Citation
Design of a Robotic Citrus Harvesting End Effector and Force Control Model Using Physical Properties and Harvesting Motion Tests

Material Information

Title:
Design of a Robotic Citrus Harvesting End Effector and Force Control Model Using Physical Properties and Harvesting Motion Tests
Creator:
FLOOD, SAMUEL J. ( Author, Primary )
Copyright Date:
2008

Subjects

Subjects / Keywords:
Crop harvesting ( jstor )
Dates ( jstor )
Diameters ( jstor )
End effectors ( jstor )
Fruits ( jstor )
Orange fruits ( jstor )
Patents ( jstor )
Punctures ( jstor )
Robotics ( jstor )
Sensors ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Samuel J. Flood. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
12/31/2011
Resource Identifier:
658230848 ( OCLC )

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

DESIGN OF A ROBOTIC CITRUS HAR VESTING END EFFECTOR AND FORCE CONTROL MODEL USING PHYSICAL PROPERTIES AND HARVESTING MOTION TESTS By SAMUEL J. FLOOD A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Samuel J. Flood

PAGE 3

To God, through whom all things are possible; and to my wife, Tania, through whom He works everyday.

PAGE 4

iv ACKNOWLEDGMENTS I would first like to thank Dr. Thomas Burks for the opportunity to be a part of this research and for all of his support in conducti ng it. I would also like to thank Dr. John Schueller for all of his efforts in advising me on my concurrent master's degree. I would also like to thank my other committee member s: Dr. Arthur Teixeira , Dr. Daniel Lee and Dr. Carl Crane. All five of these gentleme n have provided valuable insight into my study and guided my professional development. Special thanks go to all of the staff and students in the Agricultural and Biological Engineering Department for their direct and indirect support. I es pecially thank Mike Zingaro, Greg Pugh, Dr. Michael Hannan, Stev e Feagle, Mary Hall, Betty Pearson, Babu Sivaraman, Siddhartha Mehta, Aaron Franzen, Paulo Younse, Ian Han, and Jai Sikes. Lastly, I would like to thank all of my friends and family. Their love and support kept me going these past 4 years. I would especially like to thank my wife, on whom I now bestow an honorary doctorate in agricultural engineering.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................viii LIST OF FIGURES...........................................................................................................ix ABSTRACT.....................................................................................................................xi ii CHAPTER 1 INTRODUCTION........................................................................................................1 Significance of the Citrus Industry...............................................................................1 Manual Orange Harvesting...........................................................................................1 Manual Harvesting Concerns.......................................................................................2 Motivation for Robotic Citrus Harvesting....................................................................3 2 LITERATURE REVIEW.............................................................................................5 Robotic Harvesting End Effectors................................................................................5 Citrus.....................................................................................................................5 Commodities Other Than Citrus.........................................................................14 Effects of Compressive Forces on Citrus...................................................................17 Effects of Various Harvesting Mo tions on Citrus Fruit and Trees.............................20 Use of Force Feedback for Control of a Robotic Harvester.......................................25 3 OBJECTIVES.............................................................................................................40 4 METHODS AND PROCEDURES............................................................................42 Physical Properties of Oranges in Re sponse to Applied Gripping Forces for Robotic Harvesting.................................................................................................42 Studies in the Optimization of Harv esting Motion Mechanics using a 7-DOF Manipulator with a 6-Axis Force/Torque Sensor....................................................43 Robotic Citrus Harvesting End Effector Development..............................................44 Stem Detection Studies and For ce Control Model Development...............................45

PAGE 6

vi 5 PHYSICAL PROPERTIES OF ORANGES IN RESPONSE TO APPLIED GRIPPING FORCES FOR ROBOTIC HARVESTING............................................48 Introduction.................................................................................................................48 Objectives...................................................................................................................51 Materials and Methods...............................................................................................51 Results and Discussion...............................................................................................54 Conclusions.................................................................................................................57 6 OPTIMIZATION OF HARVESTING MOTION MECHANICS USING A 7DOF MANIPULATOR WITH A 6AXIS FORCE/TORQUE SENSOR..................63 Introduction.................................................................................................................63 Objectives...................................................................................................................68 Materials and Methods...............................................................................................68 Results and Discussion...............................................................................................71 Conclusions.................................................................................................................77 7 ROBOTIC CITRUS HARVESTING END EFFECTOR DEVELOPMENT............88 Introduction.................................................................................................................88 Objectives...................................................................................................................92 End Effector Design Criteria......................................................................................92 Materials and Methods...............................................................................................92 Results and Discussion...............................................................................................94 Conclusions...............................................................................................................104 8 STEM-DETECTION STUDIES AND FORCE CONTROL MODEL DEVELOPMENT.....................................................................................................112 Introduction...............................................................................................................112 Objectives.................................................................................................................117 Materials and Methods.............................................................................................117 Results and Discussion.............................................................................................119 Matlab (2004) Code for Force Control Model..................................................120 Integration of the Force Control Model into the Overall Control of the Manipulator....................................................................................................121 Conclusions...............................................................................................................123 9 SUMMARY AND CONCLUSIONS.......................................................................128 Physical Properties of Oranges in Re sponse to Applied Gripping Forces for Robotic Harvesting...............................................................................................129 Studies in the Optimization of Harv esting Motion Mechanics using a 7-DOF Manipulator with a 6-Axis Force/Torque Sensor..................................................130 Robotic Citrus Harvesting End Effector Development............................................131 Stem Detection Studies and For ce Control Model Development.............................133

PAGE 7

vii Overall Conclusions..................................................................................................134 Future Work..............................................................................................................134 APPENDIX A YIELD POINT DETERMINATION.......................................................................136 Introduction...............................................................................................................136 Objective...................................................................................................................137 Materials and Methods.............................................................................................137 Results and Discussion.............................................................................................138 Conclusions...............................................................................................................139 B HARVESTING MOTION CODE FOR MANIPULATOR CONTROL.................142 C CODE FOR FORCE/TOR QUE DATA COLLECTION.........................................147 D SCHEMATIC OF DESIGNED INFRARED SENSOR...........................................151 E DERIVATION OF ROTATION MATR IX USED IN FORCE CONTROL MODEL....................................................................................................................152 LIST OF REFERENCES.................................................................................................155 BIOGRAPHICAL SKETCH...........................................................................................162

PAGE 8

viii LIST OF TABLES Table page 2-1 Reported results for previous ly developed end effectors.........................................39 5-1 Dimensional data summary of fruit used in puncture tests......................................60 5-2 Summary of punctu re test results.............................................................................61 5-3 Dimensional data summary of fruit used in burst tests............................................61 5-4 Summary of burst test results...................................................................................62 6-1 Number of fruit in each damage categor y for each test date for the linear pull test........................................................................................................................... .86 6-2 Number of fruit in each damage cate gory in each fruit/stem angle range...............86 6-3 Number of fruit in each damage category per number of rotations.........................87 7-1 Number of fruit in each damage category for each fruit/stem angle......................111 8-1 Observed and calculated angl es from preliminary tests.........................................127 A-1 Force levels used at each corresponding punch size used......................................140 A-2 Dimensional summary of fruit used in yield point dete rmination test...................140 A-3 Force levels where staining first appeared for each punch size used.....................141 A-4 Number of fruit punctured prio r to force level being reached...............................141

PAGE 9

ix LIST OF FIGURES Figure page 2-1 Kyoto University end effector...............................................................................31 2-2 Kubota end effector................................................................................................31 2-3 Instituto de Automática e nd effector harvesting motion........................................31 2-4 First CRAM end effector.......................................................................................31 2-5 Second CRAM end effector...................................................................................32 2-6 Third CRAM end effector......................................................................................32 2-7 U.S. Patent 3165880 end effector..........................................................................32 2-8 U.S. Patent 3756001 end effector..........................................................................32 2-9 U.S. Patent 3854273 end effector..........................................................................33 2-10 U.S. Patent 4519193 end effector..........................................................................33 2-11 U.S. Patent 4718223 end effector..........................................................................33 2-12 U.S. Patent 5005347 end effector..........................................................................34 2-13 U.S. Patent 5544474 rotating finger end effector..................................................34 2-14 University of Florida end effector harvesting sequence........................................34 2-15 AID helix end effector...........................................................................................35 2-16 U.S. Patent 3925973 helix end effector.................................................................35 2-17 U.S. Patent 4674265 end effector..........................................................................35 2-18 U.S. Patent 5544474 pulling end effector..............................................................36 2-19 Twisting end effector developed as a part of a joint French and Spanish research effort........................................................................................................36

PAGE 10

x 2-20 Italian twisting end effector...................................................................................36 2-21 U.S. Patent 4532757 end effector..........................................................................37 2-22 Japanese robotic tomato harvesti ng end effector harvesting procedure................37 2-23 Hungarian robotic apple harvester end effectors...................................................38 2-24 Iwate University robotic to mato harvesting end effector......................................38 2-25 Hybrid control block diagram................................................................................38 4-1 Burst test conducted with fruit-holding die...........................................................46 4-2 Punch test conducted with the 1.27 cm diameter punch........................................46 4-3 Force gauge and fruit holding strap used in the linear pull test shown pulling a fruit on one of the test trees....................................................................................46 4-4 End effector, force/torque sensor and robotic manipulator used in testing are shown reaching towards a fruit on one of the test trees.........................................47 4-5 Characteristic diamet ers taken of fruit...................................................................47 4-6 Designed end effector............................................................................................47 5-1 Characteristic diamet ers taken of fruit...................................................................59 5-2 Burst test conducted with fruit-holding die...........................................................59 5-3 Punch test conducted with the 1.27 cm diameter punch........................................59 5-4 Sample force vs. deformation curve fr om the 16 June set of tests, using the 1.27 cm diameter punch.........................................................................................60 5-5 Ratio of increase in value from value at smallest punch.......................................60 6-1 Force gauge and fruit holding strap used in the linear pull test shown pulling a fruit on one of the test trees....................................................................................78 6-2 End effector, force/torque sensor and robotic manipulator used in testing are shown reaching towards a fruit on one of the test trees.........................................78 6-3 Network of communication and pneuma tic lines for the harvesting system.........79 6-4 Characteristic diamet ers taken of fruit...................................................................79 6-5 Fruit mass variability ov er the harvesting season..................................................80

PAGE 11

xi 6-6 Fruit diameter variability over the harvesting season............................................80 6-7 Detachment force variability over the harvesting season......................................81 6-8 Percentage of fruit in each dama ge category for test dates 1 to 8..........................81 6-9 Deflection of the fruit from its initia l position versus detachment force for the tests conducted on 27 July.....................................................................................82 6-10 Detachment force versus fr uit/stem angle at 10° groupings..................................82 6-11 Distance traveled versus fr uit/stem angle at 10° groupings...................................83 6-12 Percentage of fruit in each damage cat egory for fruit/stem angle ranges 1 to 4....83 6-13 Detachment force versus number of rotations.......................................................84 6-14 Distance traveled versus number of rotations........................................................84 6-15 Percentage of fruit in each damage category for each number of rotations...........85 6-16 Secondary pitch harvesting forces.........................................................................85 7-1 Designed end effector..........................................................................................106 7-2 Designed grasping finger.....................................................................................106 7-3 Transfer mechanism between the connecting rods and fingers...........................106 7-4 Designed drive mechanism..................................................................................107 7-5 Integrated sensors.................................................................................................108 7-6 Integrated ultrasonic sensor.................................................................................108 7-7 Communications network for the harves ting system with th e new end effector and integrated sensors..........................................................................................109 7-8 Detachment force versus fruit/stem angle for the fruit tested with the new end effector.................................................................................................................109 7-9 Percentage of fruit in each damage category for fruit/stem angles 0, 65 and 110°......................................................................................................................110 8-1 Hybrid control block diagram..............................................................................124 8-2 Fruit, stem and reaction forces.............................................................................124 8-3 Fruit and reaction forces with the fruit axis assumed to be or iented vertically...124

PAGE 12

xii 8-4 Stem angle detection tests with the white string connecting the fruit and the main supporting branch........................................................................................125 8-5 Formula for forming the rota tion matrix for a rotation of about axis r represented by unit vector components rx, ry, and rz............................................125 8-6 Flowchart of the force control model w ith assumption of a vertical fruit axis....126 8-7 Block diagram of an integration of the force control model into the overall control of the manipulator....................................................................................126 A-1 Slight staining around punch area........................................................................140 A-2 Pronounced staining around punch area..............................................................140 D-1 Schematic of IR sensor designed by t echnician Greg Pugh and integrated into the end effector of chapter 7................................................................................151

PAGE 13

xiii Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy DESIGN OF A ROBOTIC CITRUS HAR VESTING END EFFECTOR AND FORCE CONTROL MODEL USING PHYSICAL PROPERTIES AND HARVESTING MOTION TESTS By Samuel J. Flood December 2006 Chair: Thomas F. Burks Major Department: Agricultur al and Biological Engineering Because of rising costs and a shrinking labor force, it is increasingly desirable to harvest citrus fruit robotically. An integral part of the succe ss of a robotic harvester is the end effector. The studies conducted worked toward the development and implementation of a grasping robotic citr us harvesting end effector. The specific objectives of these studies were as follows: determine the safe grasping limits for the end eff ector through physical properties tests; determine the most effective harvesting motion through tests with a 6-axis force/torque sensor; design, build and test the end effector base d on the results of these tests; determine if stem location detection may be used to optimize the harv esting motion; and cons truct a force control model that would utilize feedb ack from a force/torque sensor to control the manipulator. The physical properties studies were conducte d using a universal testing machine, a set of cylindrical punches, and a flat plate. A model was developed that related the punch

PAGE 14

xiv diameter and radius of curvature of the fr uit to the puncture force. Recommendations were then made as to the design of the end effector. The harvesting motion studies examined the optimization of harvesting motion mechanics using a 7-DOF manipulator and a 6-ax is force/torque sensor. The influence of various harvesting motions was examined on each of the following: fruit damage, distance traveled, and harvesting forces. Re sults showed that a harvesting motion that imparts a 90° angle between the fruit and stem and a 180° rotation about the stem axis is the most effective. Using the results of the previously c onducted studies, the end effector was developed based on specific design criteria. The end effector was then tested in a commercial grove where it was able to harvest all of the attempted fruit successfully. A proposed control model was then devel oped that simulated the control of the manipulator based on feedback from the for ce/torque sensor. The trajectory of the manipulator could be modified based on the desired angle a nd the sensed angle between the fruit and stem. A method of integrating the control model into the overall control of the manipulator was then presented.

PAGE 15

1 CHAPTER 1 INTRODUCTION Because of the high cost and limited availa bility of human labor and the need to harvest as inexpensively as possible in a tim ely manner, it is incr easingly desirable to harvest fruit robotically. Several harveste r designs for various agricultural commodities have been tried over the past 2 decades. The success of the harvester depends on its ability to quickly detect and harvest as much fruit as possi ble without damaging the fruit or the tree. An integral part of this success is the end effector . An end effector is a tool or device attached to the end of the manipulator that carries ou t a specific task. In robotic harvesting, it is the part of the robot that actually does th e harvesting. Because of its direct interaction with the fruit and tree structure, it must be designed with the specific physical properties of the commod ity to be harvested in mind. Significance of the Citrus Industry Agriculture is the second largest indust ry in Florida (Marth and Marth, 2001). Citrus production is the largest agricultural commodity in Florida (Hodges et al., 2001). This results in a total economic impact to Florida of $9.13 billion that includes the impact of the 89,700 jobs created in Florida as a result of the industry (Hodges et al., 2001). Manual Orange Harvesting Oranges are detached manually by thr ee different methods depending on area, variety, and cultural practice. First, the laborer can use a se t of clippers to detach the fruit, usually leaving as short a stem as possible. Second, the laborer can lift the fruit so that the stem axis is rotated 90° and then pu ll down so that the force is perpendicular to

PAGE 16

2 the stem axis. Third, the laborer can add a twisting motion to th e second method. This combination of bending, pulling, and twisting wa s described by Ellis (1971) in U.S. Patent 3566594. Although the end effector n eed not necessarily follow one of these methods, an understanding of ma nual procedures offers insight into some of the potential methods. Manual Harvesting Concerns There are several concerns in manual harv esting of citrus. One concern is longterm availability of labor supply (Roka a nd Longworth, 2001). More desirable jobs in manufacturing, service, and c onstruction lure workers to le ave their manual harvesting jobs. Also, most Florida farm workers are not U.S. citizens. Increased focus on security and immigration issues may aff ect the availability of labor. There are also concerns over internationa l competition from countries with lower labor costs (Yardley, 2004). As harvesting costs are 35 to 45% of total production costs (Sanders, 2005), this dramatically lowers th e total production cost of the countries with lower labor costs. To equalize production costs, tariffs have been placed on foreign fruit entering the U.S. For Brazil, 1/3 of the cost for importing ci trus is for equalization taxes and tariffs (Muraro et al., 2003). These tari ffs are scheduled to be removed as part of the Free Trade Area of the Americas. This would create a free trade area that included all of the countries in North, South, and Central Am erica (including the Carribbean, with the exception of Cuba). Scheduled completion of negotiations for the cr eating of this free trade area was January 2005 (Zoellick, 2002). Ne gotiations are still in progress, but all countries involved still express a desire for creating this area.

PAGE 17

3 Finally, there are concerns over the safety of the manual harvesters. The largest area of worker injury in citrus is in th e harvesting operations, with the most common injuries coming from falling from ladders and strains due to lifting (Andrews, 1999). Motivation for Robotic Citrus Harvesting In the search for a solution to the concerns over manual harvesting, several mechanical harvesters have been developed. Research shows that mechanical harvesting solutions produce unacceptable levels of da mage for fresh fruit (Hannan and Burks, 2004). As the 11% of Florida citrus sold in the fresh market translates into $493.6 million in fruit output value (Hodges et al ., 2001), there is a motivation to find a harvesting solution which would be appropria te for the fresh fruit market. It was therefore the purpose of the studies in this di ssertation to work toward developing such a solution. The specific objectives of the studi es in this dissertation were as follows: 1. Determine the safe grasping limits for a grasping robotic citrus harvester end effector through physical properties tests. 2. Determine the most effective harvesti ng motion through tests with a 6-axis force/torque sensor. 3. Design a robotic citrus harvesting end eff ector based on the results of the physical properties and harvesting motion tests. 4. Build and test the designed robotic ha rvesting end effector to determine its effectiveness based on the design criteria. 5. Determine if stem location detection ma y be used to optimize the harvesting motion without the need for additional movements. 6. Construct a force control model that woul d utilize feedback from a force/torque sensor to control a robotic citrus harvesting manipulator. This chapter reviewed the motivations behi nd the studies in this dissertation. The next chapter reviews previous research that is relevant. The pr evious research is separated into four sections: robotic harvesting end effectors, the effect of compressive

PAGE 18

4 forces on citrus, the effect of various harves ting motions on citrus fr uit and trees, and the use of force feedback for control of a robotic harvester. A more in depth discussion of the objectives in this dissertation is then presented in chapter 3 followed by an overview of the materials and methods in chapter 4. Chapters 5 through 8 then present the studies conducted as a series of papers. A summary of all four paper chapte rs is then presented along with future work in chapter 9.

PAGE 19

5 CHAPTER 2 LITERATURE REVIEW The previous end effectors developed for citrus and related commodities were first reviewed. Second, the previous research performed in studying the effects of compressive forces on citrus that might be e xperienced by fruit in a grasping end effector was reviewed. Third, the previous efforts to examine the impact of various harvesting motions on the fruit and tree were reviewed. Finally the previous work using force feedback to control the motion of a robotic harvesting system was reviewed. Robotic Harvesting End Effectors Citrus Several researchers have tried to develop a robotic citrus harves ter. These included efforts in Japan, Italy, Israel, France, Spai n, and the United States. The end effectors developed by those researchers fall into four types: cutting, pulling, twisting, and twisting/pulling. The most prevalent design is the cutting end effector. This design cuts the stem, thus harvesting the fruit. Kawamura et al. (1987) and Sarig (19 93) described a cutting end effector developed at Kyoto University in Japan (Figure 2-1). This end effector uses three rubber fingers that are actuated pneumatically by rubbertuators. The r ubbertuators are rubber tubes coated by textile cords and act like pneumatic muscles. When the tube is contracted, the fingers bend in a smooth cu rve. The stem was severed by means of scissors attached to two pne umatic cylinders. The first cylinder moved the scissors forward to encompass the stem. The second cy linder actuated the sc issors, to sever the

PAGE 20

6 stem. Stems larger than 3.5 mm diameter coul d not be cut because of pressure limitations of the second cylinder. Also Kondo et al. (1998) reported that when twigs or leaves surrounded the fruit, the rubbertuators were be nt, thus preventing th e end effector from grasping the fruit. A second cutting end effector describe d by Ito (1990) and Sarig (1993) was developed by the Kubota Corporation of Japan (F igure 2-2). This end effector attaches to the fruit using a vacuum pad. It then slides comb covers over the fruit in order to protect neighboring fruit from the cutting action. A set of cutting jaws which are located underneath the comb covers swing around the to p and bottom of the fruit. The stem is severed when the saw tooth cutters meet. A third cutting end effector was developed at the Instituto de Automática and was described by Ceres et al . (1998). In this end effector a pneumatic suction cup is used to attach to the fruit. A v-shaped centering devi ce is then moved forward over the top of the fruit in order to bring the stem into the cente r of the v. Once infrar ed sensors detect the presence of the stem, it is cut using a cutt ing tool. The harvesting motion is shown in Figure 2-3. The cycle time for just the gras ping and detaching pro cedure was 2 seconds. Muscato et al. (2005) described three cutt ing end effectors proposed by researchers at the Consorzio per la Ricerca in Agricoltu ra nel Mezzogiorno ( CRAM) in Italy. The first uses a mirror placed beneath the fruit to assist the camera in moving the end effector to be centered around the fruit (Figure 2-4). A neoprene spiral would then close around the fruit, centering the stem. The stem would then be cut and the fruit would be held in the end effector through the us e of a depressurizable tube. This end effector was never constructed because of concer ns over cost and robustness.

PAGE 21

7 The second cutting end effector described in Muscato et al. (2005) utilized three pneumatically activated flexib le fingers (Figure 2-5). Th e fingers would close around the fruit as pneumatic pressure was increased. Th e end effector was then pulled back so that the stem would pull on the fruit. The stem location was then determined through the use of a force sensor mounted on the wrist of th e manipulator. The end effector was then rotated to bring the stem into the acting region of the cutter. The stem was then cut using a circular micro-saw. The end effector had a high success rate, but the cost of implementation was too high to be used in a commercial model. The last cutting end effector described in Muscato et al. (2005) involved two jaws which would close around the fruit once the en d effector was positi oned in front of the fruit by the manipulator (Figure 2-6). The jaws would guide th e stem to the middle of the upper jaw where a v-shaped cutter was locate d. This would cut the stem as the jaws closed. Buie Jr. (1965) in U.S. Patent 3165880 desc ribed a seventh cutti ng end effector in which the fruit would be drawn into a cylin der through the use of pneumatic suction (Figure 2-7). Once mechanical contacts detect ed the presence of th e fruit, a blade would swing behind the upper half of th e fruit thus severing the stem. Middleton Jr. (1971) in U. S. Patent 3564826 proposed an eighth cutting end effector developed at ILC Indus tries of Delaware. In this design the fruit was drawn into a cup using pneumatic suction. The stem woul d then be severed when a rotating cutter passed around the back of the fruit and encount ered either a counteracting cutter halfway or a stationary cutter located on the opposite side of the suction cup.

PAGE 22

8 Macidull (1973) in U.S. Patent 3756001 pr oposed a ninth cutting end effector which would utilize pneumatic suction to draw the fruit into a tube (Figure 2-8). Three cutting blades which are located behind the fruit would then be actuated to sever the stem. The blades are located at three equa lly spaced intervals along the outer perimeter of the tube and are curved so as to conform to the perimeter of the tube. When actuated the blades come together so as to force the stem to be cut between two of them. Rosenberg (1974) in U.S. Patent 3854273 proposed a tenth cutting end effector which would be positioned so as to encompass the fruit in a cylinder (Figure 2-9). Leaf plates would then be closed behind the fruit in order to capture the fruit. A cutting mechanism located on the leaf plates would th en be actuated and w ould sever the stem by passing behind the fruit. Yoshida et al. (1985) in U.S. Patent 4519193 detailed an eleventh cutting end effector developed by the Kubota Corporati on of Japan (Figure 2-10). In this end effector the fruit is drawn into a tube using pneumatic suction. Photosensors are used to detect when the calyx is in a position to be cut by cutting blades si milar to that proposed in Macidull (1973). The stem is then severe d by these blades in the same manner as in Macidull (1973). Suzuki et al. (1988) in U.S. Patent 47 18223 proposed a twelfth cu tting end effector developed by the Kubota Corporation of Japa n (Figure 2-11). The end effector would consist of a telescopic vaccum pad to extend a nd attach to the fruit. Once a vacuum seal had been attained, the rest of the end eff ector would move forward to encompass the fruit. Hemispherical covers would enclose th e fruit and end effector so as to prevent

PAGE 23

9 damage to the surrounding fruit and tree st ructure. A cutting mechanism would then swing behind the fru it and sever the stem. Kedem and Rubinstein (1991) in U.S. Pa tent 5005347 proposed a thirteenth cutting end effector developed by Du-Kedem Indus tries, Ltd. and Du-Kedem Technologies, Ltd., both of Israel (Figure 2-12). This end effector would be maneuvered in order to encompass the fruit within a cylindrical tube. Once the fruit was detected to be inside the tube, a set of half-circle bails would be actua ted to close behind the fruit. This would move the stem of the fruit to the center plane and also lock the fruit into the end effector. The end effector would then be moved downward until the surface of the fruit was against the closed bails. Once this was done a second set of bails with cutting edges that were located perpendicular to the first set of bails would close. This would sever the stem as close to the fruit as possible. Terada (1987) proposed a fourteenth cutti ng end effector that was developed by the Kubota Corporation of Japan. In this end effector, a contact sensor would determine when the end effector was beneath the fruit. The sensor would then lower inside the end effector and the manipulator would move the end effector up until the contact sensor was in contact with the fruit again. Once the fru it was inside the end effector a set of fruit pressing members would close in around the fr uit and position the stem to be in the middle of the top of the end effector. A cutte r blade would then swing over the top of the fruit, severing the stem. Finkelstein (1996) in U.S. Patent 5544474 proposed three end e ffectors. Two of them, cutting end effectors fifteen and sixteen, are substantially similar and involve a set of circularly rotating cutter fingers (Figure 2-13). Th e end effector would move into a

PAGE 24

10 position so that the cutter fingers extended above the fruit to be harvested. As this is done the stem would move into a position in between two of the cutter fingers. The fingers would then rotate, se vering the stem. The fruit w ould be held by a u-shaped mechanism located underneath the fruit. The difference between these two end effectors is that in one a single set of cutter fingers rotate whereas in the second end effector one set of fingers rotate against a stationary second set of fingers. Cutting end effectors are also prevalent in other agricultural applications as they produce the least amount of stre ss on the fruit. Problems can arise, however, if the blade is not sharp enough or does not strike the st em properly, then a pulling action results and failure at the peel, referred to as plugging, of ten occurs. Another pr oblem associated with the passing of a cutter around the fruit involves the blade size. The blade must be large enough to be able to completely envelope the fruit, but not so large that it damages the surrounding fruit or tree structur e. Regardless of protection for the surrounding fruit and tree structure that may be present, a larger end effector impedes th e ability to penetrate the canopy and harvest interior fruit. A cutting end effector is impractical for the Florida citrus industry, as it is deemed undesirable fo r any stem to be present on the citrus after harvesting. The second type of end effector is the pu lling end effector. This design harvests the fruit by pulling on the fruit in a particular direction, us ually along the stem axis. Once the force applied exceeds the binding forces along the fruit/stem interface, the fruit is harvested. This method was proposed by Pool and Harrell (1991). This end effector, developed at the University of Florida, used a rotating lip with an attached collection

PAGE 25

11 sock to encircle the fruit. As the lip ro tated around the fruit, the stem was impinged between the lip and the upper por tion of the end effector. The stem was then severed as the end effector retracted. This harvesti ng sequence is illustrated in Figure 2-14. The end effector successfully removed the fru it 69% of the time. Of these successful attempts, 63% of them caused damage to either the fruit or the tree. Damage to the tree accounted for 85% of this damage which usually involved the removal of leaves or small branches. The most common fruit damage wa s when the fruit was caught between the lip and the housing. Plugging of the fruit was rare ly observed. The residual stem lengths were between 0 and 2 inches with the occasiona l stem greater than 3.9 inches. The effect of replacing the lip with a cutt er bar was also examined, but resulted in frequent cut fruit and residual stem lengths be tween 0.8 to over 5.9 inches. A second pulling end effector was de veloped at Agricultural Industrial Development S.p.A. (AID) in Italy and was de scribed in Muscato et al. (2005) (Figure 215). This end effector consiste d of a helix which rotated when the presence of a fruit was detected using an infrared proximity sensor. This pulled the fruit into the helix which was surrounded by a cylinder. The stem was severed when it was pulled in between the helix and cylinder. This end effector had problems with da maging the fruit and the tree as well as its inability to ha ndle the fruit once it was picked. Glover (1975) in U.S. Patent 3925973 propos ed a third pulling end effector similar to that described in Muscato et al. (2005) (F igure 2-16). The end e ffector would consist of a rotating auger located in a cylindrica l tube. The fruit would be drawn in by an opening in the tube perpendicular to the axis of rotation. The stem would be severed as

PAGE 26

12 the fruit was drawn into the tube and the st em was caught between the rotating auger and the edge of the opening. Gerber (1987) in U.S. Patent 4674265 desc ribed a fourth pulli ng end effector in which the pneumatic suction would bring the fruit into a tube that was at a downward angle (Figure 2-17). The fruit would be brought up into the tube with the stem then bent at an angle against the lip of the tube. Th is pulling action on the fruit would then sever the stem. The third end effector proposed by Finkels tein (1996) in U.S. Patent 5544474 was a pulling end effector. The end effector consisted of three fingers equally spaced around the circumference of the end effector (Figure 2-18). Th e gripper would move forward until the fruit was determined to be within the grasp of the fingers. The fingers would then be actuated to close around the fruit a nd the fingers would retract within the end effector. As the fruit was pulled inside the end effector, the stem would then be bent against a plate which surrounded the fingers. This combination of bending and pulling would then sever the stem. As was stated earlier, in a pulling type end effector th e binding forces along the fruit/stem interface are often large enough that it results in plugging. These forces can be reduced if the force is applied at a 90° angle to the axis of attachment, as is often done in hand picking. This method results in less pl ugging but may increase the complexity of the motion and, thereby, increase the harves t time. Pulling actions also disturb the surrounding fruit through limb oscillation, ma king subsequent rapid harvesting more difficult since moving fruit are much harder to harvest than stationary fruit.

PAGE 27

13 The third type of end eff ector design is the twisting method. This method was recommended by Juste et al. (1988) and Rabatel et al. (1995) as the most promising of the three. This involves twisting the fruit, prefer ably about its attachme nt axis, until the stem is severed. Twisting the fruit about the stem axis reduces the amount of disturbance to the tree and thus to the surrounding fruit. A tw isting end effector was developed as a part of a joint French and Spanish research effort described in Juste et al. (1992) (Figure 219). A pneumatic suction cup gripped the fr uit from either the line of action of the telescopic arm or at a 30° angle from the line of action toward s the bottom of the fruit. The suction cup was then rotated so as to sever the stem using a hydraulic motor. The detachment rate of the end effector was 64 to 67% with the failures caused by obstacles, vacuum failures, or mechanical failures. The rate of plugging of this end effector was 2 to 4%. The residual stem length was less than 5 mm in 80 to 85% of the detached fruit. The end effector did not damage the fruit thr ough its contact with the fruit, and 4 to 5% of the detached fruit were damaged by the surrounding tree structure during removal. A second twisting end effector was proposed by Bullock (1956) in U.S. Patent 2775088 (Figure 2-20). It consisted of a cone which would frictionally grip the fruit using pneumatic suction. The cone assembly is then driven rotationally which results in a combination of twisting and pulling forces ge nerated on the stem. This combination of forces severs the stem. Similar end effect ors were proposed by Black Jr. (1969) in U.S. Patent 3460330, Connery (1971) in U.S. Patent 3591949, which was developed by the Chisholm-Ryder Company, Inc. of New York, and Guglielmino et al. (1996) which was developed in Italy.

PAGE 28

14 As in the pulling and cutting ty pe end effectors, fruit size is a consideration here as well. Generally, the twisting action is achie ved by use of a rotating suction cup. This cup must be of the right si ze to create a good seal while still providing enough force to keep the orange from slipping. Similar work was done on mushrooms by Reed et al. (2001), and it was found that the smaller suction cups tended to slip, which damaged the mushrooms. The researchers found that it wa s necessary to have the largest possible suction cup that would still be able to a dhere to the mushroom. One of the major advantages of this method is that if only a su ction cup is used, there is a large flexibility in the angle of approach. Some angles ma y be more preferential than others. Except where the stem attaches to the fruit, the c up could attach to any part of the fruit. Based on the literature available it appears that a combination of the twisting end effector and the pulling end effector holds th e most promise of success. By combining designs, both detachment time and force are possibly reduced. This method was proposed by Tutle (1985) in U.S. Patent 4532757 and Tutle (1983). This end effector was developed by Martin Marietta Corporation and consisted of a four fingered gripper (Figure 2-21). Each of the fingers were equally spaced around the circumference of the gripper. Once the presence of the fruit is dete cted within the grippe r by three of the four tactile sensors located at the rear of the gripping area on each of the fingers, the fingers are actuated to close around the fruit. To sever the stem, the gripper pivots the head downwards and then rotates. Commodities Other Than Citrus Robotic harvesting end effectors have also been developed fo r other agricultural commodities. Described here are several end effectors which were designed for commodities that resemble citr us in their shape and size.

PAGE 29

15 Monta et al. (1998) described the design of a robotic tomato harvesting end effector developed in Japan. The fina l design involves the use of a telescopic suction pad to attach to the fruit and retract the fruit into the grasping area of the end effector. Once the fruit is in a position to be grasped, two parall el plate padded grippers close on the fruit. The end effector then bent the fruit at the pedun cle in order to harvest the fruit. The final design also allowed for short peduncles by checking for the negativ e pressure in the vaccum pad. If the pressure became too ne gative, and thus approaching the point at which the fruit would be pulled from the v acuum pad, the fingers would begin moving forward at the same time that the vacuum pa d was retracting. This would prevent the pressure from becoming too negative. The harv esting procedure is illu strated in Figure 222. Kassay (1992) described the e nd effector for the Hungarian robotic apple harvester, Aufo. The end effector described uses four padded fingers to grip the fruit (Figure 223a). Compensation springs were also used on the finger actuators in order to prevent excessive grasping pressures. The pressure limit set was 100 N/cm2. The fingers were actuated by an electric motor. A pneumatic venturi end effector wa s also shown, but not described (Figure 2-23b). The motion of this end effector is de scribed in Kassay and Slaughter (1993). The end effector approaches the fruit from below at an angle of 45°. The end effector was originally programmed to return in the same direction in which it approached the fruit, however it was observed that this led to the harvester impacting branches. The retraction movement was thus mo dified to retract the end effector directly horizontal out of the tree. This movement is what harvests the fruit.

PAGE 30

16 Kataoka et al. (1999) describe d the design of a robotic ap ple harvesting end effector developed at Iwate University in Japan (F igure 2-24). The design consisted of two parallel grippers which would gr ip the peduncle of the apple. The grippers would then be rotated about the abscis sion layer in order to break the pe duncle and thus harvest the fruit. Two supporting rods were also attached whic h would support the fruit as it was rotated. The success rate of the end effector was f ound to be 92.3% on 26 replications. The failed attempts were due to dropped fruit. Grand DÂ’Esnon et al. (1987) and Pellenc et al. (1990) in U.S. Patent 4975016 described the design of Magali, which is a robotic apple harvester developed at the CEMAGREF in Montpellier, Fran ce. The end effector consis ts of a vacuum cup with a pressure sensor. Once a seal was attained on the fruit, the harvesting arm retracts to harvest the fruit. The robot was successful 50% of the time with 75% of the fruits still having their stems, which is desirable. It is recommended that further designs should include a twisting motion in order to pick all of the fruits with their stems. A summary of the reported results from all of the reviewed end effectors is provided in Table 2-1. The most promisi ng end effectors developed were the three fingered articulated end effect or described in Muscato et al. (2005), the twisting/pulling end effector proposed in Tutle (1983) a nd the three fingered pulling end effector proposed in Finkelstein (1996). It is uncertain what the re sults of these end effectors were. Muscato et al. (2005) only reported that the results were extraordinary. The other two end effectors had no reported results. Th ese were the most promising because they were all able to grasp the fruit. Although they were not all used to impart a specific motion to the fruit, a similar de sign could be used in this manner. Of those end effectors

PAGE 31

17 that did report results, none were successful enough to be im plemented in a robotic citrus harvester. It is therefore th e purpose of the studies in this dissertation to design a robotic citrus harvesting end effector that will meet specific design criteria for implementation in a robotic citrus harvester. Effects of Compressive Forces on Citrus As a part of the development of a grasping r obotic citrus harvester end effector, it is important to determine the safe handling limits for citrus fruit during harvest. Several studies have been published that examine the resistance of the rind to puncture. Ahmed et al. (1973) examined the forces required to ruptur e peel oil glands and to puncture the fruit rind. This was done for field run (unwashe d, unwaxed), commercially processed, and irradiated fruit. It was obse rved that both commercial pr ocessing and irradiation had a detrimental effect on the structural st rength, although the differences between commercially processed and field run fruit were not statistically significant. Twenty fruit were used per replicate, with three replicates performed. The punch diameters used were: 0.258, 0.264, and 0.051 cm. Churchill et al. (1980) examined the fo rces that fruit might experience during mechanical harvesting. Both puncture and burst tests were performed. The influence of variety, harvest date, harvest time, and absc ission chemical application were examined. The sample size was 20, with two to four repli cations per year. It was observed that there were varietal differences, with the Pineapple ( Citrus sinesis cv. Pineapple) variety having a greater structural strength than the Hamlin ( Citrus sinesis cv. Hamlin) variety. The harvest date and time did not have statisti cally significant differences, although the Pineapple variety required incr easing puncture and burst forces in later harvest dates. The punch diameter used was 0.64 cm.

PAGE 32

18 Coggins Jr. and Lewis (1965) examined the ch ange of puncture forces over time as well as the effect of gibbere llic acid treatments. A compar ison of compressive forces versus shearing forces was also done. They observed that for two different penetration sizes with area ratios of 2 and circumferential ratios of 1.41, a puncture force ratio (larger size puncture force over smaller size puncture force) of 1.56 was required, indicating that both compressive and shearing forces were pres ent. However, shearing forces were more dominant. The gibberellic acid treatment, if applied early enough in the growth process, had the effect of slowing the rate of decrease in puncture resistance. McDonald et al. (1987) found supporting results. They observe d that the puncture resistance decreased over time. Twenty fruit were used, with ten punctures per fruit. The punch diameters used were 0.10 and 0.1438 cm. Juste et al. (1988) examined citrus fruit prop erties as they pertained to fresh fruit robotic harvesting. In each puncture test, 50 fruit were used with a punch diameter of 0.047 cm. These tests were performed at four different times during the growth season. They found that puncture resi stance had a decreasing trend, except for the beginning and end of the growth seasons, where punctur e resistance might increase depending on the variety. Miller (1986) examined dimensional prope rties and puncture resistance, along with modulus of elasticity an d average stress. In each test, 40 fruit were measured for their dimensional properties, and 20 fruit were m easured for puncture resistance. Results from that work suggested that maximum compressive strength was related to the peel strength, whereas the deformation was related to intern al structural integrit y. The punch diameter used was 0.64 cm.

PAGE 33

19 Other punch diameters previously used we re: 0.5 and 1 cm (Chuma et al., 1978), 0.10 cm (Coggins Jr., 1969), 0.05 cm (Fidelibus et al., 2002a), 0.1 cm (McDonald et al., 1987), and 0.320 cm (Turrell et al., 1964). The aforementioned studies have an equivalent punch diameter range of 0.05 to 1 cm. Separa te burst tests were also performed by Ahmed et al. (1973), Chuma et al. (1978), C hurchill et al. (1980), Fidelibus et al. (2002b), Miller (1986), and Sarig and Orlovs ky (1974). These tests we re conducted in the same manner as the punch tests, except th e punch was replaced with a flat plate. The studies cited above were conducted on a wide range of citrus varieties, which included: Valencia ( Citrus sinesis cv. Valencia), Hamli n, Pineapple, Satsuma ( Citrus reticulata cv. Satsuma ) , Washington Navel ( Citrus sinesis cv. Washington Navel), Shamouti ( Citrus sinesis cv. Shamouti), Salustiana ( Citrus sinesis cv. Salustiana), and Temple ( Citrus sinesis cv. Temple) oranges; Duncan ( Citrus paradisi cv. Duncan) and Marsh ( Citrus paradisi cv. Marsh) grapefruit; Bearss ( Citrus limon cv. Bearss) and Eureka ( Citrus limon cv. Eureka) lemons; Persian limes ( Citrus latifolia cv. Persian); and Dancy tangerines ( Citrus reticulata cv. Dancy). The variety studied most often was the Valencia orange. These studies used punch sizes that we re well below the size that might be experienced by a fruit under robotic harvesti ng. The largest punch size studied was 1 cm. It was necessary to understand more fully how a fruit would respond to a robotic harvester by conducting studies that bridge the gap betw een the previously conducted puncture studies and the previously conduc ted burst studies. Therefore one of the objectives of this study was to develop a relationship between punch size and puncture force that would be applicable to th e development of a robotic harvester.

PAGE 34

20 Effects of Various Harvesting Motions on Citrus Fruit and Trees As a part of the development of a robotic citrus harvester end effector, it is important to determine the forces required to harvest the fruit without damaging the fruit or tree. Several studies have been perf ormed which examine the impact of various harvesting motions on the fruit and the tree. Barnes (1969) studied the detachment characteristics of lemons as they may pe rtain to the development of a mechanical harvester. In this study fruit were separate d from the tree with the stem intact. Tests were then performed by applying forces collinea r to the stem as well as at a 45° angle to the stem. A torsion test was also performed by applying loads at a 5 cm (2 in.) radius from the base of the calyx. In all three tests the load which caused failure was recorded as well as the point at which failure occurred along the fruit-stem interface. These three tests were performed on 4 diffe rent dates using the fruit from 10 different trees with 10 fruit of varying diameters for each test. Se parate collections we re performed for each side of each of the trees. Th is was done for three different locations. Two of these were located in California and the third was in Arizona. The Eureka ( Citrus limon cv. Eureka) lemon was present at the Californi a locations while the Lisbon ( Citrus limon cv. Lisbon) lemon was present at the Arizona location. The results of all three te sts indicated that the detachment force was directly correlated to the fruit size with the larger fruit requiring higher detachment forces. There was no statistical difference among observed detachment forces on different sides of the tree or among trees at a specified location. The most common form of detachment was separation of the calyx. The force required for detachment in the angle-pull tests was less th an half that required for the straight pull. In the torsion test, there was however a not iceable difference between the detachment

PAGE 35

21 torques for the Lisbon and Eureka varieties w ith the Lisbon variety re quiring consistently higher torques than the Eureka lemon. Coppock et al. (1969) studied the detach ment characteristics of Valencia ( Citrus sinesis cv. Valencia), Hamlin ( Citrus sinesis cv. Hamlin), and Pineapple ( Citrus sinesis cv. Pineapple) oranges and Marsh ( Citrus paradisi cv. Marsh) grapefruit as they may pertain to the development of a mechanical ha rvester. In their study the detachment force at 0, 45, and 90° angles from the fruit’s axis was determined as well as the form of detachment and the possible correlations to othe r citrus properties. Three samples of 40 fruit each were taken at four dates from four trees. The fruit were detached with the stems intact with a length of 10 to 15 cm (4 to 6 in.). The loading rate was approximately 44 N/s (10 lb/s). Detachment force at all three angles decreased over the harvest season except for the Valencia variety which increa sed. Detachment force and percentage of plugging decreased for a ll varieties as load ing angle increased. Fornes et al. (1993) and Just e et al. (1988) examined th e detachment characteristics of oranges as they pertained to the deve lopment of a robotic citrus harvesting end effector. Fornes et al. (1993) examined the detachment force at the beginning, middle, and end of their maturation period of Clausellina ( Citrus reticulata cv. Clausellina ) , Clemenules ( Citrus reticulata cv. Clemenules ) , and Navelina ( Citrus sinesis cv. Navelina) oranges. All thr ee varieties detachment force decreased along this period. Juste et al. (1988) examined the detach ment forces of the Washington Navel ( Citrus sinesis cv. Washington Navel) and Salustiana ( Citrus sinesis cv. Salustiana) oranges. Twenty-five fruits were used in the detachment force test and it was run on three different dates over 6 weeks. The Salustiana variety’s detachment forces slightly decreased over

PAGE 36

22 this period while the Washington Navel variet y’s increased. Juste et al. (1988) also examined the amount of rotation required to detach Washington Navel and Salustiana oranges. The average amount of rotation required for Wash ington Navel and Salustiana respectively was 2.36±0.11 and 2.48±0.12 rotations with 100% of the fruit being detached prior to three rotations. Hield et al. (1967) examined the detachme nt characteristics of oranges as they pertained to the development of a mechanical harvester and abscission chemicals. Fruit with straight stems were removed from the tree with the stem attached and loaded at a constant rate by a straight pull. Tests were performed on Navel and Valencia oranges and grapefruit taken from multiple locations, multiple rootstocks, multiple tree moisture contents, and at multiple temperatures. All three citrus varieties detachment force decreased over the sampling period which ra nged from 4 to 8 months depending on the variety. Both fruit size and fruit stem size were found to directly influence detachment force with larger fruits and stems requiring larger detachment forces. It was also observed that larger fruit gene rally correlated with larger st ems. Geographic location was found to have no effect when comparing comp arable commercial maturities. Rootstock was also found to have no effect on the detach ment force except that fruit size and stem size varied among different ro otstocks. Tree moisture c ontent as varied by applying different irrigation treatments was also f ound to have no effect on detachment forces although plugging occurrence was increased under increased tree moisture content. It was also observed that pluggi ng occurrence decreased from morning to evening. This was found to be associated with increases in temperature leading to decreased plugging occurrence. Detachment force also decrea sed with increases in temperature for both

PAGE 37

23 grapefruit and Valencia orange s. Overall the fruit and stem size had a larger effect on detachment force than temperature. Kender and Hartmond (1999) examined the detachment characteristics of Hamlin, Pineapple and Valencia orange s as they pertained to the development of mechanical harvesters and abscission chemicals. Each sampled tree was divided into between 8 and 16 sectors with 20 fruit sampled from each sector. Fruit detachment force was significantly lower for all thr ee varieties in the interior of the canopy. It was observed that stem diameters were smaller in this se ctor which contributed to this observation, although fruit size was larger. Differences in the upper and lower portions of the canopy were only observed in the Hamlin vari ety with upper sec tion exhibiting higher detachment force. Valencia and Hamlin va rieties grown in north-south hedge-rows had no significant differences between the east a nd west sides whereas the Pineapple variety exhibited lower detachment for ce on the east side than the west. The Valencia variety exhibited lower detachment force on the west than the south. Detachment force was found to be highly positively correlated with fruit size. Rumsey and Barnes (1970) and Rumsey (1967) examined the detachment characteristics of Navel and Valencia orange s and Marsh grapefruit as they pertained to the development of mechanical harvesters. Fr uit were detached from the tree with a 1 cm (0.5 in.) stem still attached for testing. Tests were conducted which applied loads collinear with the stem axis and at angles of 45 and 90°. A detachment torque test was also performed in which the fruit was loaded co llinear with the stem axis at force levels of 0, 8 and 10 pounds. The torque required to detach the fruit was then recorded by loading the fruit at a 0.64 cm (0.25 in.) radi us. A sample size of 25 fruit was used for

PAGE 38

24 each test. It was observed that the detachment force substantially decreased as the pull angle increased and that detachment force decreased over the maturation period for all varieties. However for the Navel variety the detachment force continued to decrease beyond the harvest period whereas the Valencia va riety held constant. This is consistent with the observation that Navel fruit not harvested will fall off the tree whereas Valencia fruit will remain until the next harvest. It was also observed that tension levels and maturation level had little effect on the detachment torque. In summary, the previously cited studies found detachment force to be a function of each of the following: variety, fruit maturati on, loading angle, fruit size, stem size, temperature, and distance from canopy exte rior. Detachment force was found to be positively correlated with fruit size, stem si ze, and temperature. It was found to be negatively correlated with loading angle a nd distance from canopy exterior. Occurrence of peel damage during fruit removal was found to be a function of loading angle, tree moisture content, and temperature. It wa s found to be positively correlated with tree moisture content and negatively correlated with loading angle and temperature. With the exception of Fornes et al. (1993), Juste et al. (1988) and Kender and Hartmond (1999), all of the fruit detachment studies cited above were conducted with the fruit removed from the tree. Those studies that collected data w ith the fruit still attached to the tree only examined the linear detachment force or the number of rotations for detachment. It was necessary to understand more fully the relatio nship between harvesti ng motions and each of the following: fruit damage, distance travel ed and harvesting forces . Therefore one of the objectives of this study was to conduct ha rvesting motion tests in a grove that would assist in the determination of an optimum harvesting motion.

PAGE 39

25 Use of Force Feedback for Control of a Robotic Harvester In order to optimize the eff ectiveness of a robotic citrus harvesting end effector, it was decided to explore the use of force feedback for optimizing the control of the harvesting manipulator. Force sensors have been used in robotic control since 1972 (Boubekri and Chakraborty, 2002). Generally they consist of elements such as strain gauges, piezoelectric pressure sensors, resi stive pressure sensors, capacitive pressure sensors, and optical pressure sensors. Thes e transform the forces into electrical signals which can be evaluated by the controller. Force sensors have to date been used in a wide variety of applications includi ng: part mating (Chin et al., 200 3), deburring (Pires et al., 2002), minimally invasive surgery (Tegin and Wikander, 2005), and robotic fruit harvesting (Allotta et al., 1990). Force sensors can be divided into two categ ories, those that measure forces within the manipulator/end effector system and those that directly measure the forces acting on the end effector (Tegin and Wikander, 2005). One example of a sensor that measures the forces within the manipulator/end effector sy stem is a force/torque sensor. These are sensors that use a series of strain gauges mounted along the direction of the principle forces. This results in either a 3 degree of freedom (DOF) or a 6-DOF sensor. The 3DOF sensor has the ability to sense forces in the x, y, and z directions while the 6-DOF sensor also senses the correspond ing moments about the three axes. A force/torque sensor is usually mount ed between the manipulator and the end effector, however it also may be mounted on the fingertips in order to provide better tactile information (Tegin and Wikander, 2005). The advantage of this type of sensor is that it is generally more accurate than those which directly measure the forces applied. The disadvantages of this type of sensor are that it is unable to provide force

PAGE 40

26 measurements at more than one point and may gi ve errors due to inertial effects. Since there is a mass between the sensor and the a pplied force, accelerati ons by the manipulator will give force readings that include the iner tial forces of that mass. The inability to measure more than one point of contact limits its use in tactile sensing. It is mainly used when there is only one force vector of inte rest, such as in grinding or part mating. Force/torque sensors can be used with tw o different types of controllers (Perry, 2002). The first is a controller mounted with th e sensor. It is a se lf-contained unit that can communicate directly with the robot via RS-232 or analog voltages. The second type is the computer bus sensor c ontroller that connect s with the robotÂ’s motherboard. This enables the sensor to integrate more seamlessly with the robotic system. Sensors that directly measure the forces a pplied to the end eff ector usually consist of a sensor array. These usually are capa ble of measuring pressure only, although some have been developed which can measure sh ear forces as well (Tegin and Wikander, 2005). Optical or fluid filled membranes th at deform upon applied pressure provide the best spatial resolution of these sensors. Th e amount and location of the deformation is recorded through the use of photocells or cameras . Other sensor arrays consist of a series of individual pressure sensors. These are often piezoelectric, resistive, or capacitive. All three change their electrical properties under pressure. The piezoelectric sensors produce an electric charge when deformed. The re sistive and capacitive se nsors involve passing current through resistors and capacitors resp ectively which change their resistive or capacitive properties under deformation. Alternatively, sensors can be placed on th e servomotors in order to directly measure the torque delivered by the actuators and thus calculate the applied load at the

PAGE 41

27 end effector (Sciavicco and Siciliano, 2000). This more indirect method of measuring forces at the end effector often includes nonlinearities associated with the manipulator structure and so is more inaccurate. Sin ce it also involves mounting sensors along the drive mechanisms for each of the joints it is also more complicated to implement. There are two main types of force contro l that are used: through-the-arm control and around-the-arm control (Erlbacher, 2000). Th rough-the-arm control utilizes all of the joints in order to position the arm in a manne r so as to apply the desired forces. Aroundthe-arm control utilizes the arm for positioni ng only while the end effector applies the desired forces. Through-the-arm control thus involves reso lving the desired trajectory with the trajectory needed in order to generate the desired forces. This can be accomplished through the use of hybrid control. A block di agram of a typical hybrid control scheme is shown in Figure 2-25. This type of control can lead to pos itional inaccuracies which may produce excessive loads on either the manipulator or the part. In order to safeguard from these loads occurring, compliant mechanisms are used on the end effector in order to allow for a certain amount of positional inaccuracy. The compliant mechanisms are placed on the end effector as making the enti re manipulator compliant would result in greater positional inaccuracies throughout the manipulator. Be cause of the potential for these force overloads, through-the-arm control is most often used at lower velocities in more controlled situations. Around-the-arm control uses the manipulator to position the end effector within a certain range of the desired position. This is also known as impedance control. The end effector contains additional positioning capabilities which are more flexible. The most

PAGE 42

28 widely used form is pneumatics. Pneuma tics allow the end effector to adjust for positional inaccuracies due to th e compressibility of air. There are two methods of around-the-arm cont rol, passive and active. In passive control the changing weight of the end effect or is counterbalanced either mechanically using counterweights or pneum atically using a double-acting pneumatic actuator. The mechanical counterbalance shifts as the orient ation of the end effector changes. In the pneumatic system the orientation must be known either through me asuring the degree angle or use the feedback from the manipulator actuators to calculate the degree angle of the end effector. Based on the orientation of the end effector, the known weight of the end effector and the desired applied force the pneumatic actuator is adjusted accordingly. This is an open loop control structure. Active control involves using sensors such as a load cell and an accelerometer to provide feedback to the contro l structure. The accelerometer records the orientation of the end effector as well as iner tial effects from the movement of the end effector that may produce forces at the tool tip. The load cell pr ovides feedback as to the actual force being applied by the end effector. Both the active and passive systems will need periodic calibration. The load cell and accelerometer in the active system may dr ift over time and the weight of the end effector may change in the passive system. Several sensors and force control methods have thus been presented which, when implemented into a robotic control system, allo w the manipulator to better interact with its environment. The potential for use is highest in operations where the operational environment is less structured leading to higher variability in the forces that the

PAGE 43

29 manipulator will encounter. One such unstr uctured environment is that of robotic harvesting. Allotta et al. (1990) describes the use of a force-torque sensor in determining the approximate location of the stem for cutting. The location is determined by one of two methods depending on the degrees of freedom of the robot used. In the case of a 6-DOF robot, a series of motions are performed in order to reduce the reaction for ces applied by the stem to a single pulling force. The locati on of the stem is then inferred to be at the intersection of the upper hemisphere of the fruit and the line of action of the force. In the case of a 3-DOF robot, subsequent vertical and horizontal moves are performed in order to determine two planes in which the center of the fruit and the stem lie. A line common to these two planes can then be determined a nd it can then be assume d that the location of the stem is the intersection of this line and the upper hemisphere of the fruit. It was found that the method used in the case of a 6-DOF robot provided sufficient accuracy with an error of less than 15°. The algorith m took an average of 4 seconds to converge, however this time could be improved with an improved architecture and bandwidth. The method used in the case of a 3-DOF robot was faster but had s ubstantial error which sometimes exceeded 45°. The implementation of this algorithm into a robotic orange harvester is described in Muscato et al. (2005). Allotta et al. (1990) determined stem locat ion through a series of movements. In the optimization of a fruit harvesting motion, it may not be possible to perform these motions. As such it is one of the objectives of this study to determine if stem location detection may be used to optimize the harvesting motion without the need for additional movements. There has also been no attemp t to develop a control strategy that would

PAGE 44

30 modify the harvesting motion based on the sensed fruit/stem angle. It is therefore another objective of this study to develop a force contro l model that uses feedback from the force torque sensor to control a robot ic citrus harves ting manipulator. This chapter reviewed previous research th at is relevant. The next chapter details the specific objectives of the studies in this dissertation.

PAGE 45

31 Figure 2-1. Kyoto University end effector (Sarig,1993). Figure 2-2. Kubota end effector (Sarig, 1993). Figure 2-3. Instituto de Auto mática end effector harvesting motion (Ceres et al., 1998). Figure 2-4. First CRAM end eff ector (Muscato et al., 2005).

PAGE 46

32 Figure 2-5. Second CRAM end effe ctor (Muscato et al., 2005). Figure 2-6. Third CRAM end eff ector (Muscato et al., 2005). Figure 2-7. U.S. Patent 3165880 e nd effector (Buie Jr., 1965). Figure 2-8. U.S. Patent 3756001 e nd effector (Macidull, 1973).

PAGE 47

33 Figure 2-9. U.S. Patent 3854273 e nd effector (Rosenberg, 1974). Figure 2-10. U.S. Patent 4519193 end effector (Yoshida et al., 1985). Figure 2-11. U.S. Patent 4718223 e nd effector (Suzuki et al., 1988).

PAGE 48

34 Figure 2-12. U.S. Patent 5005347 end e ffector (Kedem and Rubinstein, 1991). Figure 2-13. U.S. Patent 5544474 rotating finger end effector (Finkelstein, 1996). Figure 2-14. University of Florida end eff ector harvesting sequence (Pool and Harrell, 1991).

PAGE 49

35 Figure 2-15. AID helix end effector (Muscato, 2005). Figure 2-16. U.S. Patent 3925973 he lix end effector (Glover, 1975). Figure 2-17. U.S. Patent 4674265 end effector (Gerber, 1987).

PAGE 50

36 Figure 2-18. U.S. Patent 5544474 pulli ng end effector (F inkelstein, 1996). Figure 2-19. Twisting end effector developed as a part of a join t French and Spanish research effort (Juste et al., 1992). Figure 2-20. Italian tw isting end effector (G uglielmino et al., 1996).

PAGE 51

37 Figure 2-21. U.S. Patent 4532757 end effector (Tutle, 1983). Figure 2-22. Japanese robotic tomato harv esting end effector harvesting procedure (Monta, 1998).

PAGE 52

38 A B Figure 2-23. Hungarian robotic apple harvester end effect ors (Kassay, 1992). Figure 2-24. Iwate University robotic tomato harvesting e nd effector (Kataoka et al., 1999). Figure 2-25. Hybrid control block diagram.

PAGE 53

39 Table 2-1. Reported results for prev iously developed end effectors. End Effector Figure # or Description TypeResults 2-1CuttingCould not harvest fruit with stems larger than 3.5 mm, or those with surrounding twigs and leaves 2-3CuttingGrasping and detaching cycle time was 2 sec 2-4CuttingNot constructed due to concerns over cost and robustness 2-5CuttingHigh success rate, but cost of implementation too high for commercial use 2-14PullingThe end effector successfully removed the fruit 69% of the time. Of these successful attempts, 63% of them caused damage to either the fruit or the tree. Damage to the tree accounted for 85% of this damage which usually involved the removal of leaves or small branches. The most common fruit damage was when the fruit was caught between the lip and the housing. Plugging of the fruit was rarely observed. The residual stem lengths were between 0 and 2 inches with the occasional stem greater than 3.9 inches. The effect of replacing the lip with a cutter bar was also examined, but resulted in frequent cut fruit and residual stem lengths between 0.8 to over 5.9 inches. 2-15PullingProblems with damaging the fruit and the tree as well as its inability to handle the fruit once it was picked 2-19Twistin g The detachment rate of the end effector was 64 to 67% with the failures caused by obstacles, vacuum failures, or mechanical failures. The rate of plugging of this end effector was 2 to 4%. The residual stem length was less than 5 mm in 80 to 85% of the detached fruit. The end effector did not damage the fruit through its contact with the fruit, and 4 to 5% of the detached fruit were damaged by the surrounding tree structure during removal. 2-24Twistin g The success rate of the end effector was found to be 92.3% on 26 replications. The failed attempts were due to dropped fruit. Magali end effector Twistin g The robot was successful 50% of the time with 75% of the fruits still having their stems, which is desirable.

PAGE 54

40 CHAPTER 3 OBJECTIVES Because of the motivations described previ ously, it is the purpose of the studies in this dissertation to work towards the developm ent of a robotic citrus harvester. This chapter details the specific objectives of these studies. The main objectives and subobjectives were as follows: 1. Determine the safe grasping limits for a grasping robotic citrus harvester end effector through physical properties tests. Bridge the gap in testing of the punch sizes previously used to the burst tests. Quantify the relationship between punch size and puncture force. Make recommendations on the impact of the design of a grasping robotic citrus harvester end effector. 2. Determine the most effective harvesti ng motion through tests with a 6-axis force/torque sensor. Study the influence of harvesting angle on each of the following: fruit damage, distance traveled and harvesting forces. Study the influence of a harvesting mo tion involving rotation about the stem axis on each of the following: fruit damage, distance traveled and harvesting forces. Study the influence of a secondary pitch harvesting motion on each of the following: fruit damage, distance traveled and harvesting forces. 3. Design a robotic citrus harvesting end eff ector based on the results of the physical properties and harvesting motion tests. End Effector Design Criteria Lightweight (<17.78 N (4 lb)), based on the manipulator payload capacity (155 N) and the weight of the end ef fector used in th e harvesting motion tests (24.5 N)

PAGE 55

41 Cost-effective (<$1000, not including la bor), based on economic estimates of the overall cost of an econom ically viable harvesting system Fast (<0.5 sec), based on a total cycle time of 2 sec Produce minimal damage on fruit and tree Able to reach fruit inside the canopy Able to integrate sensing technolog y such as vision, ultrasonic and infrared sensors 4. Build and test the designed robotic ha rvesting end effector to determine its effectiveness based on the design criteria 5. Determine if stem location detection ma y be used to optimize the harvesting motion without the need for additional movements. 6. Construct a force control model that woul d utilize feedback from a force/torque sensor to control a robotic citrus harvesting manipulator. This chapter has outlined the specific objectiv es of the studies in this dissertation. The next chapter provides an overview of the methods and procedures used to accomplish these objectives.

PAGE 56

42 CHAPTER 4 METHODS AND PROCEDURES This chapter provides an overview of the methods and procedures used to accomplish the previously stated objectives. It also provides an outline of the material discussed in the remaining chapters. Physical Properties of Oranges in Response to Applied Gripping Forces for Robotic Harvesting The first main objective and associated sub-objectives stated in the previous chapter deal with the safe compressive limits of handling for a grasping robotic citrus harvester end effector. These objectives were accomplished by performing puncture and burst tests using an Instron universal testing machine. Punch diameters were selected to expand the range of punches previously used, to more closely repr esent the surface area that might be encountered in robotic fingers . Burst tests were conducted using a flat plate. The fruit were supported with an alum inum die, for the punch tests and one set of burst tests. For the other set of burst test s, the fruit were deformed using two parallel plates. The burst test with the die was conduc ted in order to compare the results of the test directly with the results of the punch tests. The burst test with the parallel plates was conducted according to ASAE Standard S368.4 ( ASAE Standards , 2000). An illustration of a burst test is shown in Figure 4-1 and an illustration of a punch te st is shown in Figure 4-2. The fruit were oriented with the fruit stem parallel to the plate in order to mimic the actions of the proposed harvester. They were loaded until failure, which was defined as

PAGE 57

43 the point at which the peel was compromise d, either through puncture or bursting. This was visually determined. Once the peel had been compromised, the test data were saved for later determination of th e exact point at which the peel was punctured through the use of the force deformation curve. Studies in the Optimization of Harv esting Motion Mechanics using a 7-DOF Manipulator with a 6-Axis Force/Torque Sensor The second main objective and associated s ub-objectives stated in the previous chapter deal with determining an optimu m harvesting motion for a robotic citrus harvester. These objectives were accomplished by performing four sets of tests: a linear pull test, a retraction angle test, a rotational test and a secondary pitch test. For the linear pull test a force gauge and fruit holding strap (Figure 4-3), were used to determine the force required to harvest the fruit while pu lling along the stem axis. A surveyor's pole was also used to determine the amount of fruit deflection. In the retraction angle test a 6-axis fo rce/torque sensor was mounted on a 7-DOF manipulator (Figure 4-4). This system was used to record the reaction of the fruit/stem interface to various harvesting motions. A pne umatic end effector with pivoting gripping fingers was used to grasp the fruit. The ma nipulator was then used to perform motions which would put the fruit and stem at various angles. The force/torque sensor would then record the resultant forces and torques. In the rotational test the force/torque se nsor and manipulator were again used to record the reaction of the fruit/stem interf ace of pre-twisted fruit to various harvesting motions. The fruit was first rotated by hand about the stem axis by either 0, 0.5, 1, 1.5 or 2 rotations. The manipulator then performed a motion to impart a relative angle between the fruit and stem as in the retraction angle test.

PAGE 58

44 In the secondary pitch test the force/torque sensor and manipulator were again used to record the reaction of the fruit/stem interface to various harvesting motions. The manipulator performed an initial motion until the force/torque sensor determined that a prescribed amount of tension had develope d at which point a secondary motion was performed. After the fruit were harvested the mass was recorded using a balance. The major, minor, and intermediate diameters were taken (Figure 4-5), where a is the largest diameter, c is the smallest diameter perpendicular to a , and b is the intermediate diameter perpendicular to both a and c . The amount of damage to the fruit was determined by visually inspecting each fruit for damage a nd assigning a severity level to each fruit. Robotic Citrus Harvesting End Effector Development The third and fourth main objectives stat ed in the previous chapter deal with designing and testing a robotic citrus harvesting end effect or. These objectives were accomplished by first designing an end effect or based on the physical properties and harvesting motion tests using SolidWorks ( 2003). Once the end effector had been designed (Figure 4-6), it was assembled, construc ted and then tested in an orange grove in order to compare the act ual performance with the design criteria. Tests were performed which put the fruit and stem at vari ous angles. The ability of the end effector to harvest the fruit without damaging the fru it or tree was evaluated. Second, the ability of the end effector to use the integrated pos ition sensors to accurately control the end effector was evaluated. Third, th e ability of the end effector to harvest the fruit even if the end effector was not accurately positione d was evaluated. Lastly, the end effector actuation time, gripping force, weight and outer profile dimensi ons were recorded.

PAGE 59

45 Stem Detection Studies and Force Control Model Development The fifth main objective stated in the prev ious chapter deals with the determination of the feasibility of using st em location detection to optimi ze the harvesting motion. This was accomplished through performing a set of tests to determine the ability of the force/torque sensor to accura tely determine the angle betw een the fruit and stem. The sixth main objective stated in the previous ch apter deals with the de velopment of a force control model. A model was developed in Ma tlab (2004) that utilized the stem location detection algorithms tested as a part of the fi fth objective tests to modify the trajectory of the manipulator.

PAGE 60

46 Figure 4-1. Burst test conducte d with fruit-holding die. Figure 4-2. Punch test conducted with the 1.27 cm diameter punch. Figure 4-3. Force gauge and fruit holding stra p used in the linear pull test shown pulling a fruit on one of the test trees.

PAGE 61

47 Figure 4-4. End effector, force /torque sensor and robotic mani pulator used in testing are shown reaching towards a fruit on one of the test trees. Figure 4-5. Characteristic di ameters taken of fruit. Figure 4-6. Designed end effector.

PAGE 62

48 CHAPTER 5 PHYSICAL PROPERTIES OF ORANGES IN RESPONSE TO APPLIED GRIPPING FORCES FOR ROBOTIC HARVESTING Introduction Because of rising costs and a shrinking labor force, it is increasingly desirable to harvest citrus fruit robotically. Currently, re searchers at the University of Florida are working to develop such a harvester. It is important to determine the safe handling limits for citrus fruit during harvest. Specifically, this chapter examines the result of compressive force testing on oranges us ing both punch tests and burst tests. Several studies have been published that examine the resistance of the rind to puncture. Ahmed et al. (1973) examined the forc es required to rupture peel oil glands and to puncture the fruit rind. This was done for field run (unwashed, unwaxed), commercially processed, and irradiated fru it. It was observed that both commercial processing and irradiation had a detrimental e ffect on the structural strength, although the differences between commercially processed a nd field run fruit were not statistically significant. Twenty fruit were used per replicate, with th ree replicates performed. The punch diameters used were: 0.258, 0.264, and 0.051 cm. Churchill et al. (1980) examined the fo rces that fruit might experience during mechanical harvesting. Both puncture and burst tests were performed. The influence of variety, harvest date, harvest time, and absc ission chemical application were examined. The sample size was 20, with two to four repli cations per year. It was observed that there were varietal differences, with the Pineapple ( Citrus sinesis cv. Pineapple) variety having

PAGE 63

49 a greater structural strength than the Hamlin ( Citrus sinesis cv. Hamlin) variety. The harvest date and time did not have statisti cally significant differences, although the Pineapple variety required incr easing puncture and burst forces in later harvest dates. The punch diameter used was 0.64 cm. Coggins Jr. and Lewis (1965) examined the ch ange of puncture forces over time as well as the effect of gibbere llic acid treatments. A compar ison of compressive forces versus shearing forces was also done. They observed that for two different penetration sizes with area ratios of 2 and circumferential ratios of 1.41, a puncture force ratio (larger size puncture force over smaller size puncture force) of 1.56 was required, indicating that both compressive and shearing forces were pres ent. However, shearing forces were more dominant. The gibberellic acid treatment, if applied early enough in the growth process, had the effect of slowing the rate of decrease in puncture resistance. McDonald et al. (1987) found supporting results. They observe d that the puncture resistance decreased over time. Twenty fruit were used, with ten punctures per fruit. The punch diameters used were 0.10 and 0.1438 cm. Juste et al. (1988) examined citrus fruit prop erties as they pertained to fresh fruit robotic harvesting. In each puncture test, 50 fruit were used with a punch diameter of 0.047 cm. These tests were performed at four different times during the growth season. They found that puncture resi stance had a decreasing trend, except for the beginning and end of the growth seasons, where punctur e resistance might increase depending on the variety. Miller (1986) examined dimensional prope rties and puncture resistance, along with modulus of elasticity an d average stress. In each test, 40 fruit were measured for their

PAGE 64

50 dimensional properties, and 20 fruit were m easured for puncture resistance. Results from that work suggested that maximum compressive strength was related to the peel strength, whereas the deformation was related to intern al structural integrit y. The punch diameter used was 0.64 cm. Other punch diameters previously used we re: 0.5 and 1 cm (Chuma et al., 1978), 0.10 cm (Coggins Jr., 1969), 0.05 cm (Fidelibus et al., 2002a), 0.1 cm (McDonald et al., 1987), and 0.320 cm (Turrell et al., 1964). The aforementioned studies have an equivalent punch diameter range of 0.05 to 1 cm. Separa te burst tests were also performed by Ahmed et al. (1973), Chuma et al. (1978), C hurchill et al. (1980), Fidelibus et al. (2002b), Miller (1986), and Sarig and Orlovs ky (1974). These tests we re conducted in the same manner as the punch tests, except th e punch was replaced with a flat plate. The studies cited above were conducted on a wide range of citrus varieties, which included: Valencia ( Citrus sinesis cv. Valencia), Hamli n, Pineapple, Satsuma ( Citrus reticulata cv. Satsuma ) , Washington Navel ( Citrus sinesis cv. Washington Navel), Shamouti ( Citrus sinesis cv. Shamouti), Salustiana ( Citrus sinesis cv. Salustiana), and Temple ( Citrus sinesis cv. Temple) oranges; Duncan ( Citrus paradisi cv. Duncan) and Marsh ( Citrus paradisi cv. Marsh) grapefruit; Bearss ( Citrus limon cv. Bearss) and Eureka ( Citrus limon cv. Eureka) lemons; Persian limes ( Citrus latifolia cv. Persian); and Dancy tangerines ( Citrus reticulata cv. Dancy). The variety studied most often was the Valencia orange. These studies used punch sizes that we re well below the size that might be experienced by a fruit under robotic harvesti ng. The largest punch size studied was 1 cm. It was necessary to understand more fully how a fruit would respond to a robotic

PAGE 65

51 harvester by conducting studies that bridge the gap betw een the previously conducted puncture studies and the previously conducted burst studies. Therefor e the purpose of this study was to develop a relations hip between punch size and punc ture force that would be applicable to the developmen t of a robotic harvester. Objectives Specific objectives of the work reported in this chapter were to: Bridge the gap in testing of the punch sizes previously used to the burst tests. Quantify the relationship between punch size and puncture force. Make recommendations on the impact of the design of a grasping robotic citrus harvester end effector. Materials and Methods All fruit were of the Valencia variety a nd were obtained from the University of Florida's Citrus Research and Education Cent er in Lake Alfred, Florida. The Valencia variety was the only variety te sted in order to focus the studies on one of the largest varieties potentially affected by the development of a robotic ci trus harvester. The fruit were freshly harvested, and thus they ha d not been washed, waxed, or sized. After harvest, the fruit were placed in an environm ental chamber that was held at 4°C and 78% relative humidity through the us e of an automatic controller. These conditions were as close to the USDA-recommended storage cond itions as the chamber permitted. Ritenour (2004) lists the USDA-recommende d storage conditions as 0°C to 1°C and 85% to 90% relative humidity. The fruit remained in the chamber until they were tested, the total duration of which was 1 week or less, except in one instance where fruit remained in the chamber for 3 weeks. This occurred on the first test date with the para llel plate burst test.

PAGE 66

52 Ritenour (2004) states that fr uit may be stored up to 12 weeks at the specified optimum storage conditions. Samples from the three different dates for th e puncture tests were determined not to be statistically different at the 95% significance level from analysis of variance (PROC GLM; SAS, 2004). However, the burst test samp les were statistically different at the 95% significance level, with the burst values from the second date statistically lower than those from the other two dates. Further expl anation of this can be found in the Results and Discussion section. Three sets of tests were run: the fi rst began on 30 March, the second on 15 May, and the third on 16 June 2004. This was done not only to provide several replications of the experiments, but also to examine the change in puncture fo rce over the growing season. The major, minor, and intermediate diameters were taken (Figure 5-1), where a is the largest diameter, c is the smallest diameter perpendicular to a , and b is the intermediate diameter perpendicular to both a and c . The radius of curvature was measured according to the procedure set forth in ASAE Standard S368.4 ( ASAE Standards , 2000). The mass was recorded using an Ohaus (Pine Brook, NJ) model AR5120 balance with 0.01 g resolution. The sample size for each punch size was approximately 30 fruit. The sample size chosen was based on the sample size range of 20 to 50 fruit used in the previously cited studi es. In one case of the puncture test and one case of the burst test, the sample size was 29 fruit because of data corruption. In another burst test, only 25 fruit were used because of a lack of fruit. Puncture and burst tests were carried out using an Instron univ ersal testing machine (model 5566) with a 10 kN load cell on the cr osshead, following the procedure set forth

PAGE 67

53 in ASAE Standard S368.4 ( ASAE Standards , 2000). The setups for burst and punch testing are shown in Figures 5-2 and 5-3, respectively. The machine was set at 1.9 cm/min crosshead speed and 0.04 N or 0.03 cm capture interval. Punch diameters were 0.323, 0.632, 0.964, 1.27, 1.90, and 2.540 cm. These punch diameters were selected to expand the range of punches previously used, to more closely repr esent the surface area that might be encountered in robotic finge rs. Burst tests were conducted using a 10.16 cm diameter plate. The fruit were supported with an aluminum die, which had a radius of curvature of 4.45 cm for the punch tests and one set of burst tests. For the other set of burst tests, the fruit were deformed using tw o parallel plates. The bur st test with the die was conducted in order to compare the results of the test directly wi th the results of the punch tests. The burst test with the parall el plates was conducted according to ASAE Standard S368.4 ( ASAE Standards , 2000). The fruit were oriented with the fruit stem parallel to the plate in order to mimic the actions of the proposed harvester. They were loaded until failure, which was defined as the point at which the peel was compromise d, either through puncture or bursting. This was visually determined. Once the peel had b een compromised, the test data were saved for later determination of th e exact point at which the peel was punctured through the use of the force deformation curve. Once the data were acquired, Matlab (2004) was used to reduce the data collected from a single test to just the puncture force from all of the data points taken by the Instron, as well as to graphica lly analyze the testing result s. Microsoft Excel (2003) was then used to graphically analyze how all of the different tested fruit compared to each other and to determine potential trends and correlations. The results were then analyzed

PAGE 68

54 statistically at the 95% significance level us ing SAS (2004) to determine correlations and develop statistical models. Results and Discussion A summary of the dimensional data for the fr uit used in the puncture tests is given in Table 5-1. The results of the dimensional measurements are similar to that reported by Miller (1986). Thus, it is hypot hesized that the fruits used were characteristic of the variety tested. An example force vs. deformation plot is shown in Figure 5-4. A yield stress point is not readily apparent, as the deformation curve is linear up until the point of failure. Some of the deformation curves had slight variances in the linear curve prior to the failure point, which might indicate a potential yield point. An example of this is shown in Figure 5-4. This may indicate when the peel of the fruit was initially penetrated but not completely punctured. Studies were conducted in order to determine the existence of a yield point and are presented in appendix A. As shown in this figure, load was not initially present. The punch was initially placed just above the fruit, and so a slight delay was present between when the punch initia lly began extending and the fruit began experiencing loading. The results from all te st dates are given in Table 5-2, where the force, pressure, and deformation values ar e at the puncture poin t. The amount of deformation at the puncture point was take n as the difference between the recorded deformation at the puncture point and the reco rded deformation at 2.22 N of loading. This was done in order to account for deformati on under the initial settling period when the punch was first engaging the fruit. The results of the puncture tests were anal yzed using SAS (2004) with analysis of variance (PROC GLM). Results showed that th e samples from the three different dates

PAGE 69

55 were not statistically different at the 95% significance level, as the p-value for this test was 0.148. This meant that the decreasing punc ture force trend that was observed by Coggins Jr. and Lewis (1965), Ju ste et al. (1988), and McDona ld et al. (1987) was not observed in this study. In order to account fo r the effects of increasing variability with increasing punch size, a model was developed with the respon se being the natural log of the puncture force (lnpf). The natural log was also taken of the punch diameter (lnpd), thus generating a log-log re lationship between the two. Regression analysis (PROC REG) was used to determine the best combination of variables in order to predict the response. The variables considered were average diameter, radius of curvature, mass, and lnpd. Since the average diameter, mass, and radius of curvature are all hi ghly correlated, it was necessary to include only one in the model. It was determined that radius of cu rvature in combination with lnpd yielded the highest r2 value, and thus the best model. The ln pd term was the predominant contributor to the model, as the r2 value using that term alone wa s 0.923. The addition of the radius of curvature term increased the r2 value to 0.924. The correlation coefficient between lnpf and the radius of curvature was very low (Pearson coefficient = 0.084, Spearman coefficient = 0.103), but it explai ned enough of the variability in the response to be able to better fit the model when it was combined with the lnpf term. The model was verified to have constant variance and normality thr ough analysis of the residuals. The resultant model has the form: lnpf = K1 * lnpd + K2 * rc + C (Eq. 5-1) where rc = radius of curvature, C = 3.977, K1 = 0.999, K2 = 0.078.

PAGE 70

56 Using the same procedure as was reported earlier by Coggins Jr. and Lewis (1965), the increase in puncture force was observed to be correla ted more closely with punch diameter than area, as puncture force was a linear function of punch diameter. An illustration of the similarity in the relationships is shown in Figure 5-5. Ratios of the value at each punch diameter over the value at the smallest punch diameter were taken in order to obtain similar units. This indicates th at the contact perimeter associated with the shearing forces influenced the puncture force more than the contact area. A summary of the dimensional data for the fr uit used in the burst tests is given in Table 5-3. The results of the dimensional measurements were similar to those reported by Miller (1986). This supported th e hypothesis that the fruits us ed were characteristic of the variety tested. Burst tests produced force vs. deformation cu rves similar to that shown in Figure 54. The results of the burst tests performed w ith and without the holding die are shown in Table 5-4. The bursting test results without the holding die were observed to be significantly lower at the 95% significance level. This most likely reflected the contribution of the holding die to preventing th e fruit from deforming as quickly as when the die was not present. Failure during burst tests was a result of the fruit diameter parallel to the loading plate expanding beyond the limits of th e fruit strength. This caused the outer skin to split (Figure 5-2). It was determined that the results from the three different dates were shown to be statistically different using analysis of variance (PROC GLM), with the burst values from the second da te statistically lower than those from the other two dates.

PAGE 71

57 The fruit tested during the second date were also observed to be smaller than the other two dates for both the puncture and burst tests, as shown in Tables 5-1 and 5-3. However, there was not as large a difference in the size of fruit used in the puncture tests among the three test dates as there was in the burst tests, even though the fruit were randomly selected from the same group. The sma ller size difference in the puncture tests, coupled with the dominance of the punch diamet er term in the model, made date an insignificant parameter for the puncture te sts. Conversely, the absence of the punch diameter term, coupled with the larger size difference among dates, made date a significant parameter in the burst tests. Since th e smaller fruit used in the second test date were still within the expected range for the Valencia variety, the values obtained from this test were included in th e results presented in Table 5-4. Conclusions As expected, the force required to puncture or burst a fruit is directly related to the contact area. This is a function of two variab les: the punch diameter used, and the radius of curvature of the fruit. The larger the radius of curvature, and thus the flatter the fruit at the point of contact, the more fruit will be in immediate contact with the punch. This results in a larger puncture force, which implies that the fruit can withstand higher contact forces when using larger punch sizes, as would be expected. Most of the model is described by the punch diameter term, as the size of the punch is the variable with the greatest influence on the puncture force. As th e punch diameter size increases, the punch diameter approaches one, and lnpd goes to zero. This left lnpf as a function of the radius of curvature only. This correlated well w ith physical observations, in that punch diameters beyond 2.540 cm approached the behavior of a flat plate, where puncture force is no longer a function of punch diameter but solely of the fruit properties. Principle

PAGE 72

58 among the determining fruit properties was the radius of curvature. The model, however, did not account for this plat eau. Therefore, it was only va lid for punches where failure was at the perimeter of the punch. The limits on overall loading force were determined in the burst tests and were dependent on how much the fruit was allowed to deform as it was loaded. Based on the results of these tests, reco mmendations can be made for the design of a grasping robotic citrus harves ter end effector. The end effector should be made so that the grasping of the fruit does not exceed th e bursting limits or the puncture limits, where the portion of the end effector in contact with the fruit may be expressed in equivalent punch diameter. This may be obt ained by using the perimeter th at is in contact with the fruit. As discussed previously, the contact perimeter influences the puncture force more than the contact area. The area is associated with preventing the fruit from deforming under applied loading, which would correlate mo re with determining the bursting limits. Further studies need to be conducted using va rious shaped punches in order to further define to what extent each of these geomet rical properties plays a role in the puncturing of the fruit, and the impact of these role s on the design of a grasping robotic citrus harvester end effector.

PAGE 73

59 Figure 5-1. Characteristic di ameters taken of fruit. Figure 5-2. Burst test conducte d with fruit-holding die. Figure 5-3. Punch test conducted with the 1.27 cm diameter punch.

PAGE 74

60 Figure 5-4. Sample force vs. deformation curv e from the 16 June set of tests, using the 1.27 cm diameter punch. 0 1 2 3 4 5 6 7 8 9 0 0.5 1 1.5 2 2.5 3 Punch Diameter (cm)Ratio Puncture force ratio Circumference ratio Figure 5-5. Ratio of increase in va lue from value at smallest punch. Table 5-1. Dimensional data summary of fruit used in puncture tests. Mass (g) Average Diameter (cm) Radius of Curvature (cm) Test Date (2004) Number o f Fruit Tested Mean Std. Dev.MeanStd. Dev.Mean Std. Dev. 30 March 180 203.0333.79 7.2060.463 3.736 0.269 15 May 180 176.0428.05 6.8900.400 3.478 0.223 16 June 179 205.5635.09 7.3290.461 3.710 0.300

PAGE 75

61 Table 5-2. Summary of puncture test results. Puncture Force (N)[a]Strength (N/cm2)[b]Deformation (cm)[c] Punch Diameter (cm) Test Date (2004) Mean Std. Dev.Mean Std. Dev.Mean Std. Dev. 0.323 30 March 21.68 3.15 264.58 38.42 0.504 0.068 0.323 15 May 22.31 4.43 272.23 54.07 0.558 0.119 0.323 16 June 25.09 3.37 306.16 41.12 0.586 0.077 0.632 30 March 51.47 9.68 164.07 30.85 0.902 0.114 0.632 15 May 43.90 9.61 139.93 30.62 0.876 0.133 0.632 16 June 49.57 8.18 158.02 26.07 0.978 0.145 0.964 30 March 71.66 11.73 98.19 16.07 1.060 0.148 0.964 15 May 65.81 13.16 90.17 18.03 1.158 0.227 0.964[d] 16 June 69.30 11.23 94.94 15.39 1.114 0.187 1.27 30 March 93.61 16.24 73.90 12.82 1.223 0.178 1.27 15 May 78.45 17.10 61.93 13.50 1.156 0.180 1.27 16 June 84.64 14.33 66.82 11.31 1.277 0.224 1.9 30 March 138.69 18.07 48.92 6.37 1.469 0.174 1.9 15 May 125.98 24.97 44.43 8.81 1.520 0.177 1.9 16 June 142.65 18.41 50.31 6.49 1.714 0.188 2.54 30 March 209.77 29.47 41.40 5.82 1.818 0.166 2.54 15 May 166.09 46.67 32.78 9.21 1.711 0.294 2.54 16 June 202.78 40.91 40.02 8.07 1.936 0.272 [a] Force is the load recorded at puncture. [b] Strength is calculated from th e puncture force and punch diameter. [c] Deformation is the difference in reco rded extension of the punch between 2.2 N load and the puncture point. [d] Sample size was 29. Table 5-3. Dimensional data summar y of fruit used in burst tests. Mass (g) Average Diameter (cm) Radius of Curvature (cm) Test Date (2004) Number o f Fruit Tested MeanStd. Dev.MeanStd. Dev.MeanStd. Dev. 30 March 60 197.3640.87 7.0970.530 3.7210.315 15 May 55 179.8132.71 6.9640.448 3.4660.231 16 June 59 199.5028.54 7.2850.374 3.6280.217

PAGE 76

62 Table 5-4. Summary of burst test results. Force w/ Die (N)[a] Force w/o Die (N)[a] Deformation w/ Die (cm)[b] Deformation w/o Die (cm)[b] Test Date (2004) Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. 30 March 403.70 94.69 317.52 64.61 1.823 0.240 2.412 0.330 15 May 289.04 59.64 221.06[c]43.02[c]1.697 0.199 2.141[c] 0.326[c]16 June 459.00[d] 92.13[d]336.40 53.81 1.928[d]0.193[d]2.576 0.305 [a] Force is the load recorded at the burst point. [b] Deformation is the difference in record ed extension of the plate between 2.2 N load and the burst point. [c] Sample size was 25. [d] Sample size was 29.

PAGE 77

63 CHAPTER 6 OPTIMIZATION OF HARVESTING MO TION MECHANICS USING A 7-DOF MANIPULATOR WITH A 6-AXI S FORCE/TORQUE SENSOR Introduction As a result of rising cost and a shrinking labor force, it is increasingly desirable to harvest citrus fruit robotically. Currently re searchers at the University of Florida are working to develop such a harvester. It is important to determine the forces required to harvest the fruit without damaging the fruit or tree. Specifically th is chapter examines the forces generated at the fruit-stem inte rface by various harvesting motions, the travel distance required to perform these motions a nd the impact of these motions on the fruit and the tree. Barnes (1969) studied the detachment char acteristics of lemons as they may pertain to the development of a mechanical harvester. In this study fruit we re separated from the tree with the stem intact. Tests were then performed by applying forces collinear to the stem as well as at a 45° angle to the stem. A torsion test was also performed by applying loads at a 5 cm (2 in.) radius from the base of the calyx. In all thr ee tests the load which caused failure was recorded as well as the point at which failure occurred along the fruitstem interface. These three tests were perf ormed on 4 different dates using the fruit from 10 different trees with 10 fruit of varying di ameters for each test. Separate collections were performed for each side of each of the trees. This was done for three different locations. Two of these were located in Ca lifornia and the third was in Arizona. The Eureka ( Citrus limon cv. Eureka) lemon was present at the California locations while the

PAGE 78

64 Lisbon ( Citrus limon cv. Lisbon) lemon was present at the Arizona location. The results of all three tests indicated that the detachment force was directly correlated to the fruit size with the larger fruit requiring higher de tachment forces. There was no statistical difference among observed detachment forces on different sides of the tree or among trees at a specified location. The most common form of detachment was separation of the calyx. The force required for detachment in th e angle-pull tests was less than half that required for the straight pull. In the to rsion test, there was however a noticeable difference between the detachment torques fo r the Lisbon and Eureka varieties with the Lisbon variety requiring consistently higher torques than the Eureka lemon. Coppock et al. (1969) studied the detach ment characteristics of Valencia ( Citrus sinesis cv. Valencia), Hamlin ( Citrus sinesis cv. Hamlin), and Pineapple ( Citrus sinesis cv. Pineapple) oranges and Marsh ( Citrus paradisi cv. Marsh) grapefruit as they may pertain to the development of a mechanical ha rvester. In this study the detachment force at 0, 45, and 90° angles from the fruit’s axis was determined as well as the form of detachment and the possible correlations to other citrus propertie s. Three 40 fruit samples were taken at four dates from four tr ees. The fruit were detached with the stems intact with a length of 10 to 15 cm (4 to 6 in.). The loading rate was approximately 44 N/s (10 lb/s). Detachment force at all th ree angles decreased over the harvest season except for the Valencia variety which increa sed. Detachment force and percentage of plugging (where part of the peel is removed w ith the stem) decreased for all varieties as loading angle increased. Fornes et al. (1993) and Just e et al. (1988) examined th e detachment characteristics of oranges as they pertained to the deve lopment of a robotic citrus harvesting end

PAGE 79

65 effector. Fornes et al. (1993) examined the detachment force at the beginning, middle, and end of their maturation period of Clausellina ( Citrus reticulata cv. Clausellina ) , Clemenules ( Citrus reticulata cv. Clemenules ) , and Navelina ( Citrus sinesis cv. Navelina) oranges. All thr ee varieties detachment force decreased along this period. Juste et al. (1988) examined the detach ment forces of the Washington Navel ( Citrus sinesis cv. Washington Navel) and Salustiana ( Citrus sinesis cv. Salustiana) oranges. Twenty-five fruits were used in the detachment force test and it was run on three different dates over 6 weeks. The Salustiana variety’s detachment forces slig htly decreased over this period while the Washington Navel variet y’s increased. Juste et al. (1988) also examined the amount of rotation required to detach Washington Navel and Salustiana oranges. The average amount of rotation required for Wash ington Navel and Salustiana respectively was 2.36±0.11 and 2.48±0.12 rotations with 100% of the fruit being detached prior to three rotations. Hield et al. (1967) examined the detachme nt characteristics of oranges as they pertained to the development of a mechanical harvester and abscission chemicals. Fruit with straight stems were removed from the tree with the stem attached and loaded at a constant rate by a straight pull. Tests were performed on Navel and Valencia oranges and grapefruit taken from multiple locations, multiple rootstocks, multiple tree moisture contents, and at multiple temperatures. All three citrus varieties detachment force decreased over the sampling period which ra nged from 4 to 8 months depending on the variety. Both fruit size and fruit stem size were found to directly influence detachment force with larger fruits and stems requiring larger detachment forces. It was also observed that larger fruit gene rally correlated with larger st ems. Geographic location was

PAGE 80

66 found to have no effect when comparing comp arable commercial maturities. Rootstock was also found to have no effect on the detach ment force except that fruit size and stem size varied among different ro otstocks. Tree moisture c ontent as varied by applying different irrigation treatments was also f ound to have no effect on detachment forces although plugging occurrence was increased under increased tree moisture content. It was also observed that pluggi ng occurrence decreased from morning to evening. This was found to be associated with increases in temperature leading to decreased plugging occurrence. Detachment force also decrea sed with increases in temperature for both grapefruit and Valencia orange s. Overall the fruit and stem size had a larger effect on detachment force than temperature. Kender and Hartmond (1999) examined the detachment characteristics of Hamlin, Pineapple and Valencia orange s as they pertained to the development of mechanical harvesters and abscission chemicals. Each sampled tree was divided into between 8 and 16 sectors with 20 fruit sampled from each sector. Fruit detachment force was significantly lower for all thr ee varieties in the interior of the canopy. It was observed that stem diameters were smaller in this se ctor which contributed to this observation, although fruit size was larger. Differences in the upper and lower portions of the canopy were only observed in the Hamlin vari ety with upper sec tion exhibiting higher detachment force. Valencia and Hamlin va rieties grown in north-south hedge-rows had no significant differences between the east a nd west sides whereas the Pineapple variety exhibited lower detachment for ce on the east side than the west. The Valencia variety exhibited lower detachment force on the west than the south. Detachment force was found to be highly positively correlated with fruit size.

PAGE 81

67 Rumsey and Barnes (1970) and Rumsey (1967) examined the detachment characteristics of Navel and Valencia orange s and Marsh grapefruit as they pertained to the development of mechanical harvesters. Fr uit were detached from the tree with a 1 cm (0.5 in.) stem still attached for testing. Tests were conducted which applied loads collinear with the stem axis and at angles of 45 and 90°. A detachment torque test was also performed in which the fruit was loaded co llinear with the stem axis at force levels of 0, 8 and 10 pounds. The torque required to detach the fruit was then recorded by loading the fruit at a 0.64 cm (0.25 in.) radi us. A sample size of 25 fruit was used for each test. It was observed that the detachment force substantially decreased as the pull angle increased and that detachment force decreased over the maturation period for all varieties. However for the Navel variety the detachment force continued to decrease beyond the harvest period whereas the Valencia va riety held constant. This is consistent with the observation that Navel fruit not harvested will fall off the tree whereas Valencia fruit will remain until the next harvest. It was also observed that tension levels and maturation level had little effect on the detachment torque. In summary, the previously cited studies found detachment force to be a function of each of the following: variety, fruit maturati on, loading angle, fruit size, stem size, temperature, and distance from canopy exte rior. Detachment force was found to be positively correlated with fruit size, stem si ze, and temperature. It was found to be negatively correlated with loading angle a nd distance from canopy exterior. Occurrence of peel damage during fruit removal was found to be a function of loading angle, tree moisture content, and temperature. It wa s found to be positively correlated with tree moisture content and negatively correlated with loading angle and temperature. With the

PAGE 82

68 exception of Fornes et al. (1993), Juste et al. (1988) and Kender and Hartmond (1999), all of the fruit detachment studies cited above were conducted with the fruit removed from the tree. Those studies that collected data w ith the fruit still attached to the tree only examined the linear detachment force or the number of rotations for detachment. It was necessary to understand more fully the rela tionship between harvesting motions and each of the following: fruit damage, distance trav eled and harvesting forces. Therefore the purpose of this study was to conduct harvesting motion tests in a grove that would assist in the determination of an optimum harvesting motion. Objectives Specific objectives of the work reported in this chapter were to: Study the influence of harvesting angle on each of the following: fruit damage, distance traveled and harvesting forces. Study the influence of a harvesting motion i nvolving rotation about the stem axis on each of the following: fruit damage, distance traveled and harvesting forces. Study the influence of a secondary pi tch harvesting moti on on each of the following: fruit damage, distance traveled and harvesting forces. Materials and Methods The harvesting tests were performed on Va lencia variety trees in a commercial grove located in Umatilla, FL. The trees we re arranged in east-west hedge rows. Four sets of tests were performed: a linear pull test, a retraction an gle test, a rotational test and a secondary pitch test. For the linear pull test a Dillon Quantrol (Fairmont, MN) 200 N (45 lb) force gauge and fruit holding strap (Fi gure 6-1), were used to determine the force required to harvest the fruit while pulling along the stem axis. The linear pull test was performed on 9 March 2006, 30 March, 12 April, 27 April, 17 May, 7 June, 6 July and 27 July. Five fruit were used per tree with a sa mple size of ten trees for all but the 27 July

PAGE 83

69 test where because of one tree's abnormally small and dehydrated fruit only nine trees were able to be tested. A lthough the 9 March test was perf ormed on different trees than the other tests, they were in the same block with identical rootstoc k and scion. All fruit were taken from the east and west sides of the trees. On July 27 the initial height of the fruit and the height at detachment were measured using a 3.66 m surveyor's pole with 3.05 cm graduations. The difference of these two measurements was taken in order to determine the amount of deflection for each fruit. In the retraction angle test an ATI Indu strial Automation (Apex, NC) Mini45 6-axis force/torque sensor was mounted on a Robo tics Research (Cincinnati, Ohio) model 1207 7-DOF manipulator (Figure 6-2) . This system was used to record the reaction of the fruit/stem interface to various harvesting motions. The sensor had an uncertainty of 2.67 N in the vertical direction, 2.00 N in the hor izontal direction and 4.00 N along the axis of the end effector. A Schunk (Lauffen/Neck ar, Germany) model DPZ80 pneumatic end effector with pivoting grippi ng fingers was used to grasp the fruit. The network of communication and pneumatic actuation lines for the harvesting system is shown in Figure 6-3. The end effector grasped the fruit and was pitched back by 20°. The manipulator was then moved back and down along a direction of either 50, 65, 80, or 110° from vertical in the end effector's coordi nate frame. This put the stem axis of the fruit at a range of angles between 35 and 135° which was dependent on the original pitch of the end effector. The code used to contro l these motions is described in Appendix B. The code used to collect the force/torque da ta is described in Appendix C. Three fruit were used per tree per angle with a sample size of ten trees. All fruit were taken from the south side of the trees. The tests were performed on 27 to 28 April 2006.

PAGE 84

70 In the rotational test the force/torque se nsor and manipulator were again used to record the reaction of the fruit/stem interf ace of pre-twisted fruit to various harvesting motions. A variation on the code describe d in Appendix B was used to control the motions and the same code described in Appe ndix C was used to collect the force/torque data. The fruit was first rotated by hand about the stem axis by either 0, 0.5, 1, 1.5 or 2 rotations. No more than two rotations were used as it was obser ved that a significant amount of fruit were harvested with additi onal rotation. This observation agrees with Juste et al. (1988). The end e ffector then grasped the fruit and was pitched back by 20°. The manipulator was then moved back and down along a direction of 90° from vertical in the end effector's coordinate frame. Unlike in the retraction angle test, the end effector's pitch was controlled to initially be zero so as to perform the same angle each time. The sum of these angles put the stem axis of the fruit at an angle of 90°. Three fruit were used per tree per rotation per angle with a sa mple size of three trees. All fruit were taken from the north side of the trees. Th e tests were performed on 27 July 2006. In the secondary pitch test the force/torque sensor and manipulator were again used to determine when a prescribed amount of tension had developed in the fruit/stem interface during harvesting at which point a secondary motion would be performed. A variation on the code described in Appendix B was used to control the motions and a variation on the code described in Appendix C was used to coll ect the force/torque data. The end effector grasped the fruit and was pitched back by 20°. The manipulator was then moved back and down along a direction of 80° from vertical in the end effector's coordinate frame. The sum of these angles pu t the stem axis of the fruit at an angle of 80°. As in the rotational test the end effect or's pitch was controlled to initially be zero

PAGE 85

71 each time. Once 22.24 N (5 lb) of tension had been developed a secondary pitch back by 30° was performed. The sum of these angles th en put the stem axis of the fruit at an angle of 110°. Eleven fruit were harvested with all fruit coming from the same tree. All fruit were taken from the north side of th e tree. The tests were performed on 17 May 2006. After the fruit were harvested the mass was recorded using an Ohaus (Pine Brook, NJ) model AR5120 balance with 0.01 g resolution. The major, minor, and intermediate diameters were taken (Figure 6-4), where a is the largest diameter, c is the smallest diameter perpendicular to a , and b is the intermediate diameter perpendicular to both a and c . The amount of damage to the fruit wa s determined by visually inspecting each fruit for damage and assigning a severity level to each fruit. The damage levels were: no stem/no damage, stem, slight damage, dama ge, significant damage. The stem category indicates a portion of the stem was still att ached to the fruit. The damage categories occurred when the fruit peel was separated from the fruit along with the stem. The slight damage category includes all fruit with an obs ervable tear in the peel approximately less than 4 mm2. The damage category includes all fruit with a tear in the peel between 4 and 100 mm2. The large damage category includes all frui t with a tear in the peel larger than 100 mm2. No stem/No damage indicates a clean se paration of the stem from the fruit. Once the data from all tests were acquire d, Matlab (2004) was us ed to extract the detachment force and detachment time from the force/torque sensor data. Microsoft Excel (2003) was then used to graphically analyze the results. Results and Discussion Figures 6-5 and 6-6 illustrate the differenc es in fruit mass and diameter respectively among test dates for the linear pull test. Fi gure 6-7 illustrates the differences in linear

PAGE 86

72 detachment force over the test dates. The increase in detachment force agrees with Coppock et al. (1969), but disa grees with Hield et al. ( 1967), and Rumsey and Barnes (1970) and Rumsey (1967). The increase in detachment force agrees with Barnes (1969), Hield et al. (1967), and Ke nder and Hartmond (1999) when taking into account the positive correlation between fruit size and detachment force and the observed increase in fruit size over the testing dates as indicat ed by mass and average diameter. This is characteristic of the Valencia variety in that fruit not harveste d during the harvesting season will remain on the tree until next ye ar, thus the natural abscission process is slightly reversed. Table 6-1 indicates the num ber of fruit observed to be in each damage category at each test date for the linear pull test. Figure 6-8 illustrates the percentage of fruit in each damage category at each test da te. It can be observed that the largest number of no stem/no damage fruit occurred during the late April and early May test dates, which coincide to the traditional occurr ence of maximal fruit abscission. The average displacement for the 27 July fr uit was 24 cm with a standard deviation of 13 cm. There also appears to be a positive correlation between detachment force and displacement (Figure 6-9). The fruit were grouped into 20 N increments in order to account for the variability. This indicates th at the added tension necessary to develop higher detachment forces is di rectly related to th e amount that the fruit is deflected from its initial position. Therefore a motion which would re quire lower harvesting forces should also require less deflection and thus take less time to complete. Figure 6-10 illustrates the relationship between fruit/stem angle and detachment force as determined in the retraction angle te st. Although there is si gnificant va riability in the data, an apparent trend is presen t indicating a negative correlation between

PAGE 87

73 fruit/stem angle and detachment force. This agrees with Barnes (1969), Coppock et al. (1969), Rumsey and Barnes (1970), and Rumsey (1967) w ho all found that increased fruit/stem angle resulted in a lower detachment force. These tests were performed during the natural abscission period as was illustrated in Figure 67. The variability in the degree of abscission of each fruit may have contributed to the observed variability in detachment force. Figure 6-11 illustrates a less pronounced ne gative correlation between fruit/stem angle and distance traveled. Distance travel ed was determined by determining the linear distance between the start of the harvesti ng motion and the point where the largest harvesting forces were experienced. Since th e distance traveled is a function of both the amount of detachment force necessary as we ll as how those forces were developed, it cannot be said that it is strictly a function of fruit/stem angle. The manipulator was moving at different trajectories for each a ngle which may have effected how quickly the necessary detachment forces were deve loped. The correlation between reduced detachment forces and reduced distance traveled agrees with the resu lts of Figure 6-9 that indicated a positive correlation between the two. Table 6-2 indicates the number of fruit obs erved to be in each damage category for four ranges of fruit/stem angles. Figure 6-12 illustrates the percentage of fruit in each damage category for the same four ranges of fruit/stem angles. This indicates a strong positive correlation between the fruit/stem angle and the number of fruit in the no stem/no damage category. This agrees with Coppock et al. (1969). Th e increase in fruit exhibiting damage for angles below 110° over that of the linear pull te st suggests that the weaker fruit/stem interface, as is observed in lower harvesting forces, results in more

PAGE 88

74 frequent damage. The stem strength remain ed the same, if not increased, between the linear pull test and retraction angle tests becau se of the pulling motion being more in line with the branch structure. Thus although for angles below 85° there is a similar percentage of fruit in the no stem/no damage category as the linear pull test, there is a shift of fruit from the stem category to the various damage categories. Although physically the stem cannot be put at an angle to the fruit of more than 90°, a motion of 90° may not necessarily translate to a fruit/stem angle of 90° depending on the deflection behavior of the tree. Thus performing mo tions which may exceed fruit/stem angles of 90° will succeed in putting more of the fruit/st em angles at their maximum of 90°. This is evident in Figure 6-11 where there is li ttle change in the distance traveled among groupings above 105°. There is also little change in the detachment force for the fruit/stem angle grouping at 95° as compared with that of 135° as evident in Figure 6-10. Figure 6-13 illustrates the relationship between harvesting forces and number of prior rotations. The line plotted in Figure 6-13 connects the mean detachment force at each level of rotation. The detachment force remains relatively cons tant with increased number of rotations until two rotations where the detachment force dropped. This observation is due to the fact that at two rotations 56% (5/9 ) of the fruit were detached prior to two rotations being reached. Many of the other te st fruit were approaching the point where the torsional stresses on the stem were substantial enough to harvest them as well. This weakened the fruit/stem interf ace enough to lower the detachment forces. This also explains the lowe r number of samples for this treatment. The only other treatment where fruit were de tached prior to the specified number of rotations being reached was 1.5 rotations where 22% (2/9) of the fruit were detached.

PAGE 89

75 Figure 6-14 illustrates the relationship betw een the distance traveled during the harvesting motion and the diffe rent rotational treatments. There appears to be no significant difference in the distance traveled among the different rotational treatments. This correlates with the previous observati on of the dependence of the distance traveled on the detachment force. Since detachment force remains relatively constant for the different rotations, this correlate s into constant distance travel ed for each of the rotations. Table 6-3 indicates the number of fruit obs erved to be in each damage category per number of rotations. Figure 6-15 illustrate s the percentage of fruit in each damage category per number of rotations. It app ears that the added torsional stress on the fruit/stem interface decreases the occurrences of damage. None of the rotational treatments displayed fruit with damage, comp ared with the zero rotation treatment where almost half the fruit displayed a form of da mage. The treatment with 0.5 rotations has the largest amount of fruit in the no stem/no dama ge category, which is th e most desirable. Starting at one rotation and increasing with increased number of rotations, damage to the portion of the stem remaining on the tree was observed. The torsional stress on the stem caused it to split apart. Figure 6-16 illustrates an example secondary pitch harvesting force plot. The plateau at b) indicates a stoppa ge in motion prior to executin g the secondary pitch. This plateau is above the force value at which point the manipulator was to execute the secondary pitch. This force value is indicat ed in Figure 6-16 by the dashed line. The difference between the force value and the plateau is a result of the time lag in communications and physically stopping the mo tion of the manipulator. The additional movement of the manipulator along the harves ting direction during this time lag allowed

PAGE 90

76 higher forces to develop. The slope of detach ment is indicated in the figure as the region where the observed forces drop to c). The slope is indicated rather than the point as the changing orientation of the force/torque se nsor during the secondary pitch prevented determining the exact point of de tachment. The force level at c) is lower than that at a) because of the added weight of the fruit being he ld by the end effector as well as shifts in the end effector weight due to orientation ch anges of the force/tor que sensor during the secondary pitch. A complete study was not able to be performed, since a majority of the fruit tested detached prior to the secondary pitch being executed. Although the handheld test indicated a higher detachment force when these tests were performed, a decrease in the detachment force was observed for the se condary pitch test as compared to the retraction angle test. All of the fruit tested were from the same area of the same tree and thus it is possible that the characteristics of that area such as stem size may have influenced this result. The force value at which the secondary pitch was triggered was unable to be reduced because of sensitivitie s of the force/torque sensor to initial acceleration forces. This sensitivity woul d cause the manipulator to execute the secondary pitch even if no external forces were being exerted on the manipulator. This lack of compensation for inertial effects intr oduced further uncertainties that reduced the effectiveness of this type of motion. Even if the triggering fo rce value was adjusted to be below the detachment force of all of the fruit, the amount of variabil ity in the detachment force, fruit size, fruit mass, branch complia nce, fruit/stem orientation, and so on may prevent any observed improvement over performing a motion without any force feedback. This suggests that an open l oop control of the manipulator has the most potential for successful robotic harvesting of citrus.

PAGE 91

77 Conclusions Based on the results of the ha rvesting motion tests; the di stance traveled, harvesting forces and fruit damage may be minimized by imparting a 90° angle between the fruit and stem. This may be accomplished by performing a motion to impart at least a 110° angle between the fruit and the stem if no feedback is ava ilable to confirm whether the desired angle has been achieved. The result s also indicate that a 180 to 360° rotation about the stem axis may also minimize th e fruit damage, although having no significant impact on the distance traveled or harvesti ng forces Although a motion based on sensed magnitude of the harvesting forces from a fo rce/torque sensor, such as the secondary pitch motion, may be effective if further developed, the am ount of variability in the detachment force, fruit size, fruit mass, bran ch compliance, fruit/stem orientation, and so on may prevent any observed improvement. Th ere is however still potential that the sensed direction of the force vector may be utilized to ensure that the desired 90° angle between the fruit and stem is achieved.

PAGE 92

78 Figure 6-1. Force gauge and fruit holding stra p used in the linear pull test shown pulling a fruit on one of the test trees. Figure 6-2. End effector, force /torque sensor and robotic mani pulator used in testing are shown reaching towards a fruit on one of the test trees.

PAGE 93

79 Figure 6-3. Network of communication and pne umatic lines for the harvesting system. Figure 6-4. Characteristic di ameters taken of fruit.

PAGE 94

80 100 150 200 250 300 3503/9 3/23 4/6 4/20 5 / 4 5 / 18 6/1 6/15 6/29 7/13 7 / 2 7Test DateMass (g) Figure 6-5. Fruit mass variability over th e harvesting season. The connecting line indicates the average fruit mass at a give n test date for the fruit used in the linear pull test on 9 March 2006, 30 March, 12 April, 27 April, 17 May, 7 June, 6 July and 27 July. 5.2 5.7 6.2 6.7 7.2 7.7 8.2 8.7 9.23/9 3/23 4/6 4/20 5 / 4 5 / 18 6/1 6/15 6/29 7/13 7 / 2 7Test DateAverage Diameter (cm) Figure 6-6. Fruit diameter variability over the harvesting season. The connecting line indicates the average fruit diameter at a gi ven test date for the fruit used in the linear pull test on 9 March 2006, 30 March, 12 April, 27 April, 17 May, 7 June, 6 July and 27 July.

PAGE 95

81 0 20 40 60 80 100 120 1403/9 3/23 4/6 4/20 5 / 4 5 / 18 6/1 6/15 6/29 7/13 7 / 2 7Test DateDetachment Force (N) Figure 6-7. Detachment force variability over the harvesting season. The connecting line indicates the average fruit detachment fo rce at a given test date for the fruit used in the linear pull test on 9 Ma rch 2006, 30 March, 12 April, 27 April, 17 May, 7 June, 6 July and 27 July. Figure 6-8. Percentage of fruit in each damage category for test dates 1 to 8. These correspond to the linear pull tests conducted on 9 March 2006, 30 March, 12 April, 27 April, 17 May, 7 June, 6 July and 27 July.

PAGE 96

82 0 10 20 30 40 50 60 70 8020 40 6 0 80 100 1 20Detachment Force (N)Deflection (cm) Figure 6-9. Deflection of the fr uit from its initial position versus detachment force for the tests conducted on 27 July. The fruit ha ve been grouped in 20 N increments such that the first grouping represents fruit having detachment forces between 20 to 40 N up to the last group represen ting fruit having detachment forces between 100 to 120 N. The connectin g line indicates the mean at each grouping. 0 10 20 30 40 50 60 70 80 90 1003 0 4 0 5 0 60 70 80 90 100 1 1 0 1 20 130 140Fruit/Stem Angle (°)Detachment Force (N) Figure 6-10. Detachment force versus fr uit/stem angle at 10° groupings. The first grouping represents angles between 30 to 40°, the second grouping represents angles between 40 to 50° and so on up to the last group representing angles between 130 to 140°. The connectin g line indicates the mean of each grouping.

PAGE 97

83 0 10 20 30 40 50 60 7030 40 50 60 7 0 8 0 90 100 110 120 130 1 40Fruit/Stem Angle (°)Distance Traveled (cm) Figure 6-11. Distance traveled versus fr uit/stem angle at 10° groupings. The first grouping represents angles between 30 to 40°, the second grouping represents angles between 40 to 50° and so on up to the last group representing angles between 130 to 140°. The connectin g line indicates the mean of each grouping. A delay of as much as 200 ms in receiving the positional feedback may have contributed as much as 3.1 cm error. Figure 6-12. Percentage of fruit in each damage category for fruit/stem angle ranges 1 to 4. These correspond to fruit/stem angl e ranges 35 to 60, 60 to 85, 85 to 110 and 110 to 135 respectively.

PAGE 98

84 0 5 10 15 20 25 30 35 40 45 00.511.52 Number of RotationsDetachment Force (N) Figure 6-13. Detachment force versus number of rotations. The connecting line indicates the average fruit detachment force at a given number of rotations. 0 5 10 15 20 25 30 35 40 00.511.52 Number of RotationsDistance Traveled (cm) Figure 6-14. Distance traveled versus number of rotations. The c onnecting line indicates the average fruit detachment time at a given number of rotations. A delay of as much as 200 ms in receiving the pos itional feedback may have contributed as much as 2.8 cm error.

PAGE 99

85 Figure 6-15. Percentage of fruit in each da mage category for each number of rotations. 0 corresponds to 0 rotations, 1 to 0.5 rota tions up to 4 which corresponds to 2 rotations. Figure 6-16. Secondary pitch harvesting forces where a) is the flat region during which the end effector has alrea dy been initially pitched, but prior to retraction. The slope between a) and b) i ndicates the period of retr action until the triggering force of 22.24 N was reached. The platea u at b) indicates the region during which the end effector paused prior to executing the secondary pitch. The slope of detachment between b) and c) indicates execution of the secondary pitch during which the fruit was harvested. The flat region at c) indicates the region where the end effector was st opped after execution of the secondary pitch.

PAGE 100

86 Table 6-1. Number of fruit in each damage category for each test date for the linear pull test. Mar 9 Mar 30 Apr 12 Apr 27 May 17 Jun 7 Jul 6 Jul 27 No Stem/No Damage[a]2226262927232519 Stem[b]2619232122232518 Slight Damage[c]02000101 Damage[c]23101206 Large Damage[c]00000101 [c] Damage indicates a portion of the fruit peel was separated from the fruit along with the stem. Damage CategoryTest Date [a] No Stem/No Damage indicates a clean separation of the [b] Stem indicates a portion of the stem was still attached to the Table 6-2. Number of fruit in each damage category in each fruit/stem angle range. 35-6060-8585-110110-135 No Stem/No Damage[a] 16162324 Stem[b] 21241 Slight Damage[c]5220 Damage[c] 4440 Large Damage[c]0100 [c] Damage indicates a portion of the fruit peel was separated from the fruit along with the Damage Category Fruit/Stem Angle Range (°) [a] No Stem/No Damage indicates a clean separation of the stem from the fruit. [b] Stem indicates a portion of the stem was still attached to the fruit.

PAGE 101

87 Table 6-3. Number of fruit in each dama ge category per number of rotations. 00.511.52 No Stem/No Damage[a]58775 Stem[b] 01224 Slight Damage[c] 20000 Damage[c] 20000 Large Damage[c] 00000 [c] Damage indicates a portion of the fruit peel was separated from the fruit along with the stem. Damage CategoryNumber of Rotations [a] No Stem/No Damage indicates a clean separation of the stem from the fruit. [b] Stem indicates a portion of the stem was still

PAGE 102

88 CHAPTER 7 ROBOTIC CITRUS HARVESTING END EFFECTOR DEVELOPMENT Introduction As a result of rising costs and a shrinking la bor force, it is increasingly desirable to harvest citrus fruit robotically. Currently, re searchers at the University of Florida are working to develop such a harvester. The success of the harvester is dependent on its ability to quickly detect and harvest as much fruit as possi ble without damaging the fruit or the tree. An integral part of this success is the end effector . An end effector is a tool or device attached to the end of the manipulator that carries out a specific task. In the case of robotic harvestin g it is the part of the robot that actually does the harvesting. This chapter details the deve lopment of an end effector for r obotic harvesting of citrus based on the studies conducted in the previous chapters. Several research efforts have attempted to develop a robotic citrus harvester. These included efforts in Japan, Italy, Israel, Fr ance, Spain, and the United States. The end effectors developed in these research efforts fall into four types: cutting, pulling, twisting, and twisting/pulling. The most prevalent desi gn is the cutting end effector. This design cuts the stem, thus harvesting the fruit. Cutt ing end effectors are al so prevalent in other agricultural applications as th ey produce the least amount of stress on the fruit. Problems can arise, however, if the blade is not shar p enough or does not strike the stem properly, then a pulling action results and failure at the peel, referred to as plugging, often occurs. Another problem associated with the passing of a cutter around th e fruit involves the blade size. The blade must be large enough to be able to completely envelope the fruit,

PAGE 103

89 but not so large that it damages the surroundi ng fruit or tree structure. Regardless of protection for the surrounding fr uit and tree structure that may be present, a larger end effector impedes the ability to penetrate the can opy and harvest interior fruit. Muscato et al. (2005) reported a high success rate with a three fingered articulated cutting end effector that would grasp the fruit, detect th e stem location and then sever the stem using a micro-saw, but the implementation cost s were prohibitively high. A cutting end effector is impractical for the Florida citrus industry, as it is deem ed undesirable for any stem to be present on the citrus after harvesting. The second type of end effector is the pu lling end effector. This design harvests the fruit by pulling on the fruit in a particular direction, us ually along the stem axis. Once the force applied exceeds the binding forces along the fruit/stem interface, the fruit is harvested. These binding forces are often large enough that it results in plugging. These forces can be reduced if the force is applied at a 90° an gle to the axis of attachment, as is often done in hand pick ing. This method resu lts in less plugging but may increase the complexity of the motion and, thereby, increase the harvest time. Pulling actions also disturb the surrou nding fruit through limb oscillation, making subsequent rapid harvesting more difficult si nce moving fruit are much harder to harvest than stationary fruit. This method was propos ed by Pool and Harrell (1991). Their end effector, developed at the Univ ersity of Florida, used a ro tating lip with an attached collection sock to encircle th e fruit. As the lip rotated around the fruit, the stem was impinged between the lip and the upper portion of the end effector. The stem was then severed as the end effector retracted. The end effector successfully removed the fruit 69% of the time. Of these successful attempts , 63% of them caused damage to either the

PAGE 104

90 fruit or the tree. Eighty-five percent of this damage was to the tree which usually involved the removal of leaves or small branches. The most common fruit damage occurred when the fruit was caught between th e lip and the housing. Plugging of the fruit was rarely observed. The residual stem le ngths were between 0 a nd 2 inches with the occasional stem greater than 3.9 inches. The effect of replacing the lip with a cutter bar was also examined, but this method resulted in frequent cut fruit and residual stem lengths between 0.8 to over 5.9 inches. A th ree fingered end effector was proposed by Finkelstein (1996) that would gr asp the fruit and then retrac t the fruit inside an outer collar to create a bending and pulling action on the stem to separate it. It was developed by Greentech, Ltd., an Israeli entrepre neurial agricultura l technology company. The third type of end eff ector is the twisting end e ffector. This method was recommended by Juste et al. (1988) and Rabatel et al. (1995) as the most promising of the three. This involves twisting the fruit, prefer ably about its attachme nt axis, until the stem is severed. Twisting the fruit about the stem axis reduces the amount of disturbance to the tree and thus to the surrounding fruit. A twisting end effector was described in Juste et al. (1992). A pneumatic suction cup gripped the fruit from either the line of action of the telescopic arm or at a 30° angle from the line of action towards the bottom of the fruit. The suction cup was then rotated so as to sever the stem using a hydraulic motor. The detachment rate of the end effector was 64 to 67% with th e failures caused by obstacles, vacuum failures, or mechanical fa ilures. The rate of plugging of this end effector was 2 to 4%. The residual stem le ngth was less than 5 mm in 80 to 85% of the detached fruit. The end eff ector did not damage the fru it through its contact with the fruit, and 4 to 5% of the detached fruit were damaged by the su rrounding tree structure

PAGE 105

91 during removal. As in the pulling and cutt ing type end effectors, fruit size is a consideration here as well. Generally, the twisting action is achieve d by use of a rotating suction cup. This cup must be of the right size to create a good seal while still providing enough force to keep the orange from sli pping. Similar work was done on mushrooms by Reed et al. (2001), and it was found that th e smaller suction cups tended to slip, damaging the mushrooms. The researchers found that it was necessary to have the largest possible suction cup that would still be able to adhe re to the mushroom. One of the major advantages of this method is that if only a suction cup is used, there is a large flexibility in the angle of a pproach. Some angles may be more preferential than others. Except where the stem attaches to the fruit, th e cup could attach to a ny part of the fruit. The fourth type of end effector is the twisting/pulling end effector. This method was proposed by Tutle (1985) in U.S. Pa tent 4532757 and Tutle (1983). This end effector was developed by Martin Marietta Corporation and consisted of a four fingered gripper. Each of the fingers was equally spaced around the circumference of the gripper. Once the presence of the fruit is detected with in the gripper by three of the four tactile sensors located at the rear of the gripping area on each of the fingers, the fingers are actuated to close around the fruit. To seve r the stem, the grippe r pivots the head downwards and then rotates. The most promising end effectors develope d were the three fingered articulated end effector described in Muscato et al. (2005), the twisting/pul ling end effector proposed in Tutle (1983) and the three fingered pulling end effector proposed in Finkelstein (1996). It is uncertain what the results of these e nd effectors were. Mus cato et al. (2005) only reported that the results were extraordinary. The other two end effectors had no reported

PAGE 106

92 results. These were the most promising because they were all able to grasp the fruit. Although they were not all used to impart a sp ecific motion to the fruit, a similar design could be used in this manner. Based on the results of the harvesting motion tests it was determined that an end effector was needed which could impart a speci fic motion to the fruit. There were several limitations of the end effector used in the ha rvesting motion tests. It was limited on its ability to accurately position the end effector fo r capture partially due to the fact that the sensors were not in line with the center of the end effector. The grasping fingers also lacked robustness if the end effector was inaccurately positioned. Since the acutation mechanism was pneumatic, it required comp ressed air which was not needed by any other component of the robotic system. It was therefore necessary to develop an end effector which would improve upon the existi ng and previously reported end effectors. Objectives Design a robotic citrus harvesting end eff ector based on the results of the physical properties and harvesting motion tests Build and test the designed robotic ha rvesting end effector to determine its effectiveness based on the design criteria. End Effector Design Criteria Lightweight (<17.78 N (4 lb)), based on th e manipulator payload capacity (155 N) and the weight of the end effector used in the harvesting motion tests (24.5 N) Cost-effective (<$1000, not including labor), based on economic estimates of the overall cost of an economically viable harvesting system Fast (<0.5 sec), based on a total cycle time of 2 sec Produce minimal damage on fruit and tree Able to reach fruit inside the canopy Able to integrate sensing technology such as vision, ultrasonic and infrared sensors

PAGE 107

93 Materials and Methods In order to impart a specific motion to the fr uit, the end effector had to be able to control the position and orient ation of the fruit. The designed safe limits of handling were based on the results of the physical prope rties tests reported in chapter five. The designed maximum loading was based on the ha rvesting motion tests reported in chapter six. These design parameters were used to design the end effector using SolidWorks (2003). SolidWorks (2003) was used as a desi gn tool to determine the dimensions and placement of the various components of the end effector. Various sized fruit were modeled in SolidWorks (2003) and placed within the grasp of the end effector to confirm how the end effector would receive and grasp the fruit. The designed end effector is shown in Figure 7-1. The SolidWorks (2003) Animator was then used to examine the actuating motion and detect a ny interferences. The necessary components of the end effector were then either manufact ured or obtained from suppliers. Once the end effector had been assembled, validation tests were conducted in order to compare the end effector's performance against the design criteria. Field tests were performed on Valencia ( Citrus sinesis cv. Valencia) orange trees in a commercial grove located in Umatilla, FL. The trees were arranged in east-west hedge rows. Tests were performed which put the fruit and stem at an angle of either 0, 65 or 110°. A variation on the code described in Appendix B was used to control the motions and the same code described in Appendix C was used to collect th e force/torque data. For the 0° tests the end effector grasped the fruit and then moved vertically down. For the 65 and 110° tests the end effector was first pitched back by 20°. The manipulator then retracted along a line of 45 or 90° from vertical to put the fru it and stem at the desired angle. The tests were performed on 6 July 2006.

PAGE 108

94 First, the ability of the end effector to us e the integrated sensors to position the end effector was evaluated. This was an evalua tion of how well the sensors were integrated, not the performance of the sensors. Second, the ability of the end effector to grasp the fruit and hold onto the fruit while performing the desired harvesting motion was evaluated. Third, the condition of the fru it and tree after harves ting was evaluated. Lastly, the ability of the end e ffector to harvest the fruit even if the end effector was not accurately positioned was evaluated. The end effector was then run for 100 consecutive open/clos e cycles and the average cycle time was recorded in order to evaluate its actuation time. This was then halved to determine its half-actuation time. It was then weighed using an Instron universal testing machine (model 5566) with a 1 kN load cell on the crosshead. The amount of force that the finger was able to apply was evaluated using the same universal testing machine. The end effector finger was placed directly beneat h the load cell and a cable was attached to the finger and the load cell. The end effector was then commanded to close and the maximum load was recorded. The outer dimensions of the end effector were then taken in order to evaluate its effectiveness of reaching fruit inside the canopy. Results and Discussion The first step in the design process was to determine the overall design of the end effector. Several designs were considered including those with pneumatic, hydraulic or electrical actuation. Al l of the designs utilized a thr ee fingered grasping approach. A three fingered approach was used as it was desi red to have as little contact with the fruit as possible without damaging it, while still bein g able to securely gr asp the fruit. Three fingers is the minimum for grasping an object su ch as an orange and so it was determined to use this approach.

PAGE 109

95 Pneumatic and hydraulic designs were c onsidered. Although the current robotic system uses primarily electrical actuators, fu ture systems may utilize different forms of actuation. The three methods of actuation cons idered were: a set of three cylinders; a master actuating cylinder with three slave cy linders; and a single cylinder that would actuate all three fingers through mechanical linkage. The three cylinders could be actuated either independently or jointly. The cylinders could be linked to actuate jointly either through the use of a flow divider, m echanical linkage or a combination of both. The independently actuated cylinders provi ded the advantage that if there was an obstruction such as a branch the other two cyli nders would still be able to close. This also meant that the end effector would not necessarily center the fruit in its grasp. A motion based on imparting specific translations and rotations to the fruit might then be prone to error depending on how much the fru it was off-center. Actuating the cylinders jointly ensured that the fruit was centered. Bo th the independently and jointly actuated cylinders necessitated the use of an outside fluid power source. These designs were not selected because of the need for an additional power source and concerns over the weight and space requirements of the three cylinde rs and their hosing. The independently actuated cylinders also had the additional aforementioned concerns over performance. The advantage to a master/slave system wa s that the system had the possibility of being a closed fluid system. That is, an el ectric actuator would be used to drive the master cylinder at a location away from the end effector and then the fluid lines would be fed to the end effector. This would eliminate the need for an outside fluid power source. The cylinders could then be act uated either independently or jointly as in the previously mentioned design, with the same advantages and disadvantages. This design was not

PAGE 110

96 selected because of concerns over the we ight and space requirements of the three cylinders and their hosing. The advantage of the single cylinder with mechanical linkages to the three fingers was that it would center the fruit without the need for a flow divider. It would also reduce the amount of hoses needed at the end effector. In or der to provide equal leverage to each of the fingers, this also meant that it would most likely need to be mounted in the center of the end effector. This would then reduce the amount of space available for the centering of the sensors in the design. As in the three cylinder design it would also be necessary to provide an outside fluid power source. This de sign was not selected because of concerns over the space requ irements along the center axis of the end effector and the need for an additional power source. Electrical designs considered included using an electric li near actuator, DC motor, servo-motor, and stepper motor. All three de signs utilized a single actuator to close all three fingers. The electric linear actuator design was substantially similar to the single cylinder design. Although the space requirement along the center axis of the end effector was reduced because of the overall length of the actuator remaining constant during actuation, there were additional concerns over th e size, weight and available power of the actuator. This design was therefore no t chosen because of these concerns. The three motors all used linkages to act uate the fingers. These included: cables and pulleys, rack and pinion, g ears and screws. Cables and pulleys had the advantage of being lighter and more easily r outed in tight space constraints. However, they also had the potential to break or stretch which woul d potentially reduce the controlability of the actuator. They were therefor e not selected as a component of the design. The other three

PAGE 111

97 designs were all rigid linkage and thus were heavier and harder to integrate into tight space constraints. The rack and pinion a nd gear transmissions had the additional disadvantage of the potential for tooth wear co mplicating the control process. They were therefore not selected as a co mponent of the design. Screws had the advantage that there was frictional resistance against backdriving th e screw. This meant that a smaller motor could be used to resist the ha rvesting forces. If the fricti onal resistance was great enough, the mechanism would be self locking and so not require a large actuator or secondary locking mechanism. Screws also had the a dvantage of being able to transmit the power from the actuator to the thr ee fingers with relatively si mple linkages. Based on the aforementioned advantages to using screws, they were selected to transmit the power from the actuator to the fingers. A ballscrew was specifically selected because of its ability to run at high speeds. Of the three motor types available the servo-motor provided the most precise control, but with the current harvesting system not being able to determine the size of the fruit prior to grasping and thus the amount that the fingers must close, this level of control may not be able to be implemented. In future systems the vision system or other sensors could possibly be used to implement this. However if an open loop system is able to be implemented successfully, then a closed loop system may not be necessary. The servo-motor would also be more expens ive than the other two motors. Both the stepper and DC motors could be implemented as either an open loop or a closed loop system. Both the stepper motor and the DC motor were consid ered possible design alternatives depending on the specifics of the available motors.

PAGE 112

98 The finger design was developed concurren tly to the actuatio n method design in order to optimize the overall design. Severa l designs were consid ered including jointed, bellows and rigid fingers. The jointed and be llows designs had the potential to customize the grip to the shape of each individual fru it. However this weakened the overall strength of the finger, would possibly have added in creased actuation time and made the finger more prone to damage. The rigid design was not as flexible as the other two in terms of conforming to the individual fruit, however so me of this lack of flexibility could be regained with the use of elastomer padding which would conform to the shape of the fruit. One of the rigid finger designs consid ered involved the use of tapered fingers. The fingers would initially enclose around the fru it without grasping it, during retraction the fruit would be pulled into the taper which woul d grasp the fruit while still giving time for leaves and branches to pull out of the grasp. Since the fruit would be allowed to initially reorient itself, this would potentially indu ce error into the harvesting motion until the fruit became firmly gripped by the taper. There was also still potential for leaves and branches to be caught between the fruit and the taper. A second rigid finger design considered involved the use of fingers curved to the approximate curvature of an orange. The elastomer padding would then be used to accomm odate fruit that were either larger or smaller than the finger curvature. This was the design selected as it was the most robust and provided the best potential to firmly gr asp the fruit for control without damaging it. After the general design had been determin ed, the specific component selection and dimensioning was performed. According to Southwest Florida Res earch and Education Center (2006) the average fruit diameter for oranges ranges between 6.67 cm and 8.26 cm. Specifically the average diameter of th e fruit tested in th e harvesting motion and

PAGE 113

99 physical properties tests was 7.04 cm with a st andard deviation of 0.51 cm. The smallest fruit observed in the physica l properties and harvesting mo tion tests was 5.69 cm in diameter, while the largest fruit was 8.79 cm in diameter. The fingers were thus designed to have a radius of curvatur e of 4.13 cm, which would correla te to an 8.26 cm diameter fruit. This design would be able to accomm odate fruit throughout th e range of diameters when used in conjunction with the elastomer padding. The length of contact was evaluated for a sphere with diameter equal to that of an average Valencia variety fruit. For a 0.635 cm deep elastormer pad under compression it was determined to be approximately 6.60 cm. With a 2.54 cm finger wi dth this then transl ates into a contact area of 16.76 cm2 and a contact perimeter of 18.29 cm . This contact area and perimeter was then compared against equivalent punch di ameter sizes. As was stated in chapter five, the largest punch size tested of 2.54 cm di ameter approached the behavior of a flat plate. The equivalent punch diameter by area was 4.57 cm and by perimeter was 5.84 cm. Thus if the loading from the harves ting motion occurred on only one finger, the maximum allowable loading would approach the bursting limits. As the maximum observed detachment force in chapter six wa s 124 N in the linear pull test and the minimum observed bursting force from chapter five was 135 N in the parallel plate test, the fruit should not be damaged. The fingers were designed so as to not have any sharp edges that might puncture the fruit. The fingers wrapped around the back of the fruit in order to prevent the fruit from pulling out of the end effector. This wrap limited the amount that the fingers could close. The smallest fruit that could be gr asped was approximately 5.08 cm in diameter. Smaller fruit could be encompassed by the finge rs and as the end effector retracted the

PAGE 114

100 elastomer padding would reduce the amount that the fruit would move, however a firm grasp would not be able to be achieved. The elastomer padding selected was 0.635 cm EPDM open cell sponge rubber sheets. EPDM was selected because of its excellent weathering properties. The compression range at 25% deflection was 28 to 48 kPa which corresponded to a relatively soft compound. This allowed the elastomer padding to better conform to the shape of each fruit. There was still enough resistance so as to prevent the padding from completely collapsing under lo ading. The designed finger is shown in Figure 7-2. The arrangement of the fingers was designed so as to accept a 11.43 cm diameter fruit, have as small a profile as possi ble in order to facilitate penetration into the tree canopy and integrate the sensors into the overall design. The pi voting mechanism of the fingers was designed so as to minimize th e actuation travel distance and maximize the leverage so as to enable the use of as small of an actuator as possible. The fingers were actuated through the use of pivoting links. As the connecting point for the link is extended out, the fingers are forced outwar d. The connecting points for the links were placed on a collar with a hole in the middle to allow for the integrated sensors. The collar was then connected through a set of three rods to the collar of a ballscrew. The rods passed through the pivoting plat e for the fingers through oil impregnated sintered bronze pressbearings. The connecting rods were at tached to the finger pivoting collar by set screws and were attached to the ballscrew collar using two retaining collars. This allowed the connecting rod some movement la terally to account for any misalignment in machining. This transfer m echanism between the connecti ng rods and the fingers is shown in Figure 7-3.

PAGE 115

101 The ballscrew selected had a dynamic load rating of 605 N, a stat ic load rating of 6294 N, a torque requirement of 0.0025 N-m to act uate a collar load of 4.45 N, a lead of 0.05 cm/rad and a maximum speed rating of 42 cm/sec. The lead and specified maximum speed rating translated into a ma ximum rotational speed of 838 rad/s. The ballscrew was then driven by a stepper moto r through a belt and a set of timing pulleys. The stepper motor selected, Excitron Corp. model 42-38, op timized the balance of size and weight with power. The motor was rated at 205 rad/s maximum speed and 0.388 Nm maximum torque. The operating speed and to rque were controlled through the use of an integrated controller, Ex citron Corp. model 10A-RS232. Th e specified design criteria required the end effector to actuate in 0.5 sec which was de fined as the time required to open and close the end effector. This woul d then require a half-actuation time of 0.25 sec. With a 0.32 cm lead on the ballscrew a nd a total actuation travel length of 1.59 cm this would require an av erage speed of 251 rad/sec. The predicted lifetime of the screw was 1.35 years with the following assumptions: a total actuation travel length of 1.59 cm, one fruit harvested every 2 seconds, 2 actua tions per fruit corresponding to one open and one close, a travel life of 2.54*107 cm as specified by the manufacturer, a 12 hour workday and a 9 month work year. The actual observed lifetime may be greater as the minimum loading for the specified lifetime cu rve by the manufacturer was above that to be experienced by the ballscrew. For a design grasp loading of 22.25 N per finger, the required torque for this mechanism would be 0.12 N-m. Based on the designed mechanism a pulley drive ratio was selected of 1:1. Although the motor was underrated for speed and overrated for tor que, it was found that the torq ue rating was for very low speed, thus at the maximum speed possible fo r the motor the torque capability was just

PAGE 116

102 enough. The maximum speed was also limited by the acceleration loading on the motor. If the motor was accelerated too quickly, the lo ading on the motor would cause it to start slipping. The pulleys used were also selected so as to maximize th e amount of teeth in mesh on the motor pulley. This reduced any load variations that might cause the motor to start slipping. Since the primary objective for the end effector was to be able to harvest the fruit successfully, it was decided to sacr ifice the speed criteria . The ballscrew was supported by a set of miniature stainless steel ball bearings rated up to 5236 rad/sec and a 1081 N dynamic load carrying capacity. The be lt was tensioned with a second set of bearings rated at 5864 rad/sec. The designe d motor and drive mechanism is shown in Figure 7-4. The sensors to be integrated include d a Korea Technology and Communications model KPC-S20CP1 camera and an infrared range sensor. The camera had dimensions of 2.2 cm height by 2.2 cm length by 2.67 cm dept h and a field of view cone of 45°. It was connected to a transformer mounted on the side of the end effector which stepped the available supply voltage of 24V DC down to 12V. The infrared range sensor consisted of a transmitter and receiver pair that were m ounted on either side of the camera. The transmitter and receiver needed to be at least 1.27 cm apart in order to prevent the receiver from receiving the transmitted signals before they had encountered the objects in front of the sensor. The infrared sensor had a range of 0.95 to 7.6 cm when shaded. During testing it was observed that the sens or was unable to provide range data when operating directly in the sun. Stereo vi sion and an ultrasonic sensor, UNAM model 18U6903/S14, were used for initia l range estimates and the infr ared sensor was used to adjust these estimates. The integrated se nsors are shown in Figures 7-5 and 7-6. A

PAGE 117

103 schematic of the infrared sensor used is shown in Appendix D. A diagram of the communication network for the harvesting sy stem using the new end effector and integrated sensors is shown in Figure 7-7. Figure 7-8 illustrates the forces experien ced by the end effect or during the trial tests. The forces were characteristic of th e harvesting forces reported in chapter six. The average 0° detachment force observed in the lin ear pull test from the same test date was 78.9 N. This was characteristic of the observe d detachment forces from other test dates as was shown in Figure 6-7 of chapter six. The average 0° detachment force observed in the end effector trial tests was 52.9 N. The reason behind this discrepancy was uncertain, however it is possible that the differences in the velocities and accelerations imparted to the fruit in each of the tests may have c ontributed to the observed discrepancy. The average 65° detachment force observed in th e retraction angle test s of chapter six was approximately 45 N. The average 65° detachme nt force observed in the end effector trial tests was 37.6 N. The average 110° detachme nt force observed in the retraction angle tests of chapter six was approximately 39 N. The average 110° detachment force observed in the end effector trial test s was 37.8 N. Although both the 70 and 110° detachment forces were lower in the end effect or trial tests than the retraction angle tests, the linear pull test indicated a higher detachment force on the end effector trial test date as compared to the retraction angle test da te. Table 7-1 indicates the number of fruit observed to be in each damage category. Figure 7-9 illustrates the percentage of fruit in each damage category. The number of fruit in each damage category was comparable to that reported in chapter six. The differences in the percentage of fruit observed in each category for comparable angles in the retr action angle tests of chapter six can be

PAGE 118

104 attributed to the small sample size. This indicates that the end effector was able to perform the desired motion and achieve the e xpected results. No damage was observed because of the interaction between the fruit a nd the end effector. All of the fruit were able to be harvested without slipping out of the end effector's grasp. The sensors had limitations on being able to adequately position the end effector, however the end effector was still able to harvest the fruit ev en if it was moderately off-center. The end effector had an average half-act uation time of 0.54 sec and a weight of 20.73 N including the integrated sensors. The loading per finger as determined by the tests with the Instron was determined to be 17 N at the average fruit diameter. Using the same contact area as was determined previously this translates into a pressure of 10 kPa. The profile of the end effector could be inscribed inside a ci rcle of diameter 16 cm. The total cost for the motor, ballscrew, beari ngs, pulleys, belt, alum inum stock, elastomer padding, camera, infrared sensor, ultrasonic sensor, miscellaneous hardware, cables and power supply used in the design was approxima tely $1300. Of this cost, over half was that of the integrated sens ors and the power supply. This cost could therefore be significantly reduced by finding less expensive al ternatives, particularly for the ultrasonic sensor and the power supply. This cost a nd the design criteria specification did not reflect any labor cost associated with ma nufacturing and assembling the end effector. Conclusions Although the end effector did not meet all of the design criteria, it was able to harvest fruit successfully and the integrat ion of the sensors improved the positioning control over the previous end effector. The overall weight and cost could be reduced with further refinements of the design. The speed is not a concern for the current harvesting system as the overall system is not capable of the final design speeds.

PAGE 119

105 However, this is an area that would need to be addressed prior to implementation in an improved harvesting system. Either the actua tor or the drive mech anism would possibly need to be redesigned in order to improve th e actuation time. Alternatively, it is possible that the previous pneumatic end effector co uld be redesigned to better integrate the sensors and use the new fingers to improve th e grasping ability. Future designs may also incorporate the ability to perf orm 180° rotations about the st em axis and 110° rotations about an axis perpendicular to the stem axis. These are mo tions that were concluded in chapter six to yield the best results. The ability of the end effector to perform these motions would eliminate the need for the manipulator to perform them, which may take increased time and operational space. Alt hough there are improvements that could be made to the end effector presented in this chapter, the design pro cess of basing an end effector on the results of the tests conducted in chapters five a nd six yielded an end effector which was able to withstand the harvesting forces it experienced while not damaging the fruit or the tree.

PAGE 120

106 Figure 7-1. Designed end effector. Figure 7-2. Designed grasping fi nger with a) the finger, b) the elastomer padding and c) the pivoting link. Figure 7-3. Transfer mechanism between the connecting rods and fingers with a) the connecting rod, b) the outer collar, c) the pi voting link, d) the retaining collars and e) the ballscrew collar.

PAGE 121

107 Figure 7-4. Designed drive mechanism with a) th e stepper motor, b) th e driver pulley, c) the timing belt, d) the belt tensioner, e) the driven pulley, f) the ballscrew, g) the ballscrew collar and h) the connecting rod.

PAGE 122

108 Figure 7-5. Integrated sensors with a) the camera, b) the infrared receiver and c) the infrared transmitter. Figure 7-6. Integrated ultras onic sensor, labeled a), mounted on the side of the end effector.

PAGE 123

109 Figure 7-7. Communications network for th e harvesting system with the new end effector and integrated sensors. 0 10 20 30 40 50 60 70 80 90 1000 1 0 2 0 30 40 50 60 7 0 8 0 9 0 100 110Fruit/Stem Angle (°)Detachment Force (N) Figure 7-8. Detachment force versus fruit/ste m angle for the fruit tested with the new end effector. The connecting line indicates th e mean detachment force at each of the three angles tested.

PAGE 124

110 Figure 7-9. Percentage of fruit in each damage category for fruit/stem angles 0, 65 and 110°. The first column represents 0, th e second column represents 65 and the third column represents 110°.

PAGE 125

111 Table 7-1. Number of fruit in each dama ge category for each fruit/stem angle. 065110 No Stem/No Damage[a]587 Stem[b] 012 Slight Damage[c] 200 Damage[c] 200 Large Damage[c] 000 [c] Damage indicates a portion of the fruit peel was Damage CategoryFruit/Stem Angle (°) [a] No Stem/No Damage indicates a clean [b] Stem indicates a portion of the stem was still attached to the fruit.

PAGE 126

112 CHAPTER 8 STEM-DETECTION STUDIES AND FORC E CONTROL MODEL DEVELOPMENT Introduction In order to optimize the eff ectiveness of a robotic citrus harvesting end effector, it was decided to explore the use of force feedback for optimizing the control of the harvesting manipulator. Based on the results of the harvesti ng motion tests reported in chapter six, performing a motion that will place the fruit and stem at a specified angle will decrease the amount of damage to the fr uit as well as result in lower harvesting forces which in turn results in less trav el distance required to harvest the fruit. Force sensors have been used in robotic control since 1972 (Boubekri and Chakraborty, 2002). Generally they consist of elements such as strain gauges, piezoelectric pressure sensors, resistive pressure sensors, capacitive pressure sensors, and optical pressure sensors. These transform the forces into electrical signals which can be evaluated by the controller. Force sensors have to date been used in a wide variety of applications including: part mating (Chin et al., 2003), de burring (Pires et al., 2002), minimally invasive surgery (Tegin and Wikander, 2005), an d robotic fruit harvesting (Allotta et al., 1990). Force sensors can be divided into two categ ories, those that measure forces at a point further removed from the area where the forces are applied and those that measure the forces at the point where they are appl ied (Tegin and Wikander, 2005). One example of a sensor that measures the forces at a point further removed from the area where the forces are applied is a force/to rque sensor. These are sensors that use a series of strain

PAGE 127

113 gauges mounted along the direction of the princi ple forces. This results in either a 3DOF or a 6-DOF sensor. The 3-DOF sensor ha s the ability to sense forces in the x, y, and z directions while the 6-DOF sensor also senses the corre sponding moments about the three axes. A force/torque sensor is usually mount ed between the manipulator and the end effector, however it also may be mounted on the fingertips in order to provide better tactile information (Tegin and Wikander, 2005). The advantage of this type of sensor is that it is generally more accurate than those that measure the forces at the point where they are applied. The disadvantag es of this type of sensor ar e that it is unable to provide force measurements at more than one point and may give errors due to inertial effects. Since there is a mass between the sensor and the applied force, accelerations by the manipulator will give force readings that incl ude the inertial forces of that mass. The inability to measure more than one point of contact limits its use in tactile sensing. It is mainly used when there is only one force vector of interest, such as in grinding or part mating. Force/torque sensors can be used with tw o different types of controllers (Perry, 2002). The first is a controller mounted with th e sensor. It is a se lf-contained unit that can communicate directly with the robot via RS-232 or analog voltages. The second type is the computer bus sensor c ontroller that connect s with the robotÂ’s motherboard. This enables the sensor to integrate more seamlessly with the robotic system. Sensors that measure the forces at the poi nt where they are applied usually consist of a sensor array. These usually are capa ble of measuring pressure only, although some have been developed which can measure sh ear forces as well (Tegin and Wikander,

PAGE 128

114 2005). Optical or fluid filled membranes th at deform upon applied pressure provide the best spatial resolution of these sensors. Th e amount and location of the deformation is recorded through the use of photocells or cameras . Other sensor arrays consist of a series of individual pressure sensors. These are often piezoelectric, resistive, or capacitive. All three change their electrical properties under pressure. The piezoelectric sensors produce an electric charge when deformed. The re sistive and capacitive se nsors involve passing current through resistors and capacitors resp ectively which change their resistive or capacitive properties under deformation. Alternatively, sensors can be placed on th e servomotors in order to directly measure the torque delivered by the actuators and thus calculate the applied load at the end effector (Sciavicco and Siciliano, 2000). This more indirect method of measuring forces at the end effector often include non linearities associated with the manipulator structure and so are more inaccurate. Sin ce it also involves m ounting sensors along the drive mechanisms for each of the joints it is also more complicated to implement. There are two main types of force contro l that are used: through-the-arm control and around-the-arm control (Erlbacher, 2000). Th rough-the-arm control utilizes all of the joints in order to position the arm in a manne r so as to apply the desired forces. Around-the-arm control utilizes the arm fo r positioning only, while the end effector applies the desired forces. Through-the-arm control thus involves reso lving the desired trajectory with the trajectory needed in order to generate th e desired forces. This can be accomplished through the use of hybrid control. A block di agram of a typical hybrid control scheme is shown in Figure 8-1. This type of control can lead to positional inaccuracies which may

PAGE 129

115 produce excessive loads on either the manipulator or the part. In order to safeguard from these loads occurring, compliant mechanisms are used on the end effector in order to allow for a certain amount of positional inaccuracy. The compliant mechanisms are placed on the end effector as making the enti re manipulator compliant would result in greater positional inaccuracies throughout the manipulator. Be cause of the potential for these force overloads, through-the-arm control is most often used at lower velocities in more controlled situations. Around-the-arm control uses the manipulator to position the end effector within a certain range of the desired position. This is also known as impedance control. The end effector contains additional positioning capabilities which are more flexible. The most widely used form is pneumatics. Pneuma tics allow the end effector to adjust for positional inaccuracies due to th e compressibility of air. There are two methods of around-the-arm cont rol, passive and active. In passive control the changing weight of the end effect or is counterbalanced either mechanically using counterweights or pneum atically using a double-acting pneumatic actuator. The mechanical counterbalance shifts as the e nd effectors orientation changes. In the pneumatic system the orientation must be known either through me asuring the degree angle or use the feedback from the manipulator actuators to calculate the degree angle of the end effector. Based on the orientation of the end effector, the known weight of the end effector and the desired applied force the pneumatic actuator is adjusted accordingly. This is an open loop control structure. Active control involves using sensors such as a load cell and an accelerometer to provide feedback to the contro l structure. The accelerometer records the orientation of

PAGE 130

116 the end effector as well as iner tial effects from the movement of the end effector that may produce forces at the tool tip. The load cell pr ovides feedback as to the actual force being applied by the end effector. Both the active and passive systems will need periodic calibration. The load cell and accelerometer in the active system may dr ift over time and the weight of the end effector may change in the passive system. Several sensors and force control methods have thus been presented which, when implemented into a robotic control system, allo w the manipulator to better interact with its environment. The potential for use is highest in operations where the operational environment is less structured leading to higher variability in the forces that the manipulator will encounter. One such unstr uctured environment is that of robotic harvesting. Allotta et al. (1990) describes the use of a force-torque sensor in determining the approximate location of the stem for cutting. The location is determined by one of two methods depending on the degrees of freedom of the robot used. In the case of a 6-DOF robot, a series of motions are performed in order to reduce the reaction for ces applied by the stem to a single pulling force. The locati on of the stem is then inferred to be at the intersection of the upper hemisphere of the fruit and the line of action of the force. In the case of a 3-DOF robot, subsequent vertical and horizontal moves are performed in order to determine two planes in which the center of the fruit and the stem lie. A line common to these two planes can then be determined a nd it can then be assume d that the location of the stem is the intersection of this line and the upper hemisphere of the fruit. It was found that the method used in the case of a 6-DOF robot provided sufficient accuracy

PAGE 131

117 with an error of less than 15°. The algorith m took an average of 4 seconds to converge, however this time could be improved with an improved architecture and bandwidth. The method used in the case of a 3-DOF robot was faster but had s ubstantial error which sometimes exceeded 45°. The implementation of this algorithm into a robotic orange harvester is described in Muscato et al. (2005). Allotta et al. (1990) determined stem locat ion through a series of movements. In the optimization of a fruit harvesting motion, it may not be possible to perform these motions. As such one of the objectives of this chapter is to determine if stem location detection may be used to optimize the harvesting motion without the need for additional movements. There has also been no attemp t to develop a control strategy that would modify the harvesting motion based on the sensed fruit/stem angle. It is therefore a second objective of this chapter to develop a fo rce control model that uses feedback from the force torque sensor to control a robotic citrus harv esting manipulator. Objectives Determine if stem location detection ma y be used to optimize the harvesting motion without the need for additional movements. Construct a force control model that woul d utilize feedback from a force/torque sensor to control a robotic citrus harvesting manipulator. Materials and Methods Tests were performed on Valencia ( Citrus sinesis cv. Valencia) orange trees to examine the ability of the force/torque sensor to accurately determine the angle between the fruit and stem. The same experimental setup was used as in the harvesting motion tests described in chapter six. The same code described in Appendix C was used to collect the force/torque data. The motion of the end effector was controlled manually through the GUI. The end effector first grasped fruit where the fruit ax is was parallel to

PAGE 132

118 gravity. The end effector then retracted along a direction perpendicular to gravity. After sufficient tension developed in the stem, the end effector was stopped. The force/torque sensor's data was then taken and the sensed angle between the observed forces and the fruit axis was determined using Matlab (2004) . The angle was determined using a unit vector representing the fruit axis and a un it vector representing the direction of the applied forces. The cosine of the angle wa s then equal to the dot product of these two unit vectors. A diagram of the fruit, stem and sensed forces are shown in Figure 8-2. A diagram of what the Matlab (2004) algorithm used to calculate the angle is shown in Figure 8-3. The calculated angle was compared agains t two observed angles. The first angle measured was the angle between the fruit axis and the stem in the re gion near the fruit. The second angle measured was the angle between the fruit axis and the line between the fruit/stem interface point and the point of attach ment of the stem to the main branch. The first angle was measured by placing a protractor against the stem and perpendicular to the fruit axis with the origin at the stem/fru it juncture. The protra ctor orientation was ensured to be perpendicular to the fruit axis w ith the use of a level th at had 2° resolution. The second angle was measured in a manner sim ilar to the first except that the protractor was placed against a string that was tied betw een the point of attachment of the stem to the main branch and the fruit/stem interface point. The setup for measuring the two angles is shown in Figure 8-4. Once the accuracy of the force/torque system in predicting fruit/stem angle had been determined, a force control model was de veloped in Matlab ( 2004) that accepted as inputs the data from the force/torque sensor and the current rotation matrix from the

PAGE 133

119 manipulator. The model then determined th e angle between the sensed forces and the fruit axis. This was done using the previous ly described method with the assumption of a vertically oriented fruit. The model then determined the difference between the desired fruit/stem angle and the current fruit/stem angle. Based on the results of the harvesting motion tests performed in chapter six, it is desirable to have a fruit/stem angle of at least 110°. This is the angle used in the developed control model. The axis normal to the plane on which both the fruit axis and the re action force direction vector lie was then determined by taking the cross product of both unit vectors. A rotation matrix was then formed to rotate the end effector about an ax is normal to the current angle to reach the desired angle. The equation used to accomplis h this is shown in Figure 8-5. It would then return the new rotation matrix that was obtained by multiplying the current rotation matrix and the rotation matrix to achieve the desired fruit/stem angl e. The derivation of this equation is shown in Appendix E. Results and Discussion It was observed from the tests that the fo rce/torque sensor was able to provide a reasonable estimate of the stem angle when us ing a vertically oriented fruit. It was observed that the reaction forces at the stem/f ruit interface were more representative of the angle between the anchori ng point on the tree and the stem /fruit interface rather than the angle of the stem itself. This anc horing point was usually located around the connection between the stem and the main suppor ting branch. This agrees with the static analysis of the system after the compliance in the stem has been removed. The stem then acts as a frame transferring forces between the fruit/stem interface and the main supporting point. The variance in the locat ion of the main supporting point is due to

PAGE 134

120 variability in the compliance of the main br anch and the stem. The results from the preliminary tests are given in Table 8-1. The force control model uses the fruit/ste m angle detection method that assumes a vertically oriented fruit axis. This may not always be a correct assumption as fruit may hang in varying orientations, pa rticularly if they are located in clusters. However, the added sensing technology necessary to determin e the fruit orientation would increase cost and potentially increase the harvesting time through more complex control. Ultimately the model would need to be implemented in orde r to test its performance. It could then be compared against the performance of an open loop control. If sensing technology was able to be integrated into th e control that could determine the fruit orientation, then the vertical axis used in the current model coul d be replaced with the actual fruit axis. The force control model takes as inputs a vector of sensed forces (FT) in x, y, and z as well as the current rotation matrix (Rotationold). Based on the determined angle between the fruit axis and the force vector, the func tion returns the modified rotation matrix (Rotationnew) to acheive the desired angle of 110°. A flowchart of this model is shown in Figure 8-6. Matlab (2004) Code for Force Control Model %ForceControl.m Inputs a vector of se nsed forces (FT) in x, y, and z % as well as the current rotation matrix (Rotationold). Based on the % determined angle between the fruit axis and the force vector, the % function returns the modified rota tion matrix (Rotati onnew) to acheive % the desired angle. function [Rotationnew]= Fo rceControl(Rotationold,FT) %The fruit is assumed to be hanging with its axis parallel to gravity which % acts in the negative x directi on in the robot's global coordinate % frame. This unit vector has then been rotated to reflect the current % orientation of the end effector. fruitaxis=[-1,0,0]*Rotationold;

PAGE 135

121 %Unitize the fruit axis vector and fo rce direction vector and obtain the % angle between them where the FT forces have been realigned so as to % agree with the robot 's coordinate frame. unitforce=unit([FT(2),-FT(1),FT(3)]); anglecos=dot(unitforce,fruitaxis); angle=acos(anglecos)*180/pi; %Determine the difference between the current angle and the desired angle. diff=(110-angle)*pi/180; %Determine the normal to the plane that both the fruit axis vector and the % force vector lie in. normal=cross(unitforce,fruitaxis); %Determine the rotation matrix to ro tate by additional diff about an axis % along the determined normal. R=[normal(1)*normal(1)*( 1-cos(diff))+cos(diff),... normal(1)*normal(2)*(1-co s(diff))-normal(3)*sin(diff),... normal(1)*normal(3)*(1-co s(diff))+normal(2)*sin(diff);... normal(1)*normal(2)*(1-co s(diff))+normal(3)*sin(diff),... normal(2)*normal(2)*( 1-cos(diff))+cos(diff),... normal(2)*normal(3)*(1-co s(diff))-normal(1) *sin(diff);... normal(1)*normal(3)*(1-co s(diff))-normal(2)*sin(diff),... normal(2)*normal(3)*(1-co s(diff))+normal(1)*sin(diff),... normal(3)*normal(3)*( 1-cos(diff))+cos(diff)]; %Calculate the new rotation matrix Rotationnew=Rotationold*R; Integration of the Force Control Model in to the Overall Control of the Manipulator The developed control model could then be integrated into the overall control of the manipulator (Figure 8-7). The control of the manipulator w ould integrate path planning, obstacle avoidance, singularity avoida nce and the force/torque based motion optimization. Path planning would incorporate knowledge of the current position as well as necessary future positions such as the curre nt target fruit locati on, the fruit collection receptacle location and the location of the next target fruit. This would then be used to develop a fruit picking strategy that would harv est the fruit in such an order so as to

PAGE 136

122 minimize the amount of time necessary to co mplete the entire harvesting process. Feedback from sensors such as cameras and range sensors would be used to provide information on the environment the robot is ope rating in. A constant gain, K1 would be associated with this control. Obstacle avoidance would likewise incorporate feedback from sensors such as cameras or range sensors to determine the presence of obstacles that may impede the manipulator's path. A gain, K2, would be a ssociated with this c ontrol that would be inversely proportional to the distance betw een the manipulator and the obstacle. Singularity avoidance would be based on the kinematics of the manipulator and would have a gain K3 that would function in a similar manner to gain K2. Lastly, the force control model would use f eedback from the force torque sensor to optimize the harvesting portion of the motion of the manipulator. A gain, K4, would be associated with the control. The gain w ould be 0 whenever the manipulator was not performing a harvesting motion, before sufficien t forces had developed and after forces had dropped off indicating the fruit had been ha rvested. The initial de lay is necessary in order to allow the stem to reorient itself in the direction of the harvesting motion as well as for the magnitude of harvesting forces to be large enough to re duce the influence of any noise. This point could be determined either through a set force level, distance traveled or rate of change of the harvesting forces. Because of the amount of variability in the detachment force and distance traveled observed in chapter six, a threshold based on the rate of change of harvesting forces ma y be the most effective. Alternatively the rate of change of the harvesting angle coul d be used to determine when the stem had finished reorienting itself. During the in terim harvesting period where the specified

PAGE 137

123 threshold has been reached but the fruit has not yet been harvested a constant gain will be associated with the control. All four of th ese control elements will then be combined to form a single motion command to be sent to the manipulator. The specific values assigned to each of the gains would be determined based on the desired overall performance. These could then be furthe r tuned during testing. After encountering manipulator nonlinearities and the operating e nvironment, the actual motion will be fed back to the four control elements. Conclusions Stem angle detection was shown to be possibl e if the fruit orientation is able to be accurately determined. The proposed method of determining fruit orientation assumes that the fruit are oriented vertically. Inaccu racies in this assumption will then influence the accuracy of the control. Further studies would need to be performed in the field to study the induced error of this assumption. Conclusions co uld then be drawn as to the whether sufficient improvement in the contro l is observed over ope n loop control. Additional sensing technology may be able to more effectively determine the fruit orientation in the future which could then replace the assumption of a vertical fruit axis in the control model.

PAGE 138

124 Figure 8-1. Hybrid control block diagram. Figure 8-2. Fruit, stem and reaction forces. Figure 8-3. Fruit and reaction forc es with the fruit axis assume d to be oriented vertically.

PAGE 139

125 Figure 8-4. Stem angle detection tests with the white string connect ing the fruit and the main supporting branch. ) cos( )) cos( 1 ( ) sin( )) cos( 1 ( ) sin( )) cos( 1 ( ) sin( )) cos( 1 ( ) cos( )) cos( 1 ( ) sin( )) cos( 1 ( ) sin( )) cos( 1 ( ) sin( )) cos( 1 ( ) cos( )) cos( 1 ( ) , (2 2 2 z x z y y z x x z y y z y x y z x z y x xr r r r r r r r r r r r r r r r r r r r r r R Figure 8-5. Formula for forming the rotation matrix for a rotation of about axis r represented by unit vector components rx, ry, and rz.

PAGE 140

126 Figure 8-6. Flowchart of the force control mode l with assumption of a vertical fruit axis. Figure 8-7. Block diagram of an integration of the force control model into the overall control of the manipulator.

PAGE 141

127 Table 8-1. Observed and calculated angles from preliminary tests. Fruit #1Fruit #2Fruit #3 Observed Fruit/Stem Angle (°)353840 Calculated Fruit/Stem Angle (°)43.1342.746.7 Observed Fruit/Main Branch Connection (°) 43.53851 Error in Calculating Fruit/Stem Angle (°)8.134.76.7 Error in Calculating Fruit/Main Branch Connection (°) 0.374.74.3

PAGE 142

128 CHAPTER 9 SUMMARY AND CONCLUSIONS Because of the high cost and limited availa bility of human labor and the need to harvest as inexpensively as possible in a tim ely manner, it is incr easingly desirable to harvest fruit robotically especially for fresh fr uit. An integral part of the success of a robotic harvester is the end eff ector. Previous efforts to de velop robotic citrus harvesting end effectors as well as end effectors for other selected commodities were reviewed in chapter two. Although many end effectors ha ve been proposed or developed, none has emerged as a successful, economically viable solution. The specific objectives of the studies in this dissertation were therefore as follows: 1. Determine the safe grasping limits for a grasping robotic citrus harvester end effector through physical properties tests. 2. Determine the most effective harvesti ng motion through tests with a 6-axis force/torque sensor. 3. Design a robotic citrus harvesting end eff ector based on the results of the physical properties and harvesting motion tests. 4. Build and test the designed robotic ha rvesting end effector to determine its effectiveness based on the design criteria 5. Determine if stem location detection ma y be used to optimize the harvesting motion without the need for additional movements. 6. Construct a force control model that woul d utilize feedback from a force/torque sensor to control a robotic citrus harvesting manipulator.

PAGE 143

129 Physical Properties of Oranges in Response to Applied Gripping Forces for Robotic Harvesting In order to understand more fully how an orange would respond to a robotic harvester, studies were conducted that br idged the gap between previously conducted puncture studies and previously conducted bur st studies. Field run (unwashed, unwaxed) Valencia oranges ( Citrus sinesis cv. Valencia) were test ed on March 30, May 15, and June 16, 2004, using an Instron universal testi ng machine. The punch sizes used for the puncture tests were 0.323, 0.632, 0.964, 1.27, 1.90, a nd 2.540 cm. Burst tests were also performed with the whole fru it under flat plate compression. As expected, the force required to puncture or burst a fruit is direc tly related to the contact area. This is a function of two variables: the punch diameter used, and the radius of curvature of the fruit. Based on the results of these test s, a model was developed that relates punch diameter to puncture force. It was also not ed that as the punch di ameter size increased, the punch diameter term in the model approach ed zero. This left th e puncture force term as a function of the radius of curvature only. This correlated well with physical observations in that punch diameters beyond 2.540 cm approached the behavior of a flat plate, where puncture force was no longer a func tion of the punch diam eter but solely of the fruit properties. Based on the results of these tests, recommendations were made for the design of a grasping robotic citrus harveste r end effector. The end effect or should be made so that the grasping of the fruit does not exceed th e bursting limits or the puncture limits, where the portion of the end effector in contact with the fruit may be expressed in equivalent punch diameter. This may be obtained by using the perimeter that is in contact with the fruit. A model was developed that relate d puncture force to punch diameter and the

PAGE 144

130 radius of curvature of the fruit. If the ra dius of curvature is not known a representative radius of curvature may be used. The mi nimum observed bursting force was 135 N in the parallel plate test. This then defines the upper loading limit. Any further loading limitations would be based on the geometry of the grasping portion of the end effector. Further studies need to be conducted using va rious shaped punches in order to further define to what extent each of the geometrical properties plays a ro le in the puncturing of the fruit, and the impact of these roles on th e design of a grasping robotic citrus harvester end effector. Studies in the Optimization of Harv esting Motion Mechanics using a 7-DOF Manipulator with a 6-Axis Force/Torque Sensor The studies conducted previously by other researchers did not completely examine the influence of harvesting motions on each of the following: fruit damage, distance traveled and harvesting forces. Therefore st udies were conducted in order to determine the optimum harvesting motion that would minimize damage to the fruit and tree, minimize loads transferred to the manipulator, and complete the motion in as little time as possible. The studies were conducted in a commerci al grove using a prototype harvesting system and a six-axis force/torque sensor in order to examine the impact of various harvesting motions on th e fruit, tree and manipulator. The specific tests included imparting a range of angles be tween the stem and the fruit axis, the effect of applied torsion about the stem axis and the effect of a pretensioned stem on a harvesting motion. A handheld linear pull test was also performed on seven different test dates in order to evaluate the change in fruit detachment fo rce, fruit mass and fruit diameter over the harvesting season. A second handheld linear pu ll test was performed to evaluate the influence of detachment force on distance traveled.

PAGE 145

131 Based on the results of the ha rvesting motion tests; the di stance traveled, harvesting forces and fruit damage may be minimized by imparting a 90° angle between the fruit and stem. This may be accomplished by performing a motion to impart at least a 110° angle between the fruit and the stem if no feedback is ava ilable to confirm whether the desired angle has been achieved. The results also indicate that a rotation of 180° about the stem axis may also minimize the fruit damage, although having no significant impact on the distance traveled or harvesting fo rces. Although a motion based on sensed magnitude of the harvesting forces from a fo rce/torque sensor, such as the secondary pitch motion, may be effective if further developed, the am ount of variability in the detachment force, fruit size, fruit mass, bran ch compliance, fruit/stem orientation, and so on may prevent any observed improvement. Th ere is however still potential that the sensed direction of the force vector may be utilized to ensure that the desired 90° angle between the fruit and stem is achieved. Robotic Citrus Harvesting End Effector Development Although many end effectors have been pr oposed or developed, none has emerged as a successful, economically viable solution. In an effort to develop such an end effector, a grasping robotic ci trus harvester end effector was developed based on the results of the physical properties and harves ting motion tests. The physical properties tests were used to determine the safe gras ping limits and the harvesting motion tests were used to determine the detachment forces th at the end effector w ould experience. The design criteria were as follows: Lightweight (<17.78 N (4 lb)), based on th e manipulator payload capacity (155 N) and the weight of the end effector used in the harvesting motion tests (24.5 N) Cost-effective (<$1000, not including labor), based on economic estimates of the overall cost of an economically viable harvesting system

PAGE 146

132 Fast (<0.5 sec), based on a total cycle time of 2 sec Produce minimal damage on fruit and tree Able to reach fruit inside the canopy Able to integrate sensing technology such as vision, ultrasonic and infrared sensors The end effector was then tested on its ab ility to successfully harvest fruit as well as its performance against each of the design criteria. This included performing field tests that examined the ability of the end e ffector to harvest fruit using three different harvesting motions. The end effector was able to harvest fruit without damaging the fruit or tree, reach inside the canopy and integrat e the necessary sensing technology. It was also able to harvest fruit ev en if the sensing technology a nd manipulator placed the end effector moderately out of pos ition. It was however heavier, slower and more expensive than was desired. Although the end effector did not meet all of these criteria, it was able to harvest fruit successfully and the integrat ion of the sensors improved the positioning control over the end effector used in the harvesting motion tests of chapter six. The overall weight and cost could be reduced w ith further refinements of the design. The speed is not a concern currently as the overall robotic harvesti ng system is not capable of the final design speeds. However, this is an ar ea that would need to be addressed prior to implementation in an improved harvesting sy stem. Either the actuator or the drive mechanism would possibly need to be redesigne d in order to improve the actuation time. Alternatively, it is possible that the previous pneumatic end effector could be redesigned to better integrate the sensors and use the new fingers to improve the grasping ability. Future designs may also incorporate the abil ity to perform 180° rota tions about the stem axis and 110° rotations about an axis perpendi cular to the stem axis. These are motions that were concluded in chapter six to yield the best results. The ability of the end effector

PAGE 147

133 to perform these motions would eliminate the need for the manipulator to perform them, which may take increased time and operationa l space. Although there are improvements that could be made to the end effector, the de sign process of basing an end effector on the results of the tests conducted in chapters fi ve and six yielded an end effector which was able to withstand the harvesting forces it experienced while not damaging the fruit or the tree. Stem Detection Studies and Force Control Model Development Previous efforts to harvest citrus fruit by using a force/torque sensor to determine the stem location required the manipulator to perform a series of movements. In the optimization of a fruit harvesting motion, it may not be possible to perform these motions. Studies were therefore conducted to examine the possibility of determining the angle between the fruit axis and the stem based on the feedback from a force/torque sensor. Detection of this angle was shown to be possible if the orientation of the fruit axis is able to be accurately determined. The proposed method of determining fruit orientation assumes that the fruit are oriented vertically. Inaccuracies in this assumption will then influence the accuracy of the contro l. The use of this angle detection in the optimization of the harvesting motion was then presented. The manner in which this harvesting motion optimization would be inte grated into the overall control of the manipulator was also presented. Further studies would need to be performed in the field to study the induced error of th is assumption. Conclusions coul d then be drawn as to the whether sufficient improvement in the contro l is observed over ope n loop control. Additional sensing technology may be able to more effectively determine the fruit orientation in the future which could then replace the assumption of a vertical fruit axis in the control model.

PAGE 148

134 Overall Conclusions The studies conducted added to the gl obal body of knowledge on the development of a robotic citrus harveste r end effector. The physical properties studies performed established safe grasping limits for a robotic harvester. The harvesting motion studies performed proposed an optimum harvesting motion as well as the harvesting forces, distance traveled and fruit damage experien ced for various harvesting motions. An end effector based on these studies was also developed and tested that was unlike the previously developed end effectors with repor ted results. Although the end effector did not meet all of the design cr iteria, it was able to harvest fruit successfully. Further refinements of the design may yield an end e ffector that does meet all of the criteria. Finally a method of using force feedback to optimize the motion of a robotic harvester was developed. If integrated into the overall control of a robotic harvester, this may further improve its performance. Future Work The development of a robotic citrus harves ting end effector and force control model using physical properties and harvesting motion tests was presented. Although the studies in this dissert ation have attempted to accomplish all of their objectives, future work still needs to be performed to further i nvestigate aspects of these areas of research. Further studies need to be conducted using va rious shaped punches in order to further define to what extent the geometrical prope rties of the punch such as area and perimeter play roles in the punctu ring of the fruit, and the impact of these roles on the design of a grasping robotic citrus harveste r end effector. Studies should also be done to examine the impact of velocities and accel erations in determining an optimum harvesting motion as

PAGE 149

135 well as the ability of the harvesting system to use force/torque feedback to control the harvesting motion. As the developed end effector did not meet all of the design criteria, it would be necessary to further refine the design. M odifying the actuation mechanism of the end effector may yield a design that meets all of the design criteria. The end effector was based on the concept of being able to grasp the orange for manipulation by the manipulator. Further actuation mechanisms c ould be incorporated into a future design that would allow for a 180° ro tation of the fruit about the stem axis and a motion to impart a 90° angle between the fruit and the stem. Further studies also need to be performed to examine the impact of the assu mption of a vertically oriented fruit on the accuracy of the force control model. Compar ative studies could then be performed to examine whether this method of harvesti ng motion optimization provides sufficient improvement over open loop control. The work presented here along with the described necessary future work should help the scien tific community continue to progress towards the development of a successful robotic citrus harvester.

PAGE 150

136 APPENDIX A YIELD POINT DETERMINATION Introduction Although the puncture and burst limits were de termined in chapter five, citrus fruit may be damaged prior to this force level being reached. This damage may not be readily apparent, but would appear later in the storag e process. It is therefore important to determine if the fruit would be injured at a force level below that of the puncture force level. Several methods for determining injury in fruits have been used. Chuma et al. (1978) used a near-ultraviolet lamp to detect peel oil that had b een released from the broken outer tissue of oranges. The peel oi l would flouresce in the visible spectrum. Chen et al. (1989) used nuclear magnetic resonance (NMR) also known as magnetic resonance imaging (MRI) to detect bruises in apples, peaches and pears as well as to detect dry regions in oranges. Varith et al. (2001) used a thermal camera to detect bruises in apples. Aneshansley et al. (1997) examin ed the reflectance properties of bruised and punctured apples versus unbruise d and unpunctured apples to detect damaged fruit. The wavelengths examined ranged from 540 to 1030 nm . Diener et al. ( 1970) and Schatzki et al. (1997) used X-ray imaging to detect bruises in apples. Ismail and Miller (1988) described several methods of injury detection to oranges including measuring CO2 production, weight loss measur ement and staining of injured surfaces. Two methods of staining were descri bed. One used indicator paper to denote areas where citric acid was present, the ot her used a coloring dye that reacted with

PAGE 151

137 exposed tissue. The dye used was a 0.5% solution of 2,3,5-triphenyl-2H-tetrazolium chloride (TTC). Using TTC dye to stain ar eas where living tissue was exposed was first proposed by Roistacher et al. (1956). It was later used by Ismail and Miller (1988), Eaks (1961) and Fornes et al. (1993). Fornes et al. (1993) used the dye to detect damage that may have occurred to fruit as a result of robotic harvesting. As it was desired to examine any injuries that may have occurred at a specific point, namely the area of loading, a general method of damage indication would not be feasible. This eliminated the possibility of using CO2 production, weight loss measurement and reflectance properties. As the use of NMR, X-ray and thermal imaging all require expensive pieces of equipment it was decided not to use these methods. Preliminary studies were done to examine the po ssibility of using near -ultraviolet light to determine injured areas, however this was met with little success. It was therefore decided to use the TTC dye staining method. Objective To further define the safe limits of hand ling by determining the point at which the peel of the fruit is first broken. Materials and Methods Three punch sizes were chosen from the punc h sizes used in the puncture tests of chapter five. The fruit were then loaded to one of several predeter mined force levels in the same manner as the puncture test. Th ese predetermined levels were based on the results of the puncture test. The smallest force level used at each punch size was approximately the mean puncture force minus th ree standard deviations. The force levels then increased in increments of 4.4 N up to the mean puncture force minus two standard deviations. The other force levels used were at the mean puncture force minus one

PAGE 152

138 standard deviation, the mean puncture for ce and the mean puncture force plus one standard deviation. The largest concentrati on of force levels was chosen well below the puncture force range as the variability in puncture force would require a grasping end effector to operate well below the mean puncture force. The force levels chosen for each of the corresponding punch sizes used is shown in Table A-1. The fruit used were of the Valencia variety ( Citrus sinesis cv. Valencia) and were obtained from a commercial grove located in Umatilla, Florida. The fru it were freshly harvested, and thus they had not been washed, waxed, or sized. After ha rvest, the fruit were placed in the same storage conditions as in the puncture tests re ported in chapter five with the tests being conducted over a period of 3 days. Ten fruit were tested per punch size per force level. After loading, the fruit were submerge d in a 0.5% solution of 2,3,5-triphenyl-2Htetrazolium chloride (TTC) dye for 30 minutes. After removal from the solution the fruit were placed into a tap water bath for 5 mi nutes. The fruit were then placed in double layers of brown paper bags. After sitting fo r at least 8 hours at room temperature, the fruit were examined to see if any staining had appeared to denote a break in the fruit peel. The dye reacts with any living tissue that was exposed when the outer peel of the fruit is ruptured. The procedure used was base d on that of Ismail and Miller (1988). Results and Discussion A summary of the dimensional data fo r the fruit used in the yield point determination tests is given in Table A-2. Th e fruit were larger than those used in the puncture tests however they were still characteristic of the Valencia variety. Staining related to the rupture of the peel by the punch was not observed until one standard deviation below the mean puncture force where some fruit exhibited slight staining that may have been caused by the punc h. The levels wher e slight staining and

PAGE 153

139 pronounced staining began to appear are shown in Table A-3. All fruit exhibited some staining as a result of injury from hand ling. There was no pronounced staining observed for the 1.27 cm punch size. The lack of obser vance of staining substantially below the range of puncture forces indicates that it is sufficient to use the puncture force distribution as the limits of safe handling. An example of slight staining is shown in Figure A-1 and an example of pronounced staini ng is shown in Figure A-2. Some fruit were punctured prior to the specified force level being reached. The number of fruit punctured at each fruit where puncturing occurred is listed in Table A-4. Conclusions The test results indicate that there is no point substantially below the puncture force range where the peel is damaged. It is sufficient therefore to design a grasping end effector based on the results of the puncture te sts. If the end effector is designed such that the grasping forces are below the range of the puncture force, then it will also be below the range of the initial pe el puncture. This is expect ed as once the peel has been initially broken, the resistance strength of th e peel to puncture is substantially reduced.

PAGE 154

140 Figure A-1. Slight st aining around punch area. Figure A-2. Pronounced st aining around punch area. Table A-1. Force levels used at each corresponding punch size used. Punch Size (cm)Force Levels (N)Mean Puncture Force (N) From Chapter Five 0.96431,36,40,44,58,67,8068.92 1.2731,36,40,44,49,53,67,85,10285.57 1.967,71,76,80,85,89,93,116,138,160135.77 Table A-2. Dimensional summary of fruit used in yield point determination test. MeanStd. Dev.MeanStd. Dev.MeanStd. Dev. 260236.9544.387.7050.5344.0050.304 Mass (g)Average Diameter (cm)Radius of Curvature (cm) Number of Fruit Tested

PAGE 155

141 Table A-3. Force levels wh ere staining first appeared for each punch size used. Punch Size (cm) Force Levels Where Slight Staining First Appeared (N) Force Levels Where Pronounced Staining First Appeared (N) Mean Puncture Force From Chapter Five (N) 0.964586768.92 1.2767N/A85.57 1.9116160135.77 Table A-4. Number of fruit punctured prior to force level being reached. Punch Size (cm) Force Level (N) Number of Fruit Punctured (Out of 10) 0.964672 0.964807 1.27852 1.271024 1.91381 1.91606

PAGE 156

142 APPENDIX B HARVESTING MOTION CODE FOR MANIPULATOR CONTROL The following is the code used in the rotati onal tests to control the movement of the manipulator. For the retraction a ngle tests, there were some di fferences in the code used. The pitch and roll were not zeroed prior to executing the motion. The roll was zero regardless due to the control of the manipulator keeping the camera level. The absence of zeroing the pitch resulted in the wide range of executed angles. In the rotational test, closing the gripper was performed separately fr om the motions in order to ensure a proper grasp of the fruit. In the re traction test, this was combined with the 20° pitch movement. In both tests the 20° pitch was performed separa tely from the retraction so as to allow for the beginning of force /torque data recording. The secondary pitch test was performed in the same manner as the code listed below except for the execution of the retract ion. The retraction portion of the code was in a loop that looked for a flag to be se t by the force/torque data collection code indicating that a predetermine d amount of tension had been reached. Once 22.24 N (5 lb) of tension had been reached, a command was se nt to halt the robot and then execute an additional 30° pitch. The new end effector trial tests were performed in the same manner as the code listed below with the exception of closing th e gripper. The stepper motor controller required a command to be sent through the serial port to open or close the gripper. This command then replaced the output to the DAQ listed in the code. //Definition of velocities //Velocity during travel double vel[] = {5.0, PID2, 0.0}; //Final velocity double fVel[] = {0.0, 0.0, 0.0}; //Zero pitch and roll so that all repetitions are the same //Obtain current pitch from the robot controller double pitch = m_RbtFdbk.PyrRad[0]; //Obtain current roll fr om the robot controller double roll = m_RbtFdbk.PyrRad[2]; //Form matrix of desired new position and orientation //Xt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Xt represents the first column of the 3 //x 3 orientation matrix. Bo th m_Desired Matrix and

PAGE 157

143 //m_GripperToolMatrix were prev iously defined as homogeneous //matrices. m_DesiredMatrix.Xi = cos(-r oll)*m_GripperToolMatrix.Xi – sin(-roll)*m_GripperToolMatrix.Xj; m_DesiredMatrix.Xj = cos(-pitch)*sin(-roll)*m_GripperToolMatrix.Xi + cos(-pitch)*cos(-roll)*m_GripperToolMatrix.Xj – sin(-pitch)*m_GripperToolMatrix.Xk; m_DesiredMatrix.Xk = sin(-pitch)* sin(-roll)*m_GripperToolMatrix.Xi + sin(-pitch)*cos(-roll)* m_GripperToolMatrix.Xj + cos(-pitch)*m_GripperToolMatrix.Xk; //Yt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Yt represents the second column of //the 3 x 3 orientation matrix. Both m_Desired Matrix and //m_GripperToolMatrix were //pre viously defined as homogeneous //matrices. m_DesiredMatrix.Yi = cos(-r oll)*m_GripperToolMatrix.Yi – sin(-roll)*m_GripperToolMatrix.Yj; m_DesiredMatrix.Yj = cos(-pitch)*sin(-roll)*m_GripperToolMatrix.Yi + cos(-pitch)*cos(-roll)*m_GripperToolMatrix.Yj – sin(-pitch)*m_GripperToolMatrix.Yk; m_DesiredMatrix.Yk = sin(-pitch)* sin(-roll)*m_GripperToolMatrix.Yi + sin(-pitch)*cos(-roll)* m_GripperToolMatrix.Yj + cos(-pitch)*m_GripperToolMatrix.Yk; //Zt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Zt represents the third column of the 3 //x 3 orientation matrix. Bo th m_Desired Matrix and //m_GripperToolMatrix were prev iously defined as homogeneous //matrices. m_DesiredMatrix.Zi = cos(-roll)*m_GripperToolMatrix.Zi – sin(-roll)*m_GripperToolMatrix.Zj; m_DesiredMatrix.Zj = cos(-pitch)*sin(-roll)*m_GripperToolMatrix.Zi + cos(-pitch)*cos(-roll)*m_GripperToolMatrix.Zj – sin(-pitch)*m_GripperToolMatrix.Zk; m_DesiredMatrix.Zk = sin(-pitch)* sin(-roll)*m_GripperToolMatrix.Zi + sin(-pitch)*cos(-roll)*m_GripperToolMatrix.Zj + cos(-pitch)*m_GripperToolMatrix.Zk; //XYZ position in tool frame (no change) where m_DesiredMatrix was //previously defined as a homogeneous matrix. m_DesiredMatrix.X = 0; m_DesiredMatrix.Y = 0; m_DesiredMatrix.Z = 0;

PAGE 158

144 //Move to desired position and orientation as expressed by m_DesiredMatrix in //tool frame with specified velocities m_Robot.Setpoint(TOOL, &m_DesiredMatrix, vel, fVel); //Close gripper //Output to DAQ bool output[8] = {1,0,0,0,0,0,0,0}; //Perform 20° pitch with zero roll //Roll angle double r = 0*DEG2RAD; //Pitch angle double p = -20*DEG2RAD; //Form matrix of desired new position and orientation //Xt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Xt represents the first column of the 3 //x 3 orientation matrix. Bo th m_Desired Matrix2 and //m_GripperToolMatrix were //pre viously defined as homogeneous //matrices. m_DesiredMatrix2.Xi = cos(r)*m_GripperToolMatrix.Xi sin(r)*m_GripperToolMatrix.Xj; m_DesiredMatrix2.Xj = cos(p)*s in(r)*m_GripperToolMatrix.Xi + cos(p)*cos(r)*m_GripperToolMatrix.Xj sin(p)*m_GripperToolMatrix.Xk; m_DesiredMatrix2.Xk = sin(p)*sin(r)*m_GripperToolMatrix.Xi + sin(p)*cos(r)*m_GripperToolMatrix.Xj + cos(p)*m_GripperToolMatrix.Xk; //Yt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Yt represents the second column of //the 3 x 3 orientation matrix. Both m_Desired Matrix2 and //m_GripperToolMatrix were previ ously defined as homogeneous //matrices. m_DesiredMatrix2.Yi = cos(r)*m_GripperToolMatrix.Yi sin(r)*m_GripperToolMatrix.Yj; m_DesiredMatrix2.Yj = cos(p)*s in(r)*m_GripperToolMatrix.Yi + cos(p)*cos(r)*m_GripperToolMatrix.Yj sin(p)*m_GripperToolMatrix.Yk; m_DesiredMatrix2.Yk = sin(p)*sin(r)*m_GripperToolMatrix.Yi + sin(p)*cos(r)*m_GripperToolMatrix.Yj + cos(p)*m_GripperToolMatrix.Yk; //Zt in tool frame where m_GripperT oolMatrix is the current position and //orientation of the end effector and Zt represents the third column of the 3

PAGE 159

145 //x 3 orientation matrix. Bo th m_Desired Matrix2 and //m_GripperToolMatrix were //pre viously defined as homogeneous //matrices. m_DesiredMatrix2.Zi = cos(r)*m_GripperToolMatrix.Zi sin(r)*m_GripperToolMatrix.Zj; m_DesiredMatrix2.Zj = cos(p)*s in(r)*m_GripperToolMatrix.Zi + cos(p)*cos(r)*m_GripperToolMatrix.Zj sin(p)*m_GripperToolMatrix.Zk; m_DesiredMatrix2.Zk = sin(p)*sin(r)*m_GripperToolMatrix.Zi + sin(p)*cos(r)*m_GripperToolMatrix.Zj + cos(p)*m_GripperToolMatrix.Zk; //XYZ position in tool frame (no change) where m_DesiredMatrix2 was //previously defined as a homogeneous matrix. m_DesiredMatrix2.X = 0.0; m_DesiredMatrix2.Y = 0.0; m_DesiredMatrix2.Z = 0.0; //Move to desired position and orientation as expressed by m_DesiredMatrix2 in //tool frame with specified velocities m_Robot.Setpoint(TOOL, &m_DesiredMatrix2, vel, fVel); //Retract //Angle that the manipulator moves down and back on as expressed by alpha in //degrees. double retractangle = alpha*DEG2RAD; double dX = 0.0; double dY = 5*cos(retractangle); double dZ = -5.0*sin(retractangle); //Form matrix of desired new position and orientation //Xt in tool frame (no change) where Xt represents the first column of the //3 x 3 orientation matrix. m_Desire dMatrix3 was previously defined as a //homogeneous matrix. m_DesiredMatrix3.Xi = 1.0; m_DesiredMatrix3.Xj = 0.0; m_DesiredMatrix3.Xk = 0.0; //Yt in tool frame (no change) where Yt represents the second column of //the 3 x 3 orientation matrix. m_DesiredMatrix3 was previously defined //as a homogeneous matrix. m_DesiredMatrix3.Yi = 0.0; m_DesiredMatrix3.Yj = 1.0; m_DesiredMatrix3.Yk = 0.0;

PAGE 160

146 //Zt in tool frame (no change) where Zt represents the third column of the //3 x 3 orientation matrix. m_Desire dMatrix3 was previously defined as a //homogeneous matrix. m_DesiredMatrix3.Zi = 0.0; m_DesiredMatrix3.Zj = 0.0; m_DesiredMatrix3.Zk = 1.0; //XYZ position in tool frame wher e m_DesiredMatrix3 was previously //defined as a homogeneous matrix. m_DesiredMatrix3.X = 0.0; m_DesiredMatrix3.Y = dY; m_DesiredMatrix3.Z = dZ; //Move to desired position and orientation as expressed by m_DesiredMatrix3 in //tool frame with specified velocities m_Robot.Setpoint(TOOL, &m_DesiredMatrix3, vel, fVel);

PAGE 161

147 APPENDIX C CODE FOR FORCE/TORQUE DATA COLLECTION The following is the code used for collecting the force/torque data. The first section contains the pseudo-code for how the da ta is collected up to the point that it is written to a file. The full code for what was wr itten in each data file is then listed. There were slight variations in th e code used for the secondary pitch test and the retraction angle test. In the secondary pitch test the magnitude of the resultant force vector was constantly monitored. Once this magnitude wa s equal to or greater than 22.24 N (5 lb), a flag was set and sent to the program cont rolling the motion of the manipulator. The harvesting motion program was monitoring for this flag a nd once it was detected, the secondary pitch was executed. In the retract ion angle test the low pass filter used on the voltages had a cutoff frequency of 100 Hz. //Pseudo-code Read in voltages from force/ torque DAQ sampled at 1000 Hz Filter voltages in low pass filter with a cutoff frequency of 35 Hz 2 sec after initializing the program, the recorded voltages are set as the bias Compute FT using calibration data and the temperature compensation measurements When the save button in the dialog box is pressed, the following are saved until the button is pressed again: Joint values Tool point position and orie ntation in X,Y,Z and P,Y,R respectively DAQ voltages Force torque values When the save button is pressed again, a dialog box opens requesting the name of the file to be saved. The following are then written to a Matlab *.m file, with the explanatory sections of the code in italics //Code for writing data to file //Clear screen and variables file.WriteString("clear\nclc\n\n"); //Write number of samples file.WriteString("NumOfSamp = "); sprintf( value, "%d", m_nNumOfSamp); file.WriteString(value); file.WriteString(";\n");

PAGE 162

148 //Write vector of time from 0 to (sample tim e)*(number of samples-1) at increments of //sample time file.WriteString("Time = ["); time = 0; sprintf( value, "%d", time); file.WriteString(value); file.WriteString("; "); for(i=1; i
PAGE 163

149 for(i=0; i
PAGE 164

150 file.WriteString(value); file.WriteString("; "); } file.WriteString("];\n"); //Write matrix of FT computed values file.WriteString("FT = ["); for(i=0; i
PAGE 165

151 APPENDIX D SCHEMATIC OF DESIGNED INFRARED SENSOR Figure D-1. Schematic of IR sensor designe d by technician Greg Pugh and integrated into the end effector of chapter 7.

PAGE 166

152 APPENDIX E DERIVATION OF ROTATION MATRIX USED IN FORCE CONTROL MODEL The control model in chapter 8 uses a formula in Figure 85 to perform a rotation of angle about an axis or rotation re presented by unit vector r = [rx, ry, rz]. This formula was derived using the method of Crane III and Duffy (1998). First a coordinate system, A, was introduced. A second coordinate syst em, B, was introduced that was initially aligned with A, but was rotated by angle about r. The rotation matrix, RA B, was then solved for to obtain the formula of Figure 85. In order to do this a third coordinate system, C, was introduced whose z axis was para llel to r. The rotation matrix between A and C,RA C, was then expressed as follows where the components of the vectors a and b are unknown: z z z y y y x x x A Cr b a r b a r b a R (Eq. E-1) A fourth coordinate system, D, was introdu ced that was initially aligned with C but was rotated by angle about r. The rotation matrix between C and D,RC D, was then expressed as follows: 1 0 0 0 cos sin 0 sin cos RC D (Eq. E-2) The rotation matrix between A and B was then expressed as follows: R R R RD B C D A C A B (Eq. E-3). Noting that R RB D A Cand so T A C D BR R , then the right side of this equation expanded became: z y x z y x z y x z z z y y y x x x A Br r r b b b a a a r b a r b a r b a R 1 0 0 0 cos sin 0 sin cos (Eq. E-4) Expanded still further with c representing cos( ) and s representing sin( ) this became:

PAGE 167

153 2 2 2 2 2 2 2 2 2... ) ( ... ... ) ( ... ... ) ( ... ... ) ( ... ... ) ( ... ... ) ( ...z z z z y z y z y z y z y z x z x z x z x z x z y y z y z y z y z y y y y x y x y x y x y x z x x z x z x z x z y x x y x y x y x y x x x A Br b a c r r a b b a s b b a a c r r a b b a s b b a a c r r a b b a s b b a a c r b a c r r a b b a s b b a a c r r a b b a s b b a a c r r a b b a s b b a a c r b a c R (Eq. E-5) Since the first row of Eq. E-1 was a unit vector the following was written: 12 2 2 x x xr b a (Eq. E-6) This was then used to substitute for 2 2x xb a in the first row, first column element of Eq. E-5 to obtain: cos 1 cos 1 cos2 2 2 1 , 1 x x x A Br r r R (Eq. E-7) Because the rows in Eq. E-1 were ort hogonal to one another, the following was written: 0 y x y x y xr r b b a a (Eq. E-8) Because the third column of Eq. E-1 was the cross product of the other two, the following was written: y x y x za b b a r (Eq. E-9) Using Eq. E-8 and Eq. E-9 the second row, first column element of Eq. E-5 was then written as follows: sin cos 1 sin cos1 , 2 z y x y x z y x A Br r r r r r r r R (Eq. E-10) Because the rows in Eq. E-1 were ort hogonal to one another, the following was written: 0 z x z x z xr r b b a a (Eq. E-11) Because the third column of Eq. E-1 was the cross product of the other two, the following was written: x z x z ya b b a r (Eq. E-12) Using Eq. E-11 and Eq. E-12 the third row, first column element of Eq. E-5 was then written as follows: sin cos 1 sin cos1 , 3 y z x z x y z x A Br r r r r r r r R (Eq. E-13) Using Eq. E-8 and E-9 the first row, sec ond column element of Eq. E-5 was then written as follows: sin cos 1 sin cos2 , 1 z y x y x z y x A Br r r r r r r r R (Eq. E-14) Since the second row of Eq. E-1 was a unit vector the following was written: 12 2 2 y y yr b a (Eq. E-15) Using Eq. E-15 the second row, second column element of Eq. E-5 was then written as follows:

PAGE 168

154 cos cos 1 1 cos2 2 2 2 , 2 y y y A Br r r R (Eq. E-16) Because the rows in Eq. E-1 were ort hogonal to one another, the following was written: 0 z y z y z yr r b b a a (Eq. E-17) Because the third column of Eq. E-1 was the cross product of the other two, the following was written: z y z y xa b b a r (Eq. E-18) Using Eq. E-17 and Eq. E-18 the third row, second column element of Eq. E-5 was then written as follows: sin cos 1 sin cos2 , 3 x z y z y x z y A Br r r r r r r r R (Eq. E-19) Using Eq. E-11 and Eq. E-12 the first row, third column element of Eq. E-5 was then written as follows: sin cos 1 sin cos3 , 1 y z x z x y z x A Br r r r r r r r R (Eq. E-20) Using Eq. E-17 and Eq. E-18 the second row, third column element of Eq. E-5 was then written as follows: sin cos 1 sin cos3 , 2 x z y z y x z y A Br r r r r r r r R (Eq. E-21) Since the third row of Eq. E-1 was a unit vector the following was written: 12 2 2 z z zr b a (Eq. E-22) Using Eq. E-22 the third row, third column element of Eq. E-5 was then written as follows: cos cos 1 1 cos2 2 2 3 , 3 z z z A Br r r R (Eq. E-23) These nine elements can then be combined to form RA B as follows with c representing cos( ) and s representing sin( ): c c r s r c r r s r c r r s r c r r c c r s r c r r s r c r r s r c r r c c r Rz x z y y z x x z y y z y x y z x z y x x A B1 1 1 1 1 1 1 1 12 2 2 (Eq. E-24) This final equation is the same as that found in Figure 8-5.

PAGE 169

155 LIST OF REFERENCES Ahmed, E. M., F. G. Martin, and R. C. Fluck. 1973. Damaging stresses to fresh and irradiated citrus fruits. J Food Sci. 38(2): 230-233. Allotta, B., G. Buttazzo, P. Dario, F. Quaglia, and P. Levi. 1990. A force/torque sensorbased technique for robot harvesti ng of fruits and vegetables. In Proc. Intl. Workshop on Intelligent Robots and Syst., 1: 231-235. Piscataway, N.J.: Institute of Electrical and Electroni cs Engineers (IEEE). Andrews, L. E. 1999. Safety needs for Florid a’s citrus workers: a baseline study. MS thesis. Gainesville, Fla.: University of Florida, Department of Agricultural and Biological Engineering. Aneshansley, D. J., J. A. Throop, and B. L. Upchurch. 1997. Re flectance spectra of surface defects on apples. In Proc. Intl. Conf. Sensors for Nondestructive Testing: Measuring the Quality of Fresh Fruits and Vegetables, 143-160. Ithaca, N.Y.: Natural Resource, Agriculture, and Engineering Service. ASAE Standards. 2000. S368.4: Compression test of f ood materials of convex shape. St. Joseph, Mich.: American Society of Agricultural Engineers (ASAE). Barnes, K. K. 1969. Detachment characteristics of lemons. Trans. ASAE 12(1): 41-45. Black Jr., G. L. 1969. Apparatus for harves ting agricultural crops. U.S. Patent No. 3460330. Boubekri, N., and P. Chakraborty. 2002. R obotic grasping: grippe r designs, control methods and grasp configurati ons-a review of research. Integrated Manufacturing Syst. 13(7): 520-531. Buie Jr., A. P. 1965. Fruit picking and tr ansporting device. U.S. Patent No. 3165880. Bullock, G. E. 1956. Fruit picking apparatus. U.S. Patent No. 2775088. Ceres, R., J. L. Pons, A. R. Jiménez, J. M. Martín, and L. Calderón. 1998. Design and implementation of an aided fru it-harvesting robot (Agribot). Ind. Robot 25(5): 337346. Chen, P., M. J. McCarthy, and R. Kauten. 1989. NMR for internal quality evaluation of fruits and vegetables. Trans. ASAE 32(5): 1747-1753.

PAGE 170

156 Chin, K. S., M. M. Ratnam, and R. Manda va. 2003. Force-guided robot in automated assembly of mobile phone. Assembly Automation 23(1): 75-86. Chuma, Y., T. Shiga, and M. Iwamoto. 1978. Mechanical properties of Satsuma orange as related to the design of a c ontainer for bulk transportation. J. Texture Studies 9(4): 461-479. Churchill, D. B., H. R. Sumner, and J. D. Whitney. 1980. Peel strength properties of three orange varieties. Trans. ASAE 23(1): 173-176. Coggins Jr., C. W. 1969. Gibberellin rese arch on citrus rind aging problems. In Proc. 1st Intl. Citrus Symp. 3: 1177-1185. Riverside, Cal.: University of California. Coggins Jr., C. W., and L. N. Lewis. 1965. Some physical properties of the navel orange rind as related to ripening and to gibberellic acid treatments. J. American Soc. Hort. Sci. 86: 272-279. Connery, W. E. 1971. Fruit-harvestin g machine. U.S. Patent No. 3591949. Coppock, G. E., S. L. Hedden, and D. H. Lenker. 1969. Biophysical properties of citrus fruit related to mechanical harvesting. Trans. ASAE 12(4): 561-563. Crane III, C. D., and J. Duffy. 1998. Kinematic Analysis of Robot Manipulators. Cambridge, United Kingdom: Cambridge University Press. Diener, R. G., J. P. Mitchell, and M. L. R hoten. 1970. Using an x-ray image scan to sort bruised apples. Agric. Eng. 51(6): 356-357, 361. Eaks, I. L. 1961. Techniques to evaluate injury to citrus fruit from handling practices. In Proc. American Soc. Hort. Sci., 78: 190-196. St. Joseph, Mi ch.: American Society for Horticultural Science. Ellis, C. R. 1971. Tree fruit harvester and picking devices. U.S. Patent No. 3566594. Erlbacher, E. A. 2000. Force control basics. Ind. Robot 27(1): 20-29. Fidelibus, M. W., F. S. Davi es, and C. A. Campbell. 2002a. Gibberellic acid application timing affects fruit quality of processing oranges. HortScience 37(2): 353-357. Fidelibus, M. W., A. A. Teix eira, and F. S. Davies. 2002b. Mechanical properties of orange peel and fruit treated pr e-harvest with gibberellic acid. Trans. ASAE 45(4): 1057-1062. Finkelstein, Z. 1996. System for harvesting crop items and crop harvesting tools used therewith. U.S. Patent No. 5544474.

PAGE 171

157 Fornes, I., F. Juste, C. Santamarina, and B. Bimbo. 1993. Potential damage to citrus fruits in mechanical harvesting. In Proc. 4th Intl. Symp. Fruit, Nut, Vegetable Prod. Eng., 2: 51-59. F. Juste, ed. Valencia, Spain: Instituto Valenciano de Investigaciones Agrarias. Gerber, C. E. 1987. Fruit harvesting machine. U.S. Patent No. 4674265. Glover, G. M. 1975. Fruit picking apparatus. U.S. Patent No. 3925973. Grand D’Esnon, A., G. Rabatel, R. Pellenc, A. Journeau, and M. J. Aldon. 1987. Magali: a self-propelled robot to pick apples. ASAE Paper No. 871037. St. Joseph, Mich.: ASAE. Guglielmino, E., M. Messina, R. C. Miche lini, and G. Amodeo. 1996. A robotic fixture for orange harvesting. In 27th Intl. Symp. Ind. Robots, 173-176. Courbevoie, France: International Federation of Robotics. Hannan, M. W., and T. F. Burks. 2004. Current developments in automated citrus harvesting. ASAE Paper No. 043087. St. Joseph, Mich.: ASAE. Hield, H. Z., L. N. Lewis, and R. L. Palmer. 1967. Fruit-stem detachment forces. Cal. Citrograph 52(10): 420-424. Hodges, A., E. Philippakos, D. Mulkey, T. Spreen, and R. Muraro. 2001. Economic impact of Florida's citrus industry, 1999-2000. Extension Digital Information Source (EDIS) FE307. Gainesville, Fla.: Univ ersity of Florida, Department of Food and Resource Economics. Ismail, M. A., and W. M. Miller. 1988. Evalua tion of mechanical damage to citrus. In Proc. Fresh Citrus Quality Short Course. Ft. Pierce, Fla.: University of Florida. Ito, N. 1990. Agricultural robots in Japan. In Proc. Intl. Workshop on Intelligent Robots and Syst., 1: 249-253. Piscataway, N.J.: IEEE. Juste, F., I. Fornés, F. Plá, and F. Sevila. 1992. An approach to robotic harvesting of citrus in Spain. In Proc. Intl. Soc. Citriculture: 7th Intl. Citrus Congress, 3: 10141018. Riverside, Cal.: Internati onal Society of Citriculture. Juste, F., C. Gracia, E. Molto, R. Ibanez, a nd S. Castillo. 1988. Fr uit bearing zones and physical properties of citrus for mechanical harvesting. In Citriculture: Proc. 6th Intl. Citrus Congress, 4: 1801-1809. Riverside, Cal. : International Society of Citriculture. Kassay, L. 1992. Hungarian robotic apple harvester. ASAE Paper No. 927042. St. Joseph, Mich.: ASAE.

PAGE 172

158 Kassay, L., and D. C. Slaughter. 1993. Electri cal, mechanical and electronic problems with the Hungarian apple ha rvester robot during the 1992 fall field tests. ASAE Paper No. 933088. St. Joseph, Mich.: ASAE. Kataoka, T., D. M. Bulanon, T. Hiroma, and Y. Ota. 1999. Performance of a robotic hand for apple harvesting. ASAE Paper No. 993003. St. Joseph, Mich.: ASAE. Kawamura, N., K. Namikawa, T. Fujiura, M. Ura, and Y. Ogawa. 1987. Study on agricultural robot (V II); hand of fruit harvesting robot. Res. Report on Agric. Machinery (17): 1-7. Kyoto, Japan: Kyot o University, Laboratory of Agricultural Machinery. Kedem, D., and M. Rubinstein. 1991. Fru it picking device. U.S. Patent No. 5005347. Kender, W. J., and U. Hartmond. 1999. Vari ability in detachment force and other properties of fruit within orange tree canopies. Fruit Varieties J. 53(2): 105-109. Kondo, N., T. Fujiura, M. Monta, and F. Se vila. 1998. Robots in bioproduction in open fields. In Robotics for Bioproduction Systems, 231-274. N. Kondo and K. C. Ting, eds. St. Joseph, Mich.: ASAE. Macidull, J. C. 1973. Fruit harvesting apparatus. U.S. Patent No. 3756001. Marth, D., and M. J. Marth. 2001. Florida Almanac 2002-2003. B. McGovern, ed. Gretna, La: Pelican Publishing Company Inc. Matlab. 2004. Matlab Help. Ver. 7.0.0.19920 (R14). Natick, Mass.: The MathWorks, Inc. McDonald, R. E., P. E. Shaw, P. D. Greany, T. T. Hatton, and C. W. Wilson. 1987. Effect of gibberellic acid on certain physical a nd chemical properties of grapefruit. Tropical Sci. 27(1): 17-22. Microsoft Excel. 2003. Excel Help. Ver. 11.5612.5606. Redmond, Wash.: Microsoft Corp. Middleton Jr., W. J. 1971. Article handli ng apparatus. U.S. Patent No. 3564826. Miller, W. M. 1986. Mechanical and physical properties for postharvest handling of Florida citrus. In Proc. Fla. State Hort. Soc., 99: 122-127. Lake Alfred, Fla.: Florida State Horticultural Society. Monta, M., N. Kondo, and K. C. Ting. 1998. End-effectors for tomato harvesting robot. Artifical Intelligence Rev. 12(1-3): 11-25. Muraro, R. P., T. H. Spreen, and M. Pozzan . 2003. Comparative costs of growing citrus in Florida and Sao Paulo (Brazil) fo r the 2000-01 season. Extension Digital Information Source (EDIS) FE364. Gainesv ille, Fla.: University of Florida, Department of Food and Resource Economics.

PAGE 173

159 Muscato, G., M. Prestifilippo, N. Abbate, and I. Rizzuto. 2005. A prototype of an orange picking robot: past history, the ne w robot and experimental results. Ind. Robot 32(2): 128-138. Pellenc, R., J. L. Montoya, A. G. D’Esnor , and M. Rombaut. 1990. Automated machine for detection and grasping of objects. U.S. Patent No. 4975016. Perry, D. 2002. Optimize your robot’s performa nce by selecting the right force/torque sensor system. Ind. Robot 29(5): 395-398. Pires, J. N., J. Ramming, S. Rauch, and R. Araújo. 2002. Force/torque sensing applied to industrial robotic deburring. Sensor Rev. 22(3): 232-241. Pool, T. A., and R. C. Harrell. 1991. An end-e ffector for robotic removal of citrus from the tree. Trans. ASAE 34(2): 373-378. Rabatel, G., A. Bourely, F. Sevila, and F. Ju ste. 1995. Robotic harv esting of citrus: stateof-art and development of the Fren ch Spanish Eureka Project. In Proc. Intl. Conf. Harvest and Postharvest Technologies for Fresh Fruits and Vegetables, 232-239. St. Joseph, Mich.: ASAE. Reed, J. N., S. J. Miles, J. Butler, M. Ba ldwin, and R. Noble. 2001. Automatic mushroom harvester development. J. Agric. Eng. Res. 78(1): 15-23. Ritenour, M. A. 2004. Orange. In The Commercial Storage of Fruits, Vegetables, and Florist and Nursery Stocks. USDA Handbook No. 66. Washington, D.C.: USDAARS. Available at: http://usna.usda.gov/hb66. Accessed 15 March 2006. Roistacher, L. J., L. J. Klotz, and I. L. Eaks. 1956. Detecting surface injuries to fruits. Cal. Citrograph 41(6): 239-242. Roka, F., and S. Longworth. 2001. Labor requ irements in Florida citrus. Extension Digital Information Source (EDIS) FE304. Gain esville, Fla.: University of Florida, Department of Food and Resource Economics. Rosenberg, B. 1974. Apparatus for picking fr uit growing on a tree. U.S. Patent No. 3854273. Rumsey, J. W. 1967. Response of the fruit-stem system to fruit removing actions. MS thesis. Tucson, Arizona: University of Arizona, Department of Agricultural Engineering. Rumsey, J. W., and Barnes K. K. 1970. Det achment characteristics of desert-grown oranges and grapefruit. Trans. ASAE 13(4): 528-530. Sanders, K. F. 2005. Orange harvesting systems review. Biosystems Eng. 90(2): 115-125.

PAGE 174

160 Sarig, Y. 1993. Robotics of fruit harv esting: a state-of-the-art review. J. Agric. Eng. Res. 54(4): 265-280. Sarig, Y., and S. Orlovsky. 1974. Viscoela stic properties of Shamouti oranges. J. Texture Studies 5(3): 339-349. SAS. 2004. SAS Help and Documentation. Ver. 9.1.3. Cary, N.C.: SAS Institute, Inc. Schatzki, T. F., R. P. Haff, R. Young, I. Can, L-C. Le, and N. Toyofuku. 1997. Defect detection in apples by means of x-ray imaging. Trans. ASAE 40(5): 1407-1415. Sciavicco, L., and B. Siciliano. 2000. Modeling and Control of Robot Manipulators. 2nd ed. London, Great Britain: Springer-Verlag. SolidWorks. 2003. SolidWorks Help. Ver. 2003 SP3.0. Concord, Mass.: SolidWorks Corporation. Southwest Florida Research and Educati on Center. 2006. Citrus Budwood Foundation Grove: Citrus Varieties. Imm okalee, Fla.: University of Florida, Institute of Food and Agricultural Sciences. Available at: http://swfrec.ifas.ufl.edu/citrus/bfg/index.htm. Accessed 5 September 2006. Suzuki, H., S. Okuyama, Y. Ueda, Y. Yukish ige, and M. Hayashi. 1988. Fruit harvesting apparatus. U.S. Patent No. 4718223. Tegin, J., and J. Wikander. 2005. Tactile se nsing in intelligent robotic manipulation–a review. Ind. Robot 32(1): 64-70. Terada, T. 1987. Fruit harvesting robot hand. U.S. Patent No. 4663925. Turrell, F. M., S. P. Monselise, and S. W. Au stin. 1964. Effect of climatic district and of location in tree on tenderness and other phys ical characteristics of citrus fruit. Botanical Gazette 125(3): 158-170. Tutle, E. G. 1983. Image controlled robo tics in agricultural environments. In Proc. 1st Intl. Conf. on Robotics and Intelligent Machines in Agric., 84-95. St. Joseph, Mich.: ASAE. Tutle, E. G. 1985. Robotic fruit harvester. U.S. Patent No. 4532757. Vartih, J., G. M. Hyde, A. L. Baritelle, J. K. Fellman, and T. Sattabongkot. 2001. Thermal image bruise detection. ASAE Paper No. 016031. St. Joseph, Mich.: ASAE. Yardley, W. 2004. “New harvesting machines shaking up citrus industry.” The Miami Herald . 6 Jan. 2004: A1.

PAGE 175

161 Yoshida, J., S. Okuyama, and H. Suzuki. 1985. Fruit harvesting apparatus with television camera and monitor. U.S. Patent No. 4519193. Zoellick, R. B. 2002. Trading in freedom: the new endeavor of the Americas. Economic Perspectives 7(3): 6-12.

PAGE 176

162 BIOGRAPHICAL SKETCH Samuel Flood was born on May 20, 1980 to an agricultural engineering professor, Dr. Clifford Flood; and a librarian, Susan Fl ood, in Opelika, Alabama. He graduated from Auburn High School in Auburn, Alabama in 1998 and then enrolled at the University of Florida. He received his B achelor of Science degree in agricultural and biological engineering in May 2002. After a couple months of em ployment at the USDA National Soil Dynamics Lab in Auburn, Alabama he returned to the University of Florida to pursue a graduate degree. He received hi s Master of Engineering degree in mechanical engineering in December 2004. Finally, after 6 years of dating, he was married to a veterinarian, Dr. Tania Wadhwa, in June 2005.