of the jet engine back to the station. Therefore we

can go through the engine inspection process with less human power and time. Figure 1.3

and Figure 1.4 show the miniature mobile robots in the inspection concept.


Figure 1.3 NASA Glenn miniature mobile sensor platform. [3]





























Figure 1.4 The concept of jet engine inspection. [3]

1.1.2.2 Robot soccer competition

The concept of robots playing soccer was first introduced by Alan K. Mackworth

[4] in 1993. By using the global vision system, the main station could acquire the

orientation and position of each different robot. A wireless communication is also

required to transmit the commands to the robots. An artificial intelligence is involved to

determine the strategy to compete against the other team. In July 1997, the first official

conference and games were held in Nagoya, Japan. Followed by Paris, Stockholm,

Melbourne and Seattle where the annual events attracted many participants and spectators

[5]. There are five different leagues in the RoboCupSoccer: simulation league, small size

league, middle size league, four legged robot league, and humanoid league. Comparing to

the other MRS applications, robot soccer is highly dynamic and the state change is in

real-time. No human intervention is allowed during the period of the game. Its situation

in a MRS development phase contrast to Figure 1.2 is a more advanced phase including









the object recognition and situational awareness. The research fields cover various areas

from artificial intelligence to robotics. Such areas include real-time sensor fusion,

reactive behavior, strategy acquisition, learning, real-time planning, multi-agent systems,

context recognition, vision, strategic decision-making, motor control, autonomous robot

control and more. Figure 1.5 is the picture of a competition in the small size league in

RoboCup 2004 at Lisbon, the regulation restricts each robot must be able to fit inside a

180mm diameter cylinder. In Figure 1.5, different colors and marks at the top of each

robot is used to identify the position and orientations of different robots from cameras.






















Figure 1.5 Small size league in RoboCup 2004. [5]

Due to the dimension restriction, the control of the whole team is usually somewhat

centralized in order to reduce the onboard processing need. A global vision system is then

used to trace the robots and ball. The control diagram is in Figure 1.6. The hardware

architecture adopted in Figure 1.6 is, however, fairly similar to the wireless multi-robot

testbed we will discussed later in the thesis. An overhead camera system is used as the
testbed we will discussed later in the thesis. An overhead camera system is used as the








object positioning system for the robot localization. The image processing is dealt by a
base station, and proper control commands for the robots are transmitted by a wireless
transceiver. Therefore, the robot soccer can also be taken as a platform for the evaluations
of various MAS concepts, as the same purposes for the wireless multi-robot testbed.

Overhead
Camera




PC Base
I Station








U.-
Soccer Robots RF Transciver

Figure 1.6 Control diagram of robot soccer.
1.1.2.3 Multi robot search and rescue
The task of search and rescue can be performed by a single robot. RobotCup [5]
also has a separated domain called RoboCupRescue. The robot is required to be operated
in a dedicated scenario autonomously. However, the cooperative multi-robot search and
rescue operation could greatly increase the efficiency or enable some capabilities which
couldn't be performed by a single robot. James S. Jennings et al. [6] discussed the









cooperation of robots for a search and rescue mission. The following capabilities are

required to perform such tasks.

* Navigation and localization

* Search

* Object recognition

* Communication with other members in the team

* Cooperative manipulation of large objects

The MRS demonstrated in Figure 1.7 is named as "MOVER". The algorithm of the

proposed MOVER system [6] performs the task in Figure 1.7 by the following steps:

Step 1: A workstation and 5 robots idle for the initial status.

Step 2: The workstation initiates the program to perform the search and rescue.

Step 3: All robots start to search the house shape object.

Step 4: One robot finds the target while all the other robots still searching.

Step 5: The robot that found the target notifies all the other robots that target has

been found.

Step 6: The other robots are heading toward to the target.

Step 7: Another robot arrives the target and notifies other robots.

Step 8: The last robot arrives the target and notifies other robots.

Step 9: All the robots manipulate the target toward the intended location

collaboratively.

































Figure 1.7 Search and rescue operation by MOVER system. [6]

1.1.3 Space-Based Applications

The MRS researches have also been studied in many of the space mission projects,

for example, NASA's Mars exploration project (See NASA [7]) and the Demonstration

of Autonomous Rendezvous Technology (DART) mission (See NASA [8]). The space-

based applications varied from formation flying, cooperative control for multi-arms, to

autonomous mobile robots for space exploration. Comparing to the terrestrial-based

applications, the space-based applications usually have more serious technical concerns

in the following characteristics:

* Uncertain environments

* Communication delays

* Limited sensing and actuation

* Scalability









Dani Goldberg et al. from Carnegie Mellon University [9] discussed the

synchronization and coordination for mobile robots for the application to space

exploration. The Mars exploration scenario has been set in the discussion. The distributed

layered architecture is proposed for the highest possible scientific return on the given

tasks. Due to the limitations on the communications (bandwidth and latency), the

centralized control is not reliable. So the robots are responsible for making the decisions

based on the priority of the tasks and how the tasks are to be accomplished. The

architecture is shown in Figure 1.8. The planning layer sends the plans to the executive

layer, which could further decompose tasks into subtasks and dispatch them based on the

temporal constraints imposed by the plan. The behavior layer is responsible for the

control of the robot or updates the information from the sensors or the status. Also, the

executive layer is responsible for monitoring the tasks status and returning them to the

planning layer. This layered architecture can be also used on a single autonomous robot.

It is shown in Figure 1.8.


Mszket& &*ans
paring t planning planning planning .



TDL eave- executive executive .
Scheduer -



I)II


kin behavior -.p behavior behavior .

Robot 1 Robot 2 Robot 3

synchronization/coordination


Figure 1.8 Layered multi-robot architecture. [9]









In Figure 1.8, each robot has the mentioned three layered architecture. The

communication can occur either vertically between either different layers on the same

robot or horizontally the same layers on different robots. The information of sensor data,

plans, or tasks in this architecture could hence be exchanged and coordinated. However,

the communication performance in this architecture dominated the system performance.

Requirements for a higher bandwidth and lower latency communication in order to

coordinate the action are needed, which might not be always allowed.

1.2 Fundamental Issues

The various applications presented above show different requirements and restrictions.

A good understanding on the potential factors that might affect the performance or

feasibility of a MRS application is critical. A system design or evaluation must be

provided under these considerations. Therefore, the following key factors to the MRS

application would be addressed in the section:

* Autonomous behavior

* Cooperative operation and communications

* Hardware restriction

1.2.1 Autonomous Behavior

Some applications only require the base station to centrally control different robots.

However, since the robots don't take charge of the data analysis, the information gathered

by various sensors on different robots need to be transmitted to the base station. Also, for

the simplicity of the communication, the information will mostly be shared on only the

base station. The drawback of the non-autonomous robot is it can't be operated unless the

control command is given from the base station. Therefore, the non-autonomous robot

itself can be taken as no more than a set of sensor/actuator instead of an intelligent









"agent". Nevertheless, the obvious advantage is the design of such a MRS system can

hence be greatly simplified and much easier to be implemented. An example of non-

autonomous MRS is the small size robot soccer mentioned earlier. The hardware

requirement for an autonomous system is usually depending on the computational and

environment sensing needs. From IR sensor, voice detection, thermometer, to relatively

complicate systems as well as the vision system or laser range detector could all be

possible options for an autonomous system.

As the earlier discussion, the developments for a MRS toward system autonomy

have several different phases from pure tele-operation to full autonomy (See Figure 1.2)

during its evolution. Many of the centralized controlled systems are still capable for low

level maintenance and fault detections. However, the high level task schedules or

tactical behaviors still need to be assigned by the central station. Multi-robot search and

rescue is usually operated under this mode. Each robot in the system can perform its task

to search the possible target autonomously. The decision making is determined by the

base station. Another example is the satellite system. Navigation, attitude control, or

health monitoring modules provided onboard can have the satellite to survive without

base station under its autonomous behaviors on the orbit. However, the determination of

the flying orbit or docking with other spacecraft is still currently controlled by the

commands transmitted from the ground station.

The design of the necessary autonomy level for a system is mission dependent.

Cost effectiveness could vary case of case. For example, the cost of the implementation

for the vision on a small size robot (< 180mm) is usually expensive. Not only the

consumption for the vision module but also the improved capabilities to process the









image data. Communication or power system may also need to be enhanced in order to

meet the operation requirement. This may be a big challenge to be implemented on an

embedded system and significantly increases the cost of a system.

1.2.2 Cooperative Operation and Communications

Another critical issue for MRS is the cooperation of the robot operation.

Cooperative operation can be used on a MRS to improve the task performance or enable

additional features. The time interval spent to search over a terrain will be significantly

reduced by using multiple robots. Also, the cooperative localization can help each robot

locate itself and understand the scene better. Sometimes the assigned task is too

complicate for a single robot to accomplish and needs to be executed by multiple robots.

For example, NASA's Planetary Surface Robot Work Crew [10] coordinates the grasp,

transport and placement of extended structure using multiple robots, as in Figure 1.9. The

formation and cooperative control must occur in such a scheme.

A cooperative operation includes many different level problems from task

assignment and schedule to the cooperative control and localization. One of the most

essential components for cooperative control on MRS is communication. An autonomous

robot can have the interactions via environment or sensing without any communication

[11]. However, in most situations, the most effective method to share and update required

information with each other is a data network. Multi robot search and rescue requires the

sharing of information gathered from different sensors. As the operation procedures

mentioned in Figure 1.7, the collaboration requires robots communicate with each other

as a network. G. Kantor et al. [12] discussed the using a network of distributed mobile

sensor systems for an emergency response problem. Multiple radio beacons have been

deployed in the target building in order to estimate the gradients of temperature in the









building in the discussion. The communication network is a prerequisite for many

cooperative operations.























Figure 1.9 Planetary Surface Robot Work Crew (RWC). [10]

1.2.3 Hardware Restriction

The hardware restriction dominates the capability of the MRS. For the mobile robot,

the resources are sometimes fairly limited in the system. The power supplied on the robot

determines the capabilities of the embedded modules: the controller, all the working

sensor and actuator subsystems. Unfortunately, the battery is sometimes relatively large,

heavy and a considerable part of the total weight, which might deteriorate the overall

performance or disqualify the system requirements. However, the autonomous system or

decentralized control often requires better processing capabilities, which results in higher

power consumption. For the physical restrictions on the mobile robots, there are some

important considerations for a hardware design [11]:

* Centralization/decentralization









* Differentiation

* Communication structure

* Modeling of other agents

A careful review on all of these factors must be made and taken into consideration

while dealing with the design problem for the MRS. An example design for small

satellite for multi-spacecraft mission can be found in [13].

1.3 Methodology

In the previous section several issues for the MRS have been addressed. They are

critical factors to the MRS. The existing solutions for these issues will be discussed in

this section now.

1.3.1 Autonomous Behavior

The autonomy for the robot system involves many different disciplines. Image

process, control algorithm, artificial intelligence, and motion planning may sometimes be

needed for an autonomous system. An important aspect of the autonomous behaviors for

a mobile robot is Motion Planning (MP). Moreover, a mobile robot is often required to be

able to explore in an uncertain terrain. Hence, sensing, obstacle avoiding and planning for

the optimal path would be the most critical problems for the mobile robot.

Y. K. Hwang and N. Ahuja [14] summarize the recent developments for motion

planning problems. Before the discussion of motion planning, we need to classify the

type of the problems and problem solving algorithms so appropriate algorithms could be

highlighted regarding to specific problems. Table 1.1 and 1.2 list the classifications of

different problem types and approaches. Also, a proper method to describe the

environment, including the robots and obstacles, is essential to MP problems. Instead of

using the classical world space representation, which the physical space robots and









obstacles exist in, configuration space is more frequently mentioned in MP study.

Configuration space is a set of parameters that completely specify the positions of every

point of the robot or obstacle.

Table 1.1 Classifications of motion planning.
Yes No
Can objects change shape? Conformable Non-conformable
Time varying? Time Varying Time invariant
Restriction on the motion of robots? Constrained Unconstrained
Availability of the obstacle information Dynamic Static

Table 1.2 Classifications ofMP algorithm.
Limited Unlimited
Completeness Heuristic Exact
Scope Local Global

Various approaches have been developed for MP problems. The applicability of each

algorithm may be wide or restricted. Nevertheless, most of the algorithms could be

separated into the following approaches [14]: skeleton, cell decomposition, potential field.

Different approaches are not necessarily exclusive. Between these algorithms, the

technique used to solve the MP problem is sometimes a hybrid method. A better

performance could sometimes be obtained via the hybrid algorithms.

* Skeleton

The major advantage of skeleton approach is the simplicity of calculation. It

simplifies the MP problem into a network of one-dimensional lines so the search can be

restricted in the connections between nodes. This algorithm includes three phases. In the

first phase, the robot moves from its initial configuration to a node in the skeleton. In the

second phase the robot moves from a goal configuration to a node in the skeleton. The

third phase connects two points by using the lines in the skeleton. Figure 1.10 shows the

visibility graph of polygon in the plane. S and G in the figure represent the starting









position and goal position respectively. The visibility graph is the collection of lines in

the free space that connects different features. There are O(n2) edges in the visibility

graph. n is the number of features.


















Figure 1.10 The solution path is shown in the bold lines in the visibility graph. [14]


* Cell Decomposition

The cell decomposition, as the name suggested, decomposes the free space into a set

of simple cells. For the motion planning, we also need to compute the adjacency of the

cells. The decomposition approach can either be object-dependent or object independent.

The object-dependent method requires less cells. However, the computation complexity

for the boundaries and adjacencies is high. It is shown in the Figure 1.11. First we decide

the boundaries and adjacencies of cells by all the sidelines of the obstacles. Then we

could determine the cells including the path from S to G.

The object-independent decomposition generally uses more cells. However, the

calculation is less specifically for nontrivial objects. Quadtree [15] is used for a 2-D









motion planning problem, or octree for a 3-D motion planning problem. The motion

planning by quadtree is shown in Figure 1.12.


G

87
.&-








S2 3



Figure 1.11 Object-dependent cell decomposition. [14]


start


Figure 1.12 Quadtree motion planning. [15]

* Potential Field

The potential field constructs the environment by using the scalar function called

potential. The goal configuration is set to be minimum, and a high value at the occupied

space. The function is sloping down anywhere else toward the goal configuration. By this

potential setting, the robot could therefore reach the goal configuration by following the

negative gradient. The example of the potential field is in Figure 1.13.

















/

(a) (b)
2





-20



(c)

Figure 1.13 Function of the potential field. [14] (a) Obstacles has high potentials, (b) The
minimal potential locates at goal, and (c) The path to the goal from start could
be found along the negative gradient.

It could be observed from above discussions of various MP methods that a global

understanding to the environment is a requisite. This can be solved by either using a

global positioning system or a cooperative data sensing network, which will be

mentioned in the next section.

1.3.2 Cooperative Operation and Communication

The cooperative operations for the MRS system are diverse. Different cooperative

behaviors are developed, for example, formation, sensor fusion, cooperative localization

and control. The disciplines involved to solve these problems are also very different.

Three most common issues are mentioned here.

* Formation control

The formation problem is the most frequently discussed problem in the mobile

MRS. It is the fundamental problem to control mobile vehicles. Many applications like










search and rescue operations, robot soccer, formation flying of Unmanned Air Vehicles

(UAV) required the cooperation between vehicles. Reza OlfatiSaber and Richard Murray

[16] uses a set of parameters to present the formation graph. Then by using a structural

potential function, the local collision free stabilization of formation of multiple vehicles

can be obtained. Figure 1.14 (a) and (b) shows the formation for 3 and 6 vehicles

respectively.

* Data fusion

The MRS application usually requires the system to be operated in an uncertain

environment. Therefore, the capability to sense the environment becomes an important

function. Sensing can include simple measurements from temperature and range to

obstacles, to sophisticated multimedia data. However, the sensing information from

different robots needs to be further fused in order to gain a better understanding of the

environment. The methodology to exchange, process, or integrate the information

becomes another issue. A distributed mobile sensor system is mentioned in [12]. The

dynamic localization of all devices is estimated by Kalman filter, Markov method, and

Monte Carlo. The location estimation of unknown tags by Kalman filter is shown in

Figurel.15.

20 20

15 10

C10 .
0
0 5 CIL

S-10

-5 -2
-20
-10 -5 0 5 10 15 20 -30 -20 -10 0 10 20
x-position x-position
(a,) (b)










Figure 1.14 The stabilization of formation control. [16] (a) formation for 3 vehicles and
(b) formation for 6 vehicles


+!+i
-----------0-+--- ---+-------------------------------------



I o+
----.--------- ------- -------- --------*-------- ------- -------- -------




S............. .. ........


---------------- ---- -





Figure 1.15 Estimation of sensor positions using Kalman filter. ( "o" is the true location,
"+" is the estimation )

The temperature gradient map is generated then by the ad-hoc network of the

Mote sensors, as in Figure 1.16. Therefore, the temperature inside a scene can be

dynamically monitored and the better decision to rescue lifes can be made.


Termpnera Map at T.1400 Seconds Z






3 V11




V"O Inn I 11 i F YII M a M I t1 a 11 l



Figure 1.16 Temperature gradients inside the target building.

* Task schedule

A distributed layered architecture for mobile robot coordination is proposed in [9].

The autonomous high level task decomposition and subtask dispatch service, especially









for time stringent cases and cooperative behavior, are mentioned. However, for non-time

stringent mission, that the delay of operation will not cause catastrophic effects, a market

based auction mechanism is proposed by Brian P. Gerkey and Maja J. Mataric[17]. The

mission is constructed into a hierarchical structure first. The child task must be performed

before the parent task is performed. The parent task is responsible for assigning child task

to the robots and monitoring the status of the tasks. The auction is processed in the

following steps:

1. Task announcement

2. Metric evaluation

3. Bid submission

4. Close of auction

5. Progress monitoring/contract renewal

The purpose of the auction mechanism is choosing the most appropriate agent to

execute the task. The assumption here is the auction is always won by the most

appropriate agent. This concept will later be used for a network self-configuration

procedure.

1.4 Motivation and Scope of the Research

A common component for all level problems for cooperative behavior mentioned

above is the communication. Just as with human beings, communication is a requirement

for a successful cooperation. Also, for the mobile robot, wireless communication is

needed instead of the communication through wires. The wireless Ethernet protocol IEEE

802.11 provides a well developed solution for wireless communication network.

However, for some physical restrictions, like signal coverage, reliability and hardware

complexity, IEEE802.11 could not always occur in all of the MRS, especially for deep









space applications. The cooperative control scheme for the MRS could be mainly divided

into two types: decentralized and centralized. It also relates to the communication

architecture and many other factors. Table 1.3 compares the differences of the centralized

and decentralized control. More details about the differences for communication will be

discussed in the next chapter.

Table 1.3 Comparison of centralized control and decentralized
Centralized Decentralized
Hardware
Low High
requirement
Autonomy Low High
Communication
Cn Infrastructure Ad hoc
architecture type
Communication .
com nictin Simple Complicate
complexity
Communication .
Comn Unidirectional/Bidirectional Bidirectional
direction

The research interest in this thesis is mainly focused on the hardware implementation

and the communication network solution. For centralized control, some simple

architecture, for example, channelization is often used for communication network to

share the medium. However, a central station must be used and constantly update the

status in order to keep all the devices functioning. A service like packet switching or

virtual circuit switching is often required to be provided for a decentralized control

scheme. As the potential issues discussed above, a generic wireless Ethernet solution can

not be adapted by all the MRS design. Proper modification must be made in order to

customize the communication system to fit the specific scenario and other hardware

designs. Chapter 2 will offer a technical review on the network developments before the

presentation of the wireless network design. Several network considerations like hidden

terminal problem and ARQ will be implemented on this network. The effort provided in






26


the thesis is a minimal wireless network infrastructure for a wireless MRS testbed. It

offers a wireless networking solution for a small size MRS for the distributed control

purposes.














CHAPTER 2
NETWORK COMMUNICATIONS

The importance of communications has been laid out with the discussion of chapter

1. Network communication, as defined here, is a service which could provide the

capability to transfer the information between three or more different hosts. Network

communication is one of the essential features in MRS since the need for different robots

to work cooperatively and the information exchange between different robots. In other

words, the network services facilitate different robots to establish the interconnections

instead of to be operated individually. A very fundamental problem for network

communications is the medium sharing issue. A couple of techniques have been

developed, for example, channelization, Time Division Multiple Access (TDMA) and

Code Division Multiple Access (CDMA). For a distributed control scheme, a static

medium access control technique like channelization is not enough. The communication

of information has to be more dynamic. A packet switching or virtual circuit method is

usually considered. This chapter introduces the concepts and developments for the

network communications.

An understanding of the evolution of network developments could highlight the

possible issues and functionalities during the developments of a network communication

environment. Some existing works will also be referenced in this chapter. For instance,

the layered network concepts and necessary features for the Ethernet like requests of

retransmission could also be properly applied to any specific MRS communication

though the existing protocols may not always proper under all the circumstances.









Therefore, an introduction to the history of the network developments will be helpful to

identify the network communication problems. Moreover, in order to structure and

simplify the various problems, the layered architecture for the network is an effective

manner to organize the functionalities for network. Therefore, a brief introduction of

Open Systems Interconnection (OSI) model for discussing different functionalities of

network communications will be presented in this chapter. Along the network

developments, an assessment of the network performance would also be analyzed. Some

measurements to properly address the performance of the network are included in the last

portion of this chapter.

2.1 Evolution of Network Communications

Classical network communications, including the telegraph and telephone services,

are developed to transfer the information in text and voice. However, with the rapid

growth of semiconductor technologies, information processed by computers increase

exponentially. Modern network communication regarding to the computer data transfer,

is then developed on the basis of the classical network communication techniques. The

research objectives of this thesis are to implement a simplified and optimized network

design for the computer network communications. So the evolutions of the network

communication are important references for such purposes. Alberto Leon-Garcia and

Indra Widjaja [18] separate the network evolutions into three phases: message switching,

circuit switching, and packet switching. A brief overview for the classical

communications is mentioned here.

2.1.1 Message Switching

The most original network communication is implemented by the message

switching method. Long distance communication is relied on the messenger travels









through different locations by feet, animals, or other manners. The communication is

fairly slow and unreliable for the manual message switching. In 1837, the telegraph

service was first demonstrated by Samuel B. Morse as a practical communication of text

messages over long distance. The message was encoded by the Morse code (see Table 2.1)

with a combination of lines and dots. The transmission is made by sending electrical

current over a copper wire. The messages could therefore be transmitted almost

instantaneously from node to node. A well trained human operator can transmit 25 30

words/minute. Meanwhile, a message routing is still made by human decision according

to the destination of messages.

Table 2.1 Morse code.
A J- J --- S 2 -----
B K-.. K -.- T 3- 3 --
C L--- L --. U 4- 4 ....
D -.. M -V -- 5 .....
E N -" W 6-- 6..
F -.0 --- X 7- 7--
G --" P .--- Y 8-- 8--
H ... Q .--- Z-- 9
I R 1 0- -----

In order to increase the transmission rate for the telegraph, a technique called

multiplexing was then introduced. Multiplexing is the attempt to combine the transmitted

information over a single telegraph wire. It is also an initial attempt to share medium on

the electrical signal transmission. The Baudot system is developed and adopted in 1874

as the first multiplexing system. Baudot system used five binary symbols as a character.

The encoding system further evolves a modem alphanumeric expression ASCII code

(American Standard Code for Information Interchange). The table of ASCII code is

shown in Table 2.2.









The multiplexing can be realized by signal modulation. The modulated signal

carries different sinusoidal signals at different frequencies. The binary information could

be transmitted at a pair of frequencies, for example, f0 as "0" and f, as "1". So by using

different pairs of frequencies multiple signals could be transmitted simultaneously. This

technique is also known as Frequency Shift Keying (FSK) as a modern communication

modulation method.

Table 2.2 ASCII table.
0 1 2 3 4 5 6 7
0 NUL SOH STX ETX EOT ENQ ACK BEL
1 DLE DC1 DC2 DC3 DC4 NAK SYN ETB
2 SP !" # $ % &
3 0 1 2 3 4 5 6 7
4 A B C D E F G
5 P Q R S T U V W
6 a b c d e f g
7 p q r s t u v w
8 9 A B C D E F
0 BS HT LF VT FF CR SO SI
1 CAN EM SUB ESC FS GS RS US
2 ( + /
3 8 9 "< > ?
4 H I J K L M N 0
5 X Y Z [ \] A
6 h i j k 1 m n o
7 x y z { } DEL


Circuit Switching

The telegraph service


successfully solves the huge propagation delay of the


traditional communications. However, the manual routing of information still limits the

network communications because of the efficiency and reliability. In 1876 Alexander

Graham Bell developed a device to transmit the voice signal. A bidirectional, real-time

transmission of voice signal is called telephone service. The telephone service is an

analog transmission system compares to the telegraph service as a digital transmission


2.1.2









system. The telephone service is convenient and could be operated by the users with little

training. This important characteristic quickly leads to the exponential growth of the

telephone service.

Nevertheless, a problem was recognized very soon after the number of users grew.

The telephone service requires a dedicated line between two users. A network of the

telephone service with n users needs n(n 1)/ 2 dedicated lines, which makes the cost

significantly increases. Therefore, in 1878, the telephone switches was introduced to

reduce the number of dedicated lines by having operators manually establish the

connections based on user's demands. Figures 2.1 (a) and (b) show the differences

between the networks connections mentioned above.








S..... .. ..... .....

(-) (a) a(b)

Figure 2.1 Telephone network connections. (a) without circuit switching, and (b) with
circuit switching.

With the development of electrical switches, a hierarchical decimal telephone

numbering system was developed later for the dialing connection in the telephone

network for the compatibility of the large number of users. One of the important features

with the telephone network is called connection-oriented because a proper setup is

required before the information can be exchanged. This communication procedure is also

called circuit-switching. The connection can be separated into three phases within a

phone call. In the first phase a connection request sent by the user and proper set up needs









to be accomplished. The second phase is the actual transfer of voice from end to end.

Finally the third phase is the release for the connection. However, for the circuit

switching, the routing decision is made while establishing the services. So the routing

information is not needed after the connection has been set up in this situation.

2.1.3 Packet Switching

The message and circuit switching is frequently used by the classical

communication systems. However, they both have their deficiencies. The circuit

switching improves the efficiency of traditional manual message routing, but it is initially

designed for analog information transmission. The occupation of the dedicated lines will

also decrease the efficiency of transmission in a network. Meanwhile, the development of

computer technology after 1940s largely increases the capability of the information

process. So the need for a discrete data transfer in a more dynamic system is requested.

The modern communication network was developed on the basis of such a

communication manner.

The developments of computer networks were initiated for the military purposes.

The first wide geographical area network, Advanced Research Projects Agency

NETwork (ARPANET), was developed by Defense Advanced Research Project Agency

(DARPA). ARPANET is the ancestor of the Internet. A discussion on the history of

Internet can be found in [20]. A critical concept before the implementation of computer

networking is packet switching. It is the fundamental principle of the Internet. The packet

switching is somewhat similar to the message switching, however information

transmitted is cut into separate short segments. There are two types of packet switching:

virtual circuit and datagram. The virtual circuit requires the connection of two end nodes

to be set up before the actual transmission, and the route of the transmission is fixed










during the session. This can guarantee the frames received in order and less overhead is

required in the frame. The most common examples are the ATM and X.25. Nevertheless,

for datagram transmission, each packet is treated as an independent entity, and full

information for the recipient is included in the header of a frame. The frames received at

the end point might be out of sequence because of their various routes. The example for

datagram transmission is Transmission Control Protocol (TCP) and User Datagram

Protocol (UDP). The transmission methods for message switching, circuit switching, and

packet switching are shown in Figure 2.2.

A B C D











----_.--^ _ltfl~~
A B C I
Request








---- m Iected

(a) (b)














(c)

Figure 2.2 Network Switching. (a) message switching, (b) circuit Switching, and (c)
packet switching.









2.2 Layered Architecture

With the growing scale of the networks, the increasing need for switching message

leads to the further development of layered networks. Different functionalities are

regulated at each layer. The proper effort is made for the definitions and reorganization of

network services based on such an environment. The network architectures developed by

different vendors varied in the early designs. However, the compatibility between

different networks begin to attract more and more concerns. A regulated reference

network model can also help to bridge different network environments. This section will

discuss briefly on the developments of the layered networks. Two most frequently

mentioned models are introduced here: OSI model and TCP/IP structure. The similarities

and differences between these structures will be highlighted.

2.2.1 OSI Reference Model

The OSI reference model is developed to provide the needs discussed above. The

effort is made by International Organization for Standardization (ISO) to provide a

reference model for Open Systems Interconnection (OSI). OSI reference model regulates

the functions for networking to seven layers in a stack. Each layer only utilizes the

services in the lower layers and provides the higher-level features to the upper layers.

Figure 2.3 shows the relative position of each layer in the model stack and how each

layer can interact with the others.

In Figure 2.3, process in each layer on different machine communicates directly

with its counterpart by Protocol Data Unit (PDU). A header contains protocol

information and user information is encapsulated with the data provided by the upper

layer for each PDU. It is the dialog across the peer processes in the parallel position.

However, in order to support such a parallel communication, each lower layer will









received the entities called Service Data Unit (SDU) from the upper layer and

encapsulate the SDU with its header as the supporting information for the operation in

this layer. This is a vertical communication between layers.


Figure 2.3 OSI reference model.

The OSI reference model is a framework for any further developments on the

protocol set for different environments. The descriptions about the seven layers are listed

below:

* Physical Layers

The physical layer is responsible for the low level transmission of each bit of the

data. It includes all the electrical and mechanical hardware specifications for the









transmission of signals. The wire materials, connection interfaces, radio frequencies, and

signal modulations are all specified with in this layer.

* Data Link Layer

The data link layer provides the transfer of frames between two different nodes. It

includes the procedure to transfer the information by blocks, indicates the boundaries of

each frame. Usually the information encapsulated into the frame in the data link layer

includes the flow control and addressing information, as well as the cyclic redundancy

check bytes. The correction of transmission error within the frame is upon these

additional bytes. The error correction is especially important for a high transmission error

environment. The flow control and address provides the functions such as Medium

Access Control (MAC) so the share of medium could be handled.

* Network Layer

The transmission of information through one host to another host over one or more

networks will be handled in this layer. Usually a hierarchical addressing scheme is used

to transmit the information over the global network. Appropriate routing services are also

provided here in order to deliver the packets from the original network to the destination

network. The routing protocol is also responsible for the determination of the optimal

path for packet transmission in order to mitigate the congestion and obtain better

transmission efficiency. The error control is provided in this layer as well. Unlike the

address in data link layer, which is a hardware-based identification, the address for this

layer is a logical address. The network layer offers a broader coverage over global

network.

* Transport Layer









The transport layer is responsible for end-to-end transfer of data. The segmentation

and reassembly of data from upper layers is processed in this layer. The connection

between two ends is set up in this layer as well. Besides, the Quality of Service (QoS) is

also provided here based on different connection requirements in order to provide the

communication in a more reliable or cost effective mean.

* Session Layer

The session layer manages the manner for end user to exchange information. For

example, a half duplex dialog or full duplex operation can be assigned here. Other

functionalities like checkpointing, adjournment, termination and restart procedure are

also offered in this layer.

* Presentation Layer

Presentation layer is in charge of the encoding methods for data from the application

layer. The machine dependent codes can therefore be converted into machine

independent codes. For instance, an ASCII coded file can be converted into an EBCDIC

coded file. Different data encryption schemes are offered here as well.

* Application Layer

The codes and protocols used in this layer provide various communication services.

For example, File Transfer Protocol (FTP) provides the service for file transfer.

HyperText Transfer Protocol (HTTP) enables the access of World Wide Web (WWW)

documents. Other services offered in the application layer include virtual terminal

(TELNET), name management (DNS), and mail exchange (POP) (See Appendix A).










2.2.2 TCP/IP structure

OSI reference model has presented a well regulated framework of the layered

network structure. The model regulates all the functions and procedures of the network

communications. However, it is a quite conceptual network model. A set of protocols to

allow physical network communication under OSI reference model is still required.

TCP/IP structure is developed by DARPA research project to connect the networks from

different vendors to provide several fundamental services over a wide area network. It

later becomes a very successful network structure worldwide. In this section, a brief

introduction about TCP/IP structure is discussed as an example of network developments

under OSI reference model.

TCP/IP architecture usually contains 4 layers: application layer, transport layer,

internet layer, and network layer. The mapping between OSI model layers and TCP/IP

structure layers is in Figure 2.4.








Transport Layer

Network Layer

Data Linm Layer
Network Interface Layer
Physical Layer


Figure 2.4 Comparison of layer definition between OSI model and TCP/IP structure.

From the figure above, it could be found that the upper 3 layers in OSI model are

concluded in the application layer in TCP/IP structure. The transport layer usually









contains two major protocols: TCP and UDP. TCP provide a more reliable mean to

transfer packets. Also error recovery and flow control is allowed by TCP. UDP, however,

is a connectionless way to send packets. It is a less reliable but cost effective way to

transmit data. The Internet layer is the same as network layer in OSI model. It is mainly

responsible for packet routing. The network interface layer covers the physical data

transmission over various hardware interfaces. However, in TCP/IP structure, layers are

not as strictly defined as in OSI model so application layer is allowed to bypass the

intermediate layers and directly send data units to network interface layer.

As mentioned in OSI reference model. When the data units are passing to the lower

layers, each of them will be encapsulated with some extra information regarding to

functions in each layer. Appropriate operation could therefore be performed during the

delivery of packets from one host to another, and from one end to another end. For

example, as in Figure 2.5, the application data like HTTP or FTP request is first sent to

the transport layer. The header depending on the service established (TCP or UDP) will

be given in front of the application data unit. The encapsulated data unit would then be

forwarded to the Internet layer, an Internet Protocol (IP) header will again be attached

with the data unit in order to assign destination of the data unit. Before the frame is

actually transmitted, the Ethernet header and error checking code will be combined at the

front and end therefore the frame could be delivered to the next node correctly. An

inverse process will be performed at each of the intermediate node to deliver the

information correctly to its destination. Appropriate packet rearrangement and request of

retransmission of the lost or error packets will also be handled in order to guarantee the

correctness of the data transfer. The documents regulates the formats of TCP, UDP, and









IP headers could be found in [19], [21], and [22]. Other protocols regarding to different

services like Real Time Protocol (RTP) could also be found in other relevant RFC

documents. Moreover, except for the network protocols developed by DARPA, more

network protocols have been developed by various vendors. Table 2.3 lists some of

frequently used network protocols for each of the OSI layer.

Table 2.3 List of network protocols.
Layer Protocols
Application Layer HTTP, FTP, SMTP, RTSP, POP, TELNET
Presentation Layer SMB, XDR
Session Layer SSH, NetBIOS, ASP
Transport Layer TCP, UDP, RTP, ATP
Network Layer IP, IPv6,DHCP, ICMP, X.25
Data Link Layer ARP, RARP, DCAP, IEEE802.11
Physical Layer T1, encoding methods, signal modulation/demodulation


Figure 2.5 Encapsulation of header and error check code into data units.


\









2.3 Wireless Communications and Issues

The early developments of the network communications are mainly developed for

the wire-based environments. Not only because the wire communication is more friendly

for signal transmission but also much simpler and secure. However, the wires

significantly restrict the mobility of the network communications and applications for

communication. With the advances in computer technology, the need to use the electrical

mobile devices increases drastically. Mobile devices like laptops, PDAs and cellular

phones, as well as the mobile robots and satellites, are required to communicate with each

other wirelessly. Also, various industrial and military applications have immense needs to

send information without using wires. Therefore, the wireless communication has

become a big step in network communications.

The media used for wireless communications are typically radio, or sometimes

optical signals. By using the modulation/demodulation methods on these signals, binary

data could be transmitted over the air. The wireless communication could hence be

implemented. Morse code has also been used extensively for manual operation using

radio transceiver in early times. It played an important role for military communications

during World War I & II. However, although both wired and wireless communications

need to modulate and demodulate the signals, there are still some considerable

fundamental differences between them for reasons mentioned in the following:

* Wireless signal is susceptible to noises and Electro-Magnetic Interferences (EMI). A

higher error rate is expected in wireless communications.









* Wireless signal strength varies greatly from both different positions and time instants

because of the EMI and the multi-path propagation. Transmission collision is hence

difficult to be detected under wireless environment.

* The spectrum of wireless signal is restricted comparing to the wired signal. So

communication bandwidth and rate are limited and slower.

The differences between wired and wireless network communications are mainly in

physical layer and data link layer in OSI seven-layer model (or network interface layer in

TCP/IP structure). The upper layer protocols are mostly compatible in both network

environments. In the following sections, some discussions will be made about the specific

issues related to the wireless network communication. It also plays a very important role

in the wireless network design for multi-robot testbed later.

2.3.1 Medium Access Control Protocol

One and maybe the most significant influence of the wireless communication is the

medium sharing technique. When two or more devices need to share the same medium

for communication, a proper procedure needs to be followed in order to transmit and

receive the data correctly in each session. The medium sharing techniques are defined in

the network layer in OSI model as MAC protocols. The wired network communications

uses collision detection to avoid the collision of transmission between multiple machines.

However, since the transmission of wireless signal is susceptible to EMI and multi-path

propagation, collision detection becomes impractical to implement. Therefore, the

collision of packet transmission needs to be avoided in advance of the actual data frame.

This technique is called collision avoidance.









A widely used medium sharing protocol for wired network communication is

Carrier Sense Multiple Access with Collision Detection (CSMA-CD). As shown in

Figure 2.6, when t = 0, host A starts to send a frame to host B while it detects no

activities over the medium. After a short period of time t,, host B also transmit a frame to

host A. The transmission frames from both nodes collide at t = t2. Because of the

abnormal voltage from the frame collision hasn't reach host B, the medium status for host

B is still available. After the propagation time tpop, the transmitted frame from A reaches

host B therefore collision has been detected by host B. Also, when t = 2top t, another

transmitted frame also reaches host A so both hosts can detect that collision has occurred.

After the collision has been detected, appropriate strategies will be applied in order for

frame retransmission. Either the persistent scanning for the availability of the

communication channel or the back-off algorithm for rescheduling a retransmission will

be performed.

However, the detection of the signal abnormality is not practical for wireless

signals due to the significantly variant in magnitude and low Signal to Noise Ratio (SNR).

Moreover, the radio transmission has the uneven signal coverage issue. It is called hidden

terminal problem as shown in Figure 2.7. Both of node A and C have only the limited

signal coverage to the intermediate node B. Therefore, the transmission from either node

A or C can't be detected by each other hence the collision can't be avoided by the

CSMA-CD method.







44


t = 0


AE B













"'*- -
-

A B













Figure 2.6 CSMA-CD.







A A- C
/ ,
/ / \ \
/ /
/ / \


\* A *

\ / /
\ /
\^/ /




Figure 2.7 Hidden terminal problem for wireless network.

Therefore, instead of sensing the collision of transmitted frame, a more

conservative strategy must be applied for an effective medium sharing. In IEEE802.11, a

modification is made for CSMA-CD. In order to prevent the data frame collision, a short










signal for flow control is sent in advance. When the collision happened to such a request,

no response will be answered for the transmission request therefore the collision for data

frame could be avoided. This technique is called Carrier Sense Multiple Access with

Collision Avoidance (CSMA-CA). It is shown in Figure 2.8.

Data frame transmitted


sent sent

Collision occur





ACK CTS RTS
sent sent sent




RTS CTS ACK Data Frame

Channel busy
Channel idle

Figure 2.8 CSMA-CA.

Figure 2.8 shows the occurrence of the collision and how data frame is transmitted.

In the beginning both channels attempt to transmit Request-To-Send (RTS) signal. The

collision happens and therefore both hosts A and B will not response to the request. A

back-off algorithm is applied on both hosts. After the waiting for a shorter back-off

period, host A attempts to re-transmit the RTS frame and seize the control of the channel.

Host B responses to host A with a frame Clear-To-Send (CTS) after RTS frame is

received. The data frame is hence sent from host A to host B after RTS is recognized.

With the time elapse during the data transmission, all of the other host will remain silent









so the period is collision free. The channel will become available again after the

transmitted data frame is confirmed by sending the acknowledge frame ACK from host B.

2.3.2 Ad hoc and Infrastructure Topology

Another important aspect for wireless networks is the network topology. The

wireless network has great mobility comparing to the wired network. Therefore it has

been vastly used on mobile devices. As a result, the dynamic reallocation of the network

nodes leads to the requirement for more dynamic connections in the network. Traditional

network topologies for wired network communication are somewhat restrictive in such an

environment. In this section, the network interconnections and developments for the

dynamic topologies will be briefly introduced.

The interconnections between network nodes in a network could be divided into

two categories: physical connection and logical connection. Physical connections means

two nodes are locally connected to each other by physical hardware such as Network

Interface Card (NIC), cables, switch or router. It is a hardware based connection. Logical

connection means two network nodes are linked to each other by various network

protocols as they are in the same network (LAN). It is a software oriented manner. Two

or more nodes at different geological locations could logically be located in the same

network. Protocol such as Virtual Private Network (VPN) is an example to provide a

logical connection. Various wired network topologies could be found in Figure 2.9. The

links could be either physical or logical.















*(a) (b)






(c) (d)








Figure 2.9 Network topologies. (a) mesh topology, (b) star, (c) bus topology, (d) ring
topology, and (e) tree topology.

The traditional network topologies are basically the centralized architectures. All

the packets usually won't be transferred directly between two nodes. The transmission is

centralized so all the packets will be sent to a leading node as the network hub first.

However, for the wireless mobile environment, the communications sometimes need to

be performed in a decentralized mode when the global signal coverage for the hub is not

possible. Instead of using any fixed network topology as either one in Figure 2.9, the

network topology is required to be more dynamic. Therefore, the network topology ad

hoc mode has been used in the wireless mobile communication environment. In ad hoc

mode, transmission could be performed between different nodes directly without using an

intermediate access point. It is also known as peer-to-peer network. Moreover, this peer-

to-peer network topology is mostly used for mobile wireless network. Therefore,









sometimes it is known as the Mobile Ad hoc NETwork (MANET). The official charter

for MANET could be found in [23]. The MANET is usually implemented under the

environment with following characteristics:

The network topologies could change rapidly and unpredictably.

The network is decentralized and lack of prefix allocation agency.

The devices running in MANET are usually power limited and equipped with

lower processing capability.

The MANET provides the network some advantages such as the scalability, mobility, and

dynamic communication environment. It could solve the issues when the reallocation for

network nodes occurred frequently. However, the MANET is still under development.

Only a few experimental network standards have been proposed for MANET. These

standards mostly provide packet routing services. Some reference documents for

experimental packet routing protocols can be found in [24] [27]. There are still a

number of technical challenges that needs to be overcome such as addressing, packet

routing, network security, and QoS for MANET. Also, characteristics for different types

of network links such as standalone ad hoc network or hybrid ad hoc network could vary

greatly. Nevertheless, MANET could provide a quite attractive communication solution

for the wireless multi robot systems. In addition, the communication network proposed in

this thesis is a simple experimental network for a semi-MANET communication

environment.

2.4 Communication Performance

Another important aspect in order to understand the network protocols is the

analysis for the communication performance. Performing such an analysis could provide









a better understanding for the network capability and reliability. Any revision for the

existing protocols in order to improve the network performance could also be verified

theoretically due to such analysis. In this section, some of the fundamental network

measurements will be briefly discussed. More terminologies about some concepts of the

network performance factors could be found in [28].

2.4.1 Bandwidth

For the communication terminologies, this term is obscure. It has two different

meanings. The first definition is the width of the band of frequency used to carry data.

This will usually affect the communication rate since the carrier with higher frequency

has larger density for data. For example, the bandpass communication is usually faster

than the baseband communication. However, the concept addressed here is the second

definition, which is the communication rate for data transmission over physical carrier.

This is determined in the physical layer. Different modulation/demodulation techniques,

hardware specifications and signal carriers used decide the maximum capability of the

transmission. For example, the maximum bandwidth for the traditional dialup network

using MODEM over the phone line is usually 65536bps. The bandwidth for IEEE802.3u

is 100Mbps. The bandwidth for IEEE802.3ae is 10Gbps. The protocols adopted in the

upper layer will not have any influence on the communication bandwidth.

2.4.2 Transmission Loss

The transmission loss will deteriorate the actual transmission rate and quality. It

occurred for various reasons from hardware based factors to protocol based factors. For

instance, a low SNR environment could lead to a higher transmission error. This usually

results from either longer transmission distance between relay nodes or low quality

transmission medium (higher EMI or worse cable quality). Also, the protocols adopted in









the network design could also have impact on the transmission error. The more bits

contained in each individual frame, the more possibility for a transmission error could

occurred within a frame. Too much routing could also make the transmitted packet to be

obsolete. Two different circumstances will happen as the transmission loss: frame loss or

frame error. When a frame is lost during the transmission, no response for the frame

acknowledgement will be received therefore it could be detected. On the other hand, the

frame being received incorrectly could be found by the Cyclic Redundancy Check (CRC)

code attached at the end of each frame or packet. When either of the communication

abnormalities happened, an appropriate arrangement for re-transmission will be made so

the problem could be corrected. The protocol used for frame re-transmission is called

Automatic Repeat reQuest (ARQ) protocol. Three different types of ARQ protocols are

used for frame re-transmission: stop and wait ARQ, go-back-n ARQ, and selective repeat

ARQ. Different efficiencies and computational requirements are needed for different

ARQ protocols. The details for ARQ protocols used will be further discussed in chapter 4.

However, the efforts to guarantee the data is being transmitted and received

correctly usually need more overhead, which cause a lower effective transmission rate.

Some tradeoff must be made between the quality and quantity for communications.

2.4.3 Throughput

The communication bandwidth couldn't present the actual communication rate for

user data. The transmission loss, repeat request, and the wait for establishing the

connection could all reduce the rate. Therefore, the network throughput makes more

sense to present the real communication performance. According to the definition in [28],

the network throughput is defined as "the maximum rate at which none of the offered

frames are dropped by the device". The throughput of a network is mainly determined at









lower layers (physical layer, data link layer, and network layer). The factors which have

influences on the throughput include ARQ protocols, MAC protocols, network

congestion and packet routing protocols.

2.4.4 Latency

One of the very important issues for network performance is the latency. It is

defined [28] as "the time interval starting when the last bit of the input frame reaches the

input port and ending when the first bit of the output frame is seen on the output port".

The network latency also represent the time it takes from the time data is requested to the

time it is arrived. Sometimes, it is also used to define how much control we have for the

network. Not only the latency but also the jittering, which is the variation of the latency,

are important when we want to measure the performance. It is a fairly critical issue for

the distributed control problem or multi-agent systems. One of the methods to mitigate

such problem is QoS, which normally reserve certain amount of bandwidth for specific

services. Also, a real-time system to guarantee a minimum delay of system operation for

MRS is another solution.

In the other hand, the communication performance is also highly dependent on the

hardware specifications. The hardware design is another critical point in this thesis.

Different hardware architecture has influential impact to the performance on not only

communication but also processing and maneuvers capabilities. The next chapter will

therefore focus on the hardware specifications and their restrictions.














CHAPTER 3
WIRELESS MULTI-ROBOT TESTBED

As the discussion in the previous chapter, it could be found that the development of

the network communications is based on the hardware specifications and mission

objectives. Protocols used in the lower layers are highly dependent on the physical signal

types and modulation/demodulation techniques adopted by the hardware. Therefore it has

become fairly important to understand the physical hardware used for the experimental

testbed before the discussion of the dedicated network environment.

Meanwhile, the understanding for the concept of system operation is also critical

before designing the network. To appropriately identify the requirements for the system

operation is the essential to support the operation of the testbed in a cost effective way.

The design of the network protocols for upper layers is mainly determined by the mission

requirements. The services provided by the network and their performances are the key

factors which lead to a satisfactory network development. Therefore, all the hardware

facilities and various sub-systems, for example, positioning system and communication

system, will be mentioned in this chapter before the discussion of network protocols.

3.1 Hardware Architecture of the tested

The wireless multi-robot testbed is composed of several sub-components including

mobile robots, positioning system, the operational area which provide the region for its

operation. The hardware architecture of the testbed is shown in Figure 3.1. The mobile

robot is controlled by the independent controller on each of them. The positioning system

determines 3 dimensional positions of the LEDs and orientations of the LED-mounted







53


objects by using eight cameras mounted as shown in the figure. A central computer is

responsible to provide information for localizing all the obstacles and moving objects via

the communication system. Also any user inputs to command the system can also be

transmitted though the central computer. The detail description on each sub-system will

be provided in the rest sections of this chapter.


rUser Input



IEEE & I IlI.


Systems_; !_----I'Positioning
Communication I System
Systems I


I I


I I






II
I EEE2 1b I IEE 211bll




IR s-nwr -R


WALKER -WALKER







I I

I I

I Operational Area

Figure 3.1 Hardware architecture of the testbed.









3.2 Wireless Mobile Robot

The core components for the testbed are the mobile robots, which are considered as

intellectual (and autonomous) agents in the system. Due to the physical application

requirements, the robots in the testbed must have the following features:

* Mobility

A mobile robot system can be used to validate the concepts and algorithms for path

planning, vision servo control, leader-follower problem. It is also widely applicable for

many autonomous vehicle control applications. So the appropriate maneuvers to move

the robot itself are required.

* Wireless communication

The control of the robot needs the communications. Hence an adequate transceiver

to send/receive control commands, transmit all the required information back and forth

between different vehicles is needed. However, for a mobile robot, wires could

enormously restrict the mobility. Therefore, the wireless communication is the essentials

to remove this constraint.

* Decentralized control

A multi-agent system needs multiple processing threads/controllers to control

different physical agents. Under some restricted circumstances, a centralized control of

multiple robots is not feasible. Therefore, the design of the testbed requires a controller or

processing facility on each of the individual robot.

* Scalability

Mobile robot can dynamically change its location, therefore, the fluctuation of the

number of robots during the operation is assumed. For many multi-agent systems, the









scalability is considered during the operation. So the functional robots in the testbed

could be added or removed any time in order to keep the flexibility of the operation.

The robot WALKER is developed here for the MAS research based on the

considerations discussed above. The WALKER acronym stands for Wireless

Autonomous Linux-based Kinematic Expert Robot. It is shown in Figure 3.2.






















Figure 3.2 WALKER for multi-robot testbed.

The dimension for the WALKER is 8.75 "x 6"x 7" (L x W x H). It is a modulized

design comprising various modules so the maintenance could be easier. Also it could be

more flexible to any further development since the upgrade of the hardware can be done

by simply replacing individual modules. The WALKER has four different modules to

support its operation: power module, processing unit, communication module, and

hardware control unit. The inter-connections between each module is shown in Figure 3.3.

The hardware used for each module will be mentioned in the following subsections.

























Figure 3.3 Block diagram for modules on WALKER.

3.2.1 Power Module

The power module is composed of a lithium-polymer rechargeable battery and the

DC/DC power supply for PC/104. The output voltage for the lithium-polymer battery is

14.8V, the power capacity is 4400mAh. The battery connects to the power supply and

converts the power to the regulated 12V and 5V outputs to support the requisite operation

on the robot. The pictures for the power module are in Figure 3.4. The power module

could support the system operation for roughly 8 hours when the system is idle under

Linux.


(a) (b)

Figure 3.4 Hardware pictures for power module. (a) lithium-polymer rechargeable battery,
and (b) PC 104 DC power supply.









3.2.2 Communication Module

There are two different channels for the wireless communications. The differences

for two channels are listed in Table 3.1. However, this thesis only discusses for the low

bandwidth communication channel. The reason for this choice will be mentioned in the

next chapter. Pictures for the hardware are shown in Figure 3.5.

Table 3.1 Comparison of different communication channels.
Wireless modem IEEE802.1 lb PCMCIA card
Modulation method FSK BPSK/QPSK/CCK
Radio Frequency 900 MHz 2.4 GHz
Bandwidth 38400 bps 11 Mbps
Signal range
Signal range 300' / 1000' 300' / 1000'
(outdoor/indoor)
Power
powr 0.225 Watt 1.235 Watt
Consumption


(a) (b)

Figure 3.5 Hardware pictures for communication module. (a) Low bandwidth channel
(wireless modem), and (b) High bandwidth channel (IEEE802.1 lb PCMCIA
card).

3.2.3 Hardware Control Unit

The hardware control for the WALKER contains the controller for motors, input

D/A (Digital to Analog) and output A/D channels for various sensors. It is the

intermediate interface between processing unit and the physical hardware. The hardware

control unit uses a Motorola 68HC11 based board with 32 K SRAM as the controller.

The detail reference for the board can be found in [29]. Also, the WALKER currently

uses two servo motors to control its motion and a pair of infrared sensors for obstacle









avoidance. The servo motor is controller by Pulse Width Modulation (PWM) signals, and

the IR sensors needs to be converted by Analog to Digital Converter (ADC). However,

more optional facilities like camera and power management system could also be

controlled by this module. The pictures for the hardware control unit can be found in

Figure 3.6.










(a) il,: V (b)















(c)

Figure 3.6 Hardware and hardware control unit. (a) servo motor, (b) infrared sensor, and
(c) MC68HC11 controller.

3.2.4 Processing Unit

The processing unit is the kernel of the robot. It greatly enhances the capabilities

and features for the robot. The processing unit turns the robot into an intellectual and

autonomous agent instead of simply a mobile robot. The processing unit is responsible









for performing the required computations for control actions and the cooperative

behaviors, as well as the requisite network communication with the other agents.

The processing unit uses PC/104 board for the robot. The specifications for the

processing unit are in Table 3.2. Figure 3.7 shows the picture of PC/104 module.

Table 3.2 Specifications for processing unit.
Specifications
Manufacturer / module number Kontron / MOPlcd6
Processor Pentium MMX 266MHz
Memory 64Mbytes
Disk 512 Mbytes CF card
Operating System White Dwarf Linux 2.0


Figure 3.7 PC/104 processing unit.

3.3 Positioning System

The localizations of the mobile robots and all the possible obstacles for many of the

multi-robot testbeds are fairly computationally expensive. The onboard camera with the

simple image processing technique like feature points extraction is usually required with

the sharing of the information between each robots for localization. However, for the

wireless multi-robot testbed discussed in this thesis, a computationally cost effective









positioning system is used for the localization. The positions for the objects are provided

in the system by the PhaseSpace position measurement system. Eight cameras are

mounted on the outer aluminum frame in order to cover the whole operational area. The

LEDs are required to be attached on the targets so that the cameras can track the positions

and orientations of the LEDs and hence the targets. The block diagram of the positioning

system can be found in Figure 3.8.


Linux PC USB Hub ] Port expander






Camera 1 Camera Camera Camera Camera Camera Camera 7 Camera






LED Board LED Board LED Board






Figure 3.8 Block diagram of the PhaseSpace positioning system.

In Figure 3.8, the cameras are connected to the hub by an USB connection. The PC

is running under Linux operating system. Also, the hub has the connections to the LEDs

via proper adaptor interfaces. Different LEDs could be recognized by the system by

sending the flashing signals at specific frequencies so each LED could be identified with

its specific number. The positioning system could generate the 3-D data for all the

tracking points at a rate of 220 fps. It is, however, enough for detecting the robot motions.









The pictures of positioning system are shown in Figure 3.9. The detail discussion about

the determination of 3-D position of LED under the PhaseSpace system could be found in

[15].





















c) (d)

Figure 3.9 Pictures of hardware for PhaseSpace positioning system. (a) hub, (b) camera,
(c) LED board, and (d) LEDs.

3.4 Operational Area

By using the positioning system, the objects could be properly located within

certain range. However, due to the physical restrictions of the positioning system and

available room, the operation can only valid within a limited area. Therefore, in order to

enhance the operation of multiple robots and ensure the vehicle operations are valid, a

platform is made for restraining the region of the motions and placing the obstacles for

different research scenarios. The floor and several blocks used for the platform and

obstacles are made of foam. The figures in Figure 3.10 include the geometry of the

testbed for the platform, obstacles, and the positioning system.





















S(a) (b)

Figure 3.10 Geometry of the testbed. (a) positions of the operational area, and (b)
projections of the visible range for the positioning system.

From the description in Figure 3.1 above, it could be observed that a

communication network is required to support the operation of the wireless MRS testbed

by using the communication module mentioned in section 3.2.2. The next chapter will

describe a dedicated protocol suite using the low bandwidth channel to enable the inter-

cooperation between various robots as different individual agents.














CHAPTER 4
PROTOCOL SUITE

This chapter describes the protocol design for the self-configurable network used

for the wireless multi-robot testbed. The network interfaces and protocol suite mentioned

in this chapter are the communication implementations on the experimental multi-robot

testbed mentioned in the previous chapter. A complete network environment is outlined

in this chapter. Some considerations and hardware specific issues are addressed in section

4.1. In order to simplify the communication scenarios and problems, only LAN is

considered here. Section 4.2 introduces a dedicated LAN for the testbed. A few of the

protocols have been specified for this layered network to perform the operation of the

network communications. This will be discussed in detail in section 4.3.

4.1 Limitations and Requirements

Chapter 3 has describes the equipment used in the wireless multi-robot testbed.

Before the attention is focused on the design of network environment, explicit statements

of the network restriction are provided.

The most important principle for the network design for the testbed in this thesis is

the minimal realization for a required network communication with better hardware

compatibility. This means the optimization of the use of the onboard resources including

the power and computational needs, for example, memory and minimal processor

required. In other words, to support the system operation in a more efficient manner in

terms of the power and computational uses is the highest design principal.









The first step for the design problem is the understanding of all the constraints and

requirements. The limitations of the wireless multi-robot testbed are hence listed below:

1. Power consumption on the mobile robot needs to be minimized. So a

transceiver with lower power consumption used on the robot is preferred.

2. Lower bandwidth channel (wireless modem) is the preferred device for the

high level data transmission such as a control command because the lower

required bandwidth, lower power consumption and better signal sensibility

( and reliability).

3. Higher bandwidth channel (PCMCIA wireless Ethernet card) is the

preferred device for a low level data transmission such as the image data.

A mixed use of both low and high bandwidth channels can make the

communication to be more robust and efficient. However, the transmission

of low level data is not in the scope of this thesis.

The requirements to support the system operation are specified below:

1. Command refresh rate for mobile robot control could be less 5 Hz since the

transient response for a mechanical mechanism is relatively slow.

Therefore the data rate of 38400 bps is enough for robot control

2. The operation requires the inter-operation between agents. Therefore the

application layer Agent Communication Language (ACL) is also needed to

be defined for the network communication to handle the conversation in

lieu of simply low level data transmission.









3. The scale of the fleet in the multi-robot testbed is restrictive. Hence in

order to simplify the network, only a LAN is discussed here. The packet

routing protocol is not under consideration in this thesis.

4. The network topology is comprised of one leader and multiple followers.

5. In order to enhance the autonomous operation and decentralized control, an

automatic configuration of the topology is required by the system so the

robot doesn't definitely need the command offered from the leader to make

such a change. This procedure will be covered in the next chapter.

4.2 Local Area Network Architecture

A very important assumption made here is the scale of network. As the most

frequently used topology, the tree topology needs a hierarchical addressing technique and

ability to forward the information to its destination. This also includes sophisticated

algorithms and protocols for data routing. Therefore, a significant simplification could be

made in the design problem by confining the network to only LAN. Moreover, some of

the intermediate layers could also be skipped since only a single LAN has been

considered here. The difference of the LAN and Wide Area Network (WAN) can be

found in Figure 4.1.



LAN 1 ,I- -

--- ---



'* ,.. .

(a) (b)

Figure 4.1 Comparison of the interconnections of different networks (a) LAN, and (b)
WAN.









As it is shown in Figure 4.1(a), all of the nodes in LAN 1 could be accessed using

the local addresses in the LAN. For example, the local address for node A and node B in

Figure 4. l(a) are x and y respectively. The two nodes could therefore reach each other

by using their local addresses within the same network. The information routing under

such case is not going to be required except in the infrastructure mode. The frame

transmission from node A to B will go though the intermediate node C as the local

network hub under infrastructure mode. Otherwise, no forwarding of information will be

required in ad hoc mode under local network.

However, as shown in Figure 4. l(b), the addresses within each LAN are only valid

for the other nodes in the same LAN. A direct connection can't be made across different

LANs. The only exception is the specific bridges used for interconnection between

different LANs. Therefore, in order to have the access for each node in the WAN, a

global address needs to be given to each node. Usually, a hierarchical addressing method

is considered in this case to improve the scalability of network. For example, a local

address for node A in LAN 1 is x and the address for node B in LAN 2 is y. However,

since node A and B are not in the same network, they can't reach each other by using

their local addresses. The numbering on each of the networks is required to address a

global identification. Therefore, as the topology in Figure 4.1(b), a given global address

for the node A is 1.x and 2.y for node B. When the transmission occurred between these

two nodes, the information will first be transmitted from node A to the local network

leader B, then the received packet will be forward to the other network leader node D in

LAN2, another information routing will happened then from node D to node B in LAN 2.

As it could be observed here, the addressing is used not only for the medium sharing









purpose but also for the information routing. Nevertheless, as it has already been

discussed in chapter 2, the data link layer and network layer (in ISO model) are used to

handle the problems.

Another critical packet routing issue also happened when signal can't globally

cover the physical range of a network. When the communication needs to be made from

one node to another node in the network between two far ends, the determination of the

forwarding route becomes sophisticated, especially when the network is under ad hoc

mode or the hybrid of ad hoc and infrastructure mode. Although the routing protocols for

wired network environment are already well developed, the networking in a mixed

network environment of different communication modes is still somewhat limited. The

proper routing protocol used highly depends on the type of the network environment.

The wireless multi-robot testbed discussed in the previous chapter is comprised of

several mobile robots. The number of robots is however limited. Since the objective for

the network communication is only for the collaborations between a limited numbers of

small robots, a more cost effective manner to implement the network communication

here will be to consider the environment to be only a single LAN. lin the next section, the

network and transport layers in the dedicated network are ignored under such an

assumption. As a result, both the computational cost and complexity of the network can

be significantly reduced.

4.3 Protocol Specifications

The above discussion provides an explicit explanation of the network

communication considerations and infrastructure. The detail specifications of the

dedicated wireless network are described in this section.









First, the network layers of the wireless communication are shown in Figure 4.2. In

Figure 4.2, the protocols include the pre-programmed protocols on the wireless modem

and specified protocols. The pre-programmed protocols are hard coded on the wireless

modem and the header for the raw transmission frame cannot be accessed. The pre-

programmed protocols covers two ISO layers by using Modbus protocol [30] as the data

link layer protocols, and a specific protocol [31] for network layer. The network layer

protocol using vendor identification number, channel, destination address, and address

mask to provide the networking features and filtration. However, the hard coded network

layer protocol is basically a Point-to-Point Protocol (PPP) which is designed for a

centralized basis communication scenario. As the decentralized communication need, the

provided protocols cannot satisfy the requirements for the operation. Moreover, the

header for each frame cannot be captured or modified in a software manner. Hard coding

the protocols into the firmware on wireless modem provides great convenience for the

end user to control various devices using serial port communication. However, the lack of

the flexibility with such design method significantly restricts the possibilities for further

network programming and control. Therefore, in order to obtain the full control on the

robot network communications, the pre-programmed protocols are ignored, and the

additional protocols are specified in the thesis instead.

The specified network communication structure contains two layers, data link layer

and application layer. As discussed in the chapter 2 for the layering in TCP/IP structure,

it is feasible to skip over the intermediate layers and have the network services directly

encapsulate the application layer SDU into the data link layer frame. Hence, due to the









consideration mentioned in previous section, the global addressing is ignored and the data

link layer could have the direct access to the application layer.


Fields can't be accessed at DTE

Figure 4.2 Dedicated wireless network layers

4.3.1 Data Link Layer Protocol

The specified data link SDU contains three fields: source address, destination

address, control information, and its payload. The purpose of this layer and the frame

header is for the medium access and flow control purposes. The design of this frame is

referenced from High-level Data Link Control (HDLC). However, the HDLC is also a

PPP. Therefore the destination address is added to the frame for the wireless network









medium access. The detail explanations for the meaning of each field are defined as

below:

* Source address

It is the address or ID for the frame transmitter. This is an 8 bits long field. 255

devices can be addressed.

* Destination address

It is the address or ID for the frame receiver. This is an 8 bits long field. 255

devices can be addressed.

* Control

The control field is responsible for establishing or releasing the connection for

data transmission, as well as the functions of the frame retransmission and

acknowledgement for the reception of previously transmitted frame. The ARQ

protocol is also implemented by the control field. It is an 8 bits long field. The format

of this frame is defined in Figure 4.3. The control field used here is similar to the

HDLC procedure [32]. Three types of frame are used here: information frame,

supervisory frame, and unnumbered frame. Information frame are used for the

transmission of datagram from upper layer. N(S) represents the send sequence

number for the frame, and N(R) represents the response sequence number. The bit

P /F indicates the direction for the data transmission. The supervisory frame is for

the flow and error control. The acknowledgement (ACK) frame and negative

acknowledgement (NAK) frames could be used to confirm the correctness of the

transmission. It comes with only response sequence number at the end since no data

is contained in this frame. The unnumbered frame is used to establish or release the










connection. Three different modes of connection could be established by unnumbered

frame: Normal Response Mode (NRM), Asynchronous Balanced Mode (ABM), and

Asynchronous Response Mode (ARM).

Information frame

0 N(S) PIF N(R)


Supervisory frame

1 0 S P/F N(R)


Unnumbered frame

1 1 U U PIF U U U


Figure 4.3 Bit-wise format of the control field

4.3.1.1 Link management

The normal response mode is the only mode used here for the network

communication. It is a synchronous transmission mode. So the secondary node can only

transmit when it is instructed by the primary mode. It is used under the half duplex

environment. The transmission mode is less efficient. However, it is computational

inexpensive since less buffer is needed and no data rearrangements needs to be made.

The transmission procedure for NRM is in Figure 4.4.

In Figure 4.4, the first two entities in the square bracket are source and destination

address. The third entity is the type of the frame. I is the information frame. The fourth

and fifth entities are N(S) and N(R). The acknowledgement of the received frame can


be piggybacked with the transmitted frame sequence number.

Also, when the transmission error occurred during the transmission, no

acknowledgement will be sent. The error will be detected if no acknowledgement is







72


received for the transmission for a period. The error control mechanisms in the wireless

network will be described in the following sub-sections.

Node A Node B
SNRM


UA
[A,B,1,0,0]

[B,A,I,1,0]
[A,B,I,1,2]
[B,A,I,3,2]
[A,B,I,4,3]
[B,A,I,5,4]
[A,B,I,6,5]
[B,A,I,7,6]


DISC


UA




Figure 4.4 Normal response mode

4.3.1.2 Forward error correction

Error control is important to guarantee the correctness of signal transmission.

Usually in a communication channel, two error control methods will be used in order to

improve the reliability of transmission: Forward Error Correction (FEC) and feedback

error correction (ARQ). The FEC is, by using redundant information attached in the

transmitted frame, the transmission error can be detected and corrected. However, the

FEC can't guarantee the correctness of transmissions.

In the wireless network environment discussed here, the FEC is automatically

provided by the CRC field in pre-programmed Modbus protocol. No further FEC is made










in the protocol described in section 4.3.1. However, the FEC provided by the firmware

will only drop the received frame once error is detected. No negative acknowledgement

will be sent in the network channel. Therefore, a feedback error correction is also offered

in this network.

4.3.1.3 Feedback error correction

The feedback error correction in data link layer is called ARQ. As described in

chapter 2, it is a mechanism to improve the correctness of transmissions. For the ARQ

protocols, there are three frequently used types: stop-and-wait ARQ, go-back-n ARQ, and

selective reject ARQ. For the latter two ARQ methods, the transmission efficiency

comparing to stop-and-wait ARQ are better. However, the cost is their relatively

expensive computational needs since more buffers and data rearrangements are required.

So the stop-and-wait ARQ is selected for the dedicated network here. The ARQ methods

are shown found in Figure 4.5.

NOde A
Node A (a) (





Node B (a) 4 ,,, (b)


< U
< < (c)

Figure 4.5 ARQ methods. (a) stop-and-wait ARQ, (b) go-back-n ARQ, and (c) selective
reject ARQ









4.3.2 Agent Communication Language

Another appeared problem after two agents could have communicated with each

other: How could the agent interpret the binary data into meaningful context? A common

language for two homogenous/ heterogeneous agents is required in order to translate low

level data into useful high level information or control command for different agents. The

syntactic and semantic definitions for such a language hence have to be regulated. This

task is defined in the presentation layer and application layer in ISO model. On the other

hand, since the dedicated network environment has skipped over some intermediate

layers, the compensation for some of these functions could have been made in the

application layer. The Agent Communication Language (ACL) has therefore become

another critical protocol for the multi-agent environments.

A few of the agent communication languages have been developed to satisfy the

interoperability need between agents. It includes both semantic and syntactic definitions

for the language. Two major ACLs have been developed and widely discussed:

Knowledge Query and Manipulation Language (KQML) [33]-[34] and FIPA-ACL [35].

Usually the ACLs are comprised of different performatives with various lengths of the

arguments and contents, depends on their requested actions. The information contained in

each ACL is often encoded in the ASCII format, while the binary message encoding

method is also used once in a while. However, the introduction of the ACL is quite

lengthy and beyond the scope of this thesis. Therefore, a detail introduction for the

developments and principles of ACL will be skipped over. The proposals and

specifications for various ACLs could be found in [33] [35]. This section will only

focus the attention on the dedicated ACL proposal.









The ACL proposed here is composed of a couple of different components. Table

4.1 lists the essential components as the syntax of the language used in this thesis.

Comparing to the other generic ACLs, this ACL proposal is restricted to a specific use for

some fundamental operations for the multi-robot testbed mentioned in the previous

chapter. For example: the query for the IR sensor reading, sending a command to request

a motion on another robot, or monitoring the eligibility of a robot in the network. The

detail semantics of the language are explained below:

Table 4.1 Proposed agent communication language
arguments Bits value semantics meanings
length 8 0x01 indicate the length of the ACL
length 8 n/a
OxFF message
0x01
sender 8 OxF n/a message sender ID
OxFF
0x01 ~
receiver 8 0FF n/a message receiver ID
IOxFF
Ox01 ask ask for specific information
0x02 tell send the requested information
performative 5 .
iper e 5 nform the other agent of leaving the
0x09 unregister network
network
Ox01 eligibility value for eligibility
0x02 motor speed value for motor speed
0x03 IR reading value for proximity detection
0x04 signal values indicate wireless signal
ontology 3 strength strength
0x05 power measurement for the remaining power
x6 camera signal to turn on/off the embedded
camera
0x07 connection information about the connection
0x02 data out .
service-request 2 0x02 data ot the information flow of the message
0x03 data in
in-reply-to 3 x00 n ID of a specific process for the
in-reply-to 3 n/a
0x07 message response regarding to
S, 0x00 ~ message ID regarding to the message
reply with 3 00 n/a
0x07 itself


* Length









The total length of each ACL message is not fixed. So this field indicates the total

length of the transmitted ACL message in order for the receiver to receive the message

correctly. This is an 8 bits field.

* Sender

This field specifies the source of the information. Comparing to the source address

in the data link layer, this ID is more flexible. It could be determined by the physical

address or another specific ID due to the change of the network topology. This

identification could be either the source address in data link layer protocol or a logical

address. The range of the address is from 0x01 OxFF.

* Receiver

This field is similar to the sender mentioned above. The value indicates to receiver

of the information. The range is also from 0x01 OxFF.

* Performative

The performative assigns the type of communicative act [35]. This field is 5 bits

long, so the value can vary from 0x00 ~ OxlF. The values which have already been

assigned could be found in Table 4.1. This field indicates all the control command and

information exchange for the conversion.

* Ontology

This term is used in conjunction with performative as an auxiliary statement to

express and interpret the message. Here this field is used to specify the type of the

information either requested or sent.

* Service-request









This is used here to indicate the direction of the information flow as another

auxiliary statement for some implicit performatives.

* In-reply-to

This field indicates the ID of a specific process for the message response regarding

to.

* Reply i ilth

This field assigns a specific number for the identification when any further process

and response is being made.

According to the above explanation, when we want to acquire the eligibility value

from robot 4 to robot 8, we can send an ACL message as shown in Figure 4.6(a). The

response for such an information query is in Figure 4.6(b). In Figure 4.6(a), no content is

included in the message. In the replied message, only one value is needed for the

eligibility query.

:sender 8 :sender 4
:receiver 4 :receiver 8
-Performative 1 -Performative 2
:ontology 1 :ontology 1
:service-request 3 :service-request 2
:in-reply-to 0 :in-reply-to 3
:reply-with 3 :reply-with 7
:content (a) :content Ox7A (b)

Figure 4.6 Examples of ACL messages. (a) request for eligibility value, and (b) response
for eligibility query.

According to the explanation of the physical meaning for each field and the

example, most of the messages transmitted by using the ACL are expected to be short.

The overhead of each message then would occupy a lot of portion for the whole message

length. This fact would reduce the performance of sending useful measurements across









the robots. In order to improve the efficiency, instead of sending the ASCII coded

message, the message in the ACL used here is encoded in a bit-wise manner. The

structure of the ACL message could be found in Figure 4.2.

A significant advantage could be made for the bit-wise encoding method. As of the

example found in Figure 4.7, the robot of 16 sends a message to the robot of 8 a

command of moving forward with the speed 100 on both wheels. The ASCII encoded

message is 129 bytes long. The time spent to transmit this message for a 38400 bps

wireless modem is 0.0268 sec. However, the same message encoded in a bit-wise method

is only 7 bytes long. The time required for transmission is only 0.00146sec. So the bit-

wise encoding is only 5% long comparing to the ASCII encoding. The similar results

could also be found in some relevant efforts under both KQML [36] and FIPA-ACL [37].

:sender 16 0 0 0 1 0 0 0 0
:receiver 8
-Performative TELL 0 0 0 0 1 0 0 0
:ontology motor speed 0 0 0 1 0 0 1 0
:service-request Data out 1 0 0 1 0 0 0 0
:in-reply-to 1
:reply-with 0 00 0 00
:content L100 R100 (a) 0 1 1 0 0 1 0 0(b)

Figure 4.7 Comparison between different encoding methods for ACL. (a) ASCII
encoding, and (b) bit-wise encoding.

Figure 4.8 shows the performance comparison between the ASCII encoding and

bit-wise encoding. It could be observed that the performance has significant

improvements for the shorter message since the overhead for a shorter message is a heavy

burden comparing to the longer message. Nevertheless, the efficiency for longer message

will still be better comparing to the ASCII encoding method.








79



Comparison of different message encoding methods
900
0 0 ASCII encoded
800 -+ 0 + Pure text
0 Pure number
0
700 + o
0
0
500- +
D + 0
(iL + 0

S+ 00
~ 400 +
O +
o- 0 00000000
( 300 +++ o
2 00 00 0 0 0000000000000000
200

100 <><><><><><>....<><><><><> <>0 <><> 0 >>0 >00

0 5 10 15 20 25 30 35 40 45 50
Length in message content(bytes)


Figure 4.8 Comparison of the performance on different message encoding methods


This chapter provides a network environment for the communication needs


between each robot. However, a mobile robot always changes it physical locations and


sometimes needs to change the network topology corresponding to different incidents or


mission requirements. Chapter 5 will proposed a procedure for the MRS testbed to


configure their logical connections dynamically in order to gain the flexibility during the


system operation.














CHAPTER 5
SELF-CONFIGURABLE TOPOLOGY

As in the centralized control scenario, all of the controlled devices connected to a

single command center. The network topology is fixed and therefore less flexible. Any

change of the connection or topology must be made from the leader in the network.

However, for each single robot, the lack of the ability to configure itself in response to

any event occurred regarding to itself reduces the autonomy on such a robot. As an agent

in multi-agent system, the agent generally should have the following properties:

autonomy, flexibility, and to own the thread to control itself. Therefore, the ability to be

autonomous to the environment including network connections promotes a robot to an

intelligent agent. a self-configurable network topology can also utilize the resources more

efficiently and response to any incident faster. Moreover, the failures on the members of

the MRS will not obsolete the operation of the whole system. Hence the system could

also be more robust. Each member in the system is also easier to be replaced and

upgraded. In the section 5.1, the use of an evaluation measurement is proposed to indicate

the eligibility of each robot. The eligibility value could be determined by various mission

objectives and multiple measurements on each single agent. Eligibility List (EL) can then

be generated and broadcasted globally based on the collection of eligibility value on each

available robot. Therefore, a dynamic task allocation method could be developed based

on the awareness of eligibility on each robot. The related discussion about the dynamic

task allocation can be found in [9] and [17]. Section 5.2 explains the procedure of the

self-configuration using the EL. Two different operational modes are suggested to









indicate the robot status as either leader or follower. The physical operations are

demonstrated under various scenarios in the rest sections as the examples of the self-

configuration procedure.

5.1 Eligibility List

The topology for the network is determined by the eligibility of each agent. The

MRS is designed for homogeneous/heterogeneous robots interoperation. For each

different single robot, different capabilities and resources are assumed. For example, the

remaining power, wireless signal strength, and the location of itself could all affect the

eligibility to perform the allocated mission tasks. The modules and actuators embedded

on each robot might vary as well. For instance, a robot with only IR sensors might need

to be guided by another robot with a camera so it could become location awareness. Also,

the robot with higher processing performance might take leadership of the whole robot

team. Therefore, the election of the most eligible robot to be the leading agent in the

network can optimize the system performance. During the operation of MRS, an EL is

generated by broadcasting the query to each robot. The EL contains two rows. The first

row contains the ranking of the eligibility for each robot, and the second row contains the

eligibility value for each robot. After the EL is generated, the list will again be

broadcasted to each of the robot. While the control of the network topology is still seized

by the leading robot, the most eligible robot could be determined and configuration on

the topology based on the available eligibility information could be made. The self-

configuration mechanism could also justify the unpredictable fault when the leading

robot fails and replaces it with the next eligible robot. This self-configurable mechanism

is discussed in the next section.









5.2 Self-configuration

A self-configuration mechanism is a critical request to take the system

communication to an autonomous level. The network functions by scanning the network

frequently from the leading robot, and the sensing of a connection timeout by the

follower robot. For each of the robot in the network, the capability to adjust itself

individually can be added by such a mechanism. Therefore, the biggest difference for this

network is the control of the network topology is decentralized. The flowchart of the self-

configuration mechanism is shown in Figure 5.1.

As shown in Figure 5.1, the operation of robot communication can be separated

into two different modes: passive mode and active mode. When the robot works under

passive mode, it waits for the connection request from the other robots. During this

operation mode, the robot works as a follower under the command of the leading robot

which connects to it. However, when the connection request isn't received over a period

of time, or the leading robot un-register from the network, a follower robot will assume

the leading robot becomes invalid and eliminate its existence from the EL. Then for each

follower robot, an inspection for the eligibility list to determine the most eligible robot

will be performed. The most eligible robot (with the highest eligibility value) will take

over the lead of the network and enter the active mode. During the active mode operation,

the leading robot will constantly scan the network to find the other available robots at a

given rate. Once a more eligible robot is found, the leading robot will relinquish its

privilege to control the other robots, return to its passive mode and wait for the

connection request from the most eligible robot.

The self-configurable network contains four different scenarios: network

initialization and new robot joining, follower failure, leader failure, and control privilege






83


transfer. The execution of the procedure will be explained in the later sections as the

simulation results for the network self-configuration mechanism. Before the

demonstrations of these procedures, the test configuration for this self-configurable

network is explained in the next section first.

SNetwork
Initialization

I I
I Wait for Yes No Leading Yes
I connection Disconnected? L
Request robot unregister
I No No
Receive Yes Get connected
I connection by the leading Timeout?
request robot
I No Yes


No


Timeout?


Most eligible No


Eliminate the
-most eligible
robot from EL


Yes


Passive Moc


e Operation


No-Anym robts. Search for
-4o.undt--'' other robots

Yes

Acquire eligibliy
a d broadcast L


IUnregister Yes M sm eligible No Wait for
from network robot foundT next scan


Active Mode Operation


Figure 5.1 Flowchart of self-configuration mechanism.









5.3 Test Configuration

The WALKER is a prototype for wireless multi-robot testbed. With the proposed

hardware architecture described in chapter 3, multiple WALKERs are capable to work

cooperative within the testbed. However, due to the research progress, only one

WALKER is manufactured currently for the hardware assessment. Therefore, tests of the

communication network in the following sections are performed by similar computer

equipment. Detail test configuration including wireless modem setting, used operating

system, and software development tool chains are described below.

Table 5.1 lists all computers used for the tests. The computer one is the PC/104

processing module embedded on WALKER. Other two computers are Intel PC and SUN

workstation respectively. The purpose for using different computers in the tests is also to

validate the compatibility of the communication network between heterogeneous robots.

However, all the tests are performed under Linux environments. The kernels used in all

three computers are under the family of Linux kernel 2.4. (See Table 5.1) It is,

nevertheless, not a real time kernel. The display during the execution is shown in Figure

5.2. In Figure 5.2, the program shows the value -1 to present the empty fields in the

eligibility list.

Table 5.1 Computer configurations for the test.
Computer 1 Computer 2 Computer 3
Processor INTEL Pentium INTEL Pentium 4 SUN UltraSPARC-
266 MHz (PC/104 1.8GHz IIi 440 MHz
board)
Memory 64MB 128MB 640MB
Operating System White Dwarf Linux Debian 3.0r4 Debian3.0r4
2.0
Linux Kernel 2.4.29 2.4.27 2.4.27
Hard Drive/ FLASH 512MB 10GB 9.4GB
Volume
Compiler Version Gcc 3.2 Gcc 3.3.5 Gcc 3.3.5
































Figure 5.2 Display during the test.

5.4 Network Initialization

The network initializtiion process is shown in Figure 5.3. In the beginning both

robots are in the passive mode. However, when the first robot gets the timeout from

waiting, it will enter the active mode and scan the network for other available robots. The

eligibility value will then be queried and transmitted between all the existing robots. The

leading robot is responsible for ranking all the robots in the list and then broadcast the list

to the robots which connects to it. The connection will be closed after the negotiation is

complete in the last step and waiting for another scanning after a short period of time.

Also, the same process will also work when more robots attempt to join the

network. The difference is when the leading robot scans over the network, more available

robots will be found. The connections from the leading robot to all available robots will

therefore be established and closed after it is finished.













Y wrlng for Mo1odion
wotel tor conwehnon



Robot













Robot 2














Reeene veery
roabo Te onre
Robot 1








Y RecNeh t11 q1ey,


Robot 1 Robot 2
statu
AcltN Pans,
EIb k L
1 1 7S





5





(c)





Robot 1












Robot


,.. I.n i [ re .







ROII








RobOt 2






Robot 1








I R -eeivet eligiblty list



Robot 2

Robot IRobot 2



75 t i yL^
1 1 75 1 75
2 2 30 2 3


4






~//^


Robot 1 Robot 2


ActUe Passt e

Eligigify Lst


8(b)

Robot 1 Robot 2
satus
Aclt Pash


1 1 75 1 75
S 2 30 2 30


Figure 5.3 Network initialization process. awaitingg for timeout, (b)connection

established, (c)eligibility query, (d)EL broadcast, and (e)close connection.


5.5 Follower Failure


Another possible scenario is the failure of the follower. This incident can be


detected and updated from the scanning by the leading robot. Once any available robot


becomes invalid, the response for connection request will no longer be seen. So the failed







87


robot will be removed from the EL at the leading robot. The new eligibility list will be

broadcasted through the network. This procedure could be seen in Figure 5.4.

i RbI 1 RODo RoDOl Rol I RooII R2 RomDl
1,1 P1


(a)
Robol 1 Robol 2 Roml ?










S(c)


Y



YI 1




H-2
I^^HIRBC- ) r
RIM*: i^ jr


Robol 1 Robol 2 Roal?




SI




' : (
t -- 1 ----- 7i I 7S
5 I3


SRo l l Roll 12 Rol R Rol 2 Ro l

i -W2 I Ii M Rb I W I











Figure 5.4 Topology configuration when follower fails. (a) a network with 3 robots, (b)
robot 7 fails, (c) leader scans the network, (d) failed robot detected and
S1 S I I 1 75 1 -i




I .Wu.td EL Y

R .- (e) R. (f)


Figure 5.4 Topology configuration when follower fails. (a) a network with 3 robots, (b)
robot 7 fails, (c) leader scans the network, (d) failed robot detected and
removed from the EL, (e) broadcast the EL, and (f) finish the configuration.

5.6 Leader Failure

One of the requirements for the dedicated network is to be robust to the possible

robot failures. A decentralized system could encounter the exceptional failure on any of

its member. Therefore, for the system autonomy, the communication network must be

able to accommodate itself to such an event, not only for the failures on followers but


TI*




R-1
R- 21 ," .,--..,




I.^ ^I Rt .........I