Group Title: Department of Computer and Information Science and Engineering Technical Reports
Title: Improved decision making through simulation based planning
Full Citation
Permanent Link:
 Material Information
Title: Improved decision making through simulation based planning
Series Title: Department of Computer and Information Science and Engineering Technical Reports
Physical Description: Book
Language: English
Creator: Fishwick, Paul A.
Kim, Gyooseok
Lee, Jin Joo
Affiliation: University of Florida
University of Florida
University of Florida
Publisher: Department of Computer and Information Science and Engineering, University of Florida
Place of Publication: Gainesville, Fla.
Copyright Date: 1996
 Record Information
Bibliographic ID: UF00095369
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.


This item has the following downloads:

1996212 ( PDF )

Full Text

Improved Decision Making through Simulation Based Planning

Paul A. Fli-li- i, 1:, Gyooseok Kim, Jin Joo Lee
Computer and Information Science and Engineering Department
University of Florida

Real-time military planning and decision making involves several different mod-
eling techniques, including rule-based, operator-based and dynamic-model based ap-
proaches. While rule-based approaches are generally fast and are more appropriate
for simple scenarios, simulation methods and dynamic models, indigenous to the sim-
ulation literature, are necessary to plan within environments involving large-scale un-
certainty, multiple interacting elements and complex dynamics. Planning techniques
must inter-operate to yield the best decisions, and we have found that simulation based
planning serves as an architecture for detailed model levels for both real-time and off-
line decision making. We introduce Simulation Based Planning as a methodology for
addressing the complexity involved in Air Force missions while employing an example
of air interdiction.

1 Introduction

Decision making and planning are critical operations for all military missions. .i', ever,
planning occurs over several different time scales depending on the amount of time that one
has to plan prior to committing to a particular plan. Planning is a hierarchical enterprise
since many techniques can be used to determine near-optimal plans. For example, if one has
information on costs between events during a mission, and the goal is to minimize cost, then
a mathematical programming approach, based on lowest cost path determination, may yield
satisfactory results. An even higher level type of planning is possible by using heuristics in
the form of operators and rules [Fil: -. 72]. Rules can use certainty factors or fuzzy sets.
Our long-term goal is to explore this hierarchy of planning approaches, and our first step
toward this goal is to provide high level planners with a technique we call Simulation Based
Planning (SBP). '. il i ,r'y missions involve many interacting elements including concurrently
active adversarial tasks and uncertain information regarding ground-based anti-aircraft ca-
pability. As the complexity of a mission and knowledge-base increases, it is valuable to use
computer simulation and more detailed dynamic models to obtain an answer to the broad
question "Which is the best approach to take given our mission and all currently available
knowledge1" A good method for answering this question is to use simulation since the
simulation technique is generally useful for obtaining answers to "'-li i if" scenarios. T111
quality of the answer depends on how much time is available prior to committing to the plan.
If more time is available, more simulation experiments can be executed in real time. If time
is of the essence, higher level models will need to be simulated. In any event, our goal is to
plan using the most detailed dynamic models available rather than to limit all planning to
the use of a singular planning technique such as a decision tree.
SWhere the knowledge about the terrain, ii. I i unit motions and postulated enemy plans incorporate a
great deal of uncertainty and can change over time.

Our contribution in the area of planning is to develop a method that allows simulation to
be used in real-time, where the simulation is embedded within the decision making system.
Consider a simulation system that manages force engagements up to the battalion level.
CASTFOREM [CASTFOREM 96] provides one such modeling capability [Wargames 96].
CASTFOREM is driven by decision trees and rules to guide what actions occur at any given
time. For interactive graphical output on the state of units, JANUS [Janus 96] can be used.
Our goal to to allow a program such as CASTFOREM the ability to make decisions based,
not on rules or decision tables, but on multiple simulations that are run within the overall
simulation. Typically, simulation has been used widely in the military for offline decision
analysis and "'li~.I if" weapons effectiveness assessment. Our proposal -i l._.- -i-. that we
enable simulation to support I .-ii_-, of ,. ii i1" (COA) analysis, and embed it directly within
the force simulation. Thi approach yields a two-level simulation procedure: simulations for
COA analysis guiding the decisions that drive the simulation of units, platoons, companies
and battalions. This -iiiiiiii, i, .i within a -7ilii lt l .1" approach is novel, but it can be time
consuming. For that reason, our work stresses the use of multi-level models so that different
;.--ii_ .-,iif..l levels can be executed so that the planning can be performed with real-time
Planning, regardless of the specific domain, involves three components: 1) model type,
2) plan set, 3) plan evaluation. Thi first step in planning is to determine the modeling
language (or type) to use. For rule-based approaches, this language can be "rules" or "pred-
icate logic." For other approaches, there are many alternatives: equation sets, finite state
automata, Petri nets, functional block models, queuing models. Often the model type is
visual in structure [Fi-hli," 1 95]. Thi next step is to create a set of candidate plans. For
rules, this set is often created through backward chaining. For more detailed model types,
the set is created by creating an experimental design and performing simulation (i.e., a type
of forward chaining). Plan evaluation is the key step where elements in the plan set are
simulated to determine the best planss. For our purposes, we view all modeling approaches
as being definable hierarchically, so that model types can include both rules and detailed
queuing models, for instance, defined at different abstraction levels. T111 -I hierarchical model
types are termed multimodels [Fi-lh, i, 1 95].
We will first discuss the application of simulation-based planning in Section 2: air inter-
diction. Ti111, we define the method of simulation-based planning, and finally we illustrate
our prototype simulation application which serves as an aid to plan interdiction missions.
In Section 3, we focus on route planning using a low-level strike mission on a munitions
factory. Information on the implementation is included in Section 4, followed by conclusions
in Section 5.

2 Air Interdiction

A typical use of the application of force is air interdiction, where the purpose is to de-
stroy, delay, or disrupt existing enemy surface forces while they are far from friendly surface
forces [Drew 92]. Th, interdiction mission includes attacks against supplies and lines of
communication. T111 objective of the interdiction mission is to reduce the enemy threat by
diminishing enemy combat effectiveness or by preventing a buildup of combat capabilities.

To achieve this objective, careful and comprehensive planning is required to isolate an area
and to stop all support from reaching the theatre of conflict. One must systematically attack
the significant elements of the enemy's logistical structures (transportation lines and cen-
ters, supply depots and storage facilities, repair and modification centers, staging areas, and
industrial installations) and maintain a high degree of destruction until the desired effect is
Tlii 1, are two levels of the interdiction plan : the interdiction plan at the theatre level
and at the tactical air force level [AFM 1-7 54]. A theatre interdiction plan establishes the
general scheme of employment, and enumerates available forces by type and number. T111
plan also outlines logistical support, delineates force responsibility, establishes the general
system of targets, prescribes the priority of target systems, and describes the anticipated
results. A theatre-level plan is developed through tactical air force planning. This planning
involves the day-to-day conduct of operations for the implementation of the assigned portion
of the broad theatre interdiction task. It covers specific and detailed actions of the forces
to be employed. A large part of the mission is dependent on which particular route or air
corridor is used.
Tli, task of the attack aircraft is to strike the target swiftly and accurately with whatever
munitions are carried, and then to return safely to base. To carry out this task, we must pen-
etrate the enemy defense. i, -'. difficulties arise here because methods of penetrating enemy
defenses can vary according to the strength and sophistication of the hostile detection, re-
porting, command and control network, and how much intelligence is available on the nature
of the defense capability [Spick 87]. :. i--i .1 planning will also involve maintaining a balance
between fuel and munitions resources to determine the load to be carried. Considering these
constraints, selecting the best routes can be a complex undertaking.

2.1 Simulation Based Planning (SBP)

SBP refers to the use of computer simulation [Law 91, Fi-rli i 1: 95] to aid in the decision
making process. In much the same way that adversarial trees are employed for determining
the best course of action in board games, SBP uses the same basic iterative approach where
a model of an action is executed to determine the efficiency of an input or control decision
within the given situation. However, board game trees implement a static position evaluation
function whereas, in SBP, a model (serving as a dynamic evaluation structure) is executed to
determine the effects of making a i ,ve." With the ability to simulate models at different
abstraction levels, SBP executes detailed models of a physical phenomenon when there is
sufficient planning time, or when fast computation and parallel methods are instrumented.
T111 military has been using simulation-based planning for many decades in the form of
constructive model simulation. A constructive model is one based on equations of attrition
and, possibly, square or hexagon-tiled maps using discrete jumps in both space and time. To
decide whether to accept a course of action, one can use a constructive model (a '.,ii )
to evaluate the alternatives. Related work by Czigler et al. [Czigler 94] demonstrates the
usefulness of simulation as a decision tool. Our extension in SBP is one where we permit
many levels of abstraction for a model, not just the ;-i- regate abstraction level characterized
by Lanchester equations and combat result tables. T111 idea is to allow the planner the
flexibility to balance the need between the required level of detail and the amount of time

given to make a decision.
Although the planning system shown in Figure 1 is divided into 5 functional blocks, we
will describe the overall framework in terms of three components: the experimental design
component, the output analysis component and the simulation component. Experimental
design is a method of choosing which configurations (parameter values) to simulate so that
the desired information can be acquired with the minimal amount of simulation [Law 91].
Since we treat uncertainty in the planning domain as random variates based on probability
distributions, repeated simulations (i.e., replications) using sampled data are necessary in
order to perform the proper analysis, which includes confidence intervals about the mean
for each result. We apply both heuristic and standard experimental methods to reduce the
overall simulation time in two aspects: 1) in reducing the number of replications, and 2) in
reducing the overall computation time spent on a single simulation of a plan on a particular
Following the experimental design, we simulate models of individual objects. This com-
ponent is called Trial (block 2) as shown in Figure 1. Figure 2 displays the lower level model
of the Trial block where each entity of the planning domain is modeled individually using the
appropriate model type. We assume that there are seven types of physical objects: BlueAC,
RedAC, Radar, SAM, Wx, Tgt, Zones and one abstract object called Eval. Tli, class BlueAC
stands for Blue Aircraft and RedA C for Red Aircraft. Radar represents a ground radar site.
SAM represents a ground Surface-to-Air '.i --i! site. Wx represents the weather. Tgt repre-
sents the target that needs to be destroyed. In the current prototype, the target is the red
force munitions factory. Zones represent the area of defense zone. For the zones, we assume
that there are a set of radars strategically located inside the zone such that when an enemy
plane BlueAC flies inside the zone, it is detected.
During a typical simulation loop, every object updates its local state and perform actions,
which in turn may affect other objects in the following time slice. Tli, last object to be called
within a simulation loop is Eval. Eval is responsible for three functions:

1. ' 1, 1 .li;,i a consistent il ii iii -i.i of the world that is an ;i-. related state of all
the current states of each object.

2. Deciding the outcome of inter-object events such as an engagement event between a
BlueA C and a RedA C. By allowing either of the objects involved decide the outcome of
an event-which would normally be beyond their control- we would violate symmetry
and self containment among objects.

3. Evaluating each situation (from the planner's point of view) for every time slice and
maintaining a score which represents the goodness of the plan.

We now perform output analysis using the set of output data produced from the repli-
cations using the following blocks: Replicator (block 1), Evaluator (block 3) and Ana-
lyzer (block 4). Output analysis is concerned with obtaining the appropriate interpretations
of the output data. Tli, Replicator controls the random number streams for each replication.
Different random number streams are used for each run, so that the results are independent
across runs. We also allow for common random numbers (CRN) to provide a controlled
environment for comparison among alternatives. This is to eliminate any i. i_. 'inii 11,11

dillt iiI -" that can exist between different simulations. CRN is a standard variance reduc-
tion technique in simulation and we use it across different alternative route plans within the
same replication so that we may expedite convergence to the true placement of the mean.
In its simplest form, the Evaluator serves as the accumulator of any relevant simulation
data that is produced from the Trial Block. If the objective function within the Trial block
produces a set of scores for each alternative, a straightforward evaluation approach is to total
the scores produced from the replications for each alternatives. Using the accumulated data
produced by the Evaluator block, the Analyzer block calculates the mean, variance and the
confidence interval for each alternative. Tli, mean of the replication results serves as the
basic i.i .1" point for the response surface representing the goodness of a plan. Variance can
be a measure of predictability or stability when the variance is small. Confidence intervals
are useful because given a sample output distribution and a confidence level x, the interval
states that, within confidence, the true mean lies within the stated interval [Lee 96].

2.2 An Air Interdiction Scenario
As one of the applications of SBP, we have chosen a typical air interdiction scenario, and
developed its Simulation Based Planner (C++) and graphical user interface (Tk/Tcl) within
our .ill i in dealing Object-Oriented Simulation Environment ('. OOSE) initiative. To illus-
trate the usefulness of the SBP approach, we consider the air interdiction scenario depicted
in Figure 3. Figure 3 defines a scenario with dynamically moving objects. Tli, mission of
the blue force aircraft is to destroy a red force munitions factory. T11i, are three Radars
(Ri, R2, R3) and two Surface-to-Air .I--i! (SA-.i) sites (Sl, S2), each with different ef-
fective detection ranges. Two red force fighters (Al, A2) are located in air defense Zone2
and Zone3 respectively, while one red force fighter (A3) is located outside of the air defense
zones. At first glance, the problem of guiding the blue force around the radar, SA i. and air
defense zone coverage, and toward the factory seems like a simple problem in computational
geometry. Tli, geometry approach is used frequently for route planning problems. A typical
rule might be formed as follows "To locate a path, avoid radar and SA I. fields, and avoid
fighting against enemy fighters." Tli, problem with this simple approach to route planning
is that the reasoning becomes difficult when uncertainty and dynamics are present. This
complexity manifests itself as an increasingly large rule base which often proves difficult to
create, maintain and verify for consistency.
To illustrate the kind of uncertainty and dynamics which are involved, consider the
following available information at some point during the mission.

Uncertain location and range : Radar R1 and R2 have been identified as permanent
fixtures, but a land based scout report -il.--, -. that R3 may have mobility. '.i, ,re-
over, the ranges (track, missile, arm range) of SA '. site Sl is well known, but S2 has
been reported to have a better guidance system including swift mobility, improving its
surveillance capability.

Uncertain enemy mission : red force fighter Al and A2 are known to be on a Combat
Air Patrol (CAP) mission, since they are always detected around Zone2 and Zone3;
however, A3's mission type is unknown.

In these examples, the behavior of each object is simplified as much as possible, since
our purpose is to demonstrate how to handle uncertainty in SBP, and not to focus solely
on the complex behaviors of objects. However, we have a plan to include the sophisticated
behaviors of each object incrementally. This can be considered as an advantage of the SBP
approach: the ability to increase the level of detail of the simulation object model as desired.

3 Route Planning Examples and Results

Figure 4 shows two possible routes (Routel, Route2) under the environment defined in Figure
3. Tli, goal of blue force aircraft is to destroy the red force munitions factory while satisfying
3 constraints: time or fuel level, safety, and destruction of the target. Given the possible
routes, the role of SBP is to choose the best route minimizing time and fuel consumption,
and maximizing safety and target destruction. In Figure 4, Routel is more attractive than
Route if we value mission time above all others, but seems less safe since it is vulnerable to
an attack by red fighter Al. Route might be considered more safe and achieve higher target
destruction than Routel by avoiding the attack from fighter Al and SA site Sl. However,
it will be detected by radar R2, increasing the probability of losing blue force aircraft or
damage to blue force aircraft. 'i never, there is a big chance of being detected by radar R3
even though its location is uncertain. Tli, table at the lower left of Figure 4 shows the result
of the SBP. We display the mean score and the confidence interval half width of each mean
at a 90% confidence level. As can be expected, Route2 is more successful since it avoids
direct attacks from the highly destructive enemy fighter and the SA i. site (mean score of
Route2: 69, mean score of Routel: -54).
If we delete Routel and consider another route based on the result of the previous situa-
tion, we may have two routes we want to analyze. Figure 5 illustrates these two candidates.
RouteS was chosen to avoid direct attack from Al, but for a short time period it will be
detected by R1. RouteS also takes the blue force into the track range of Sl, but not into its
arm or missile range. Being detected in the track range of Sl does not seem very dangerous
since only tracking functions may be performed by Sl. We can expect its success to depend
largely on the result of the samplings for uncertainty factors: specifically, the location and
guidance capability of SA '. S2 and the mission type of A3. If the powerful guided system
of SA '.1 is sampled close to this route, or A3 has a intercept capability, then the chance
of success will be very small. Otherwise, the chance of mission success will be very good.
Tli, -, nondeterministic and stochastic characteristics can be resolved by multiple simulation
with varying values for the uncertainty factors. Tli, confidence interval of the mean score
of RouteS is wide in comparison to that of Route2 due to the reason previously discussed;
however, the overall mean score is better than that of Route2 because of the small chance of
being detected by S2 or intercepted by A3.
We can now delete Route2 and insert a route, Route4, which is carefully chosen to mini-
mize the amount of time that a blue force aircraft will be within the detection ranges of R2
and R3 as in Figure 6. Tli, result of the SBP shows almost the same mean score for Route3
and Route4 (Route3 : 110.36, Route4 : 103.08) with Route3 being slightly better2. But
2The goal is to maximize the mean score for determining the better plan.

we can select Route4 as the best overall route based on its more narrow confidence interval
(Route4 : 1.3, RouteS : 6.0).

4 Implementation

In this section, we briefly introduce an example of the multimodel which we developed for the
Air Force Route Simulation and two analysis methods in the multiple simulations for dealing
with uncertainties, which arise from Simulation Based Planning. Additional implementation
issues and their potential solutions can be found in [Lee 96].
Thi purpose of the multimodel is to create a heterogeneous collection of connected sub-
models so that one can simulate different parts of the system at different abstraction levels.
TIl, choice of a dynamic selection of abstraction level provides flexibility to the simulation
based planning activity; real-time constraints can be met by tuning the multimodel. To
implement the multimodel, the generic Route Simulation '. idel in Figure 1 was instantiated
to the Air Force 'i.--i, i Route Simulator, and each object resides inside the Trial block as in
Figure 2. Among the seven types of physical objects, we chose one object, BlueA C, to explain
how we could capture the object's multimodel behavior. Ti, model presented here is not
complete since it has not been validated by a Subject '.i,11 r Expert (-I'.iiL). However, the
model represents the kind of model that one could obtain from the -:s. i through knowledge
acquisition methods. Since building sophisticated and realistic models is not the issue in our
current research, simple yet sensible models were built to prove our SBP approach.
ThI toplevel model of the BlueAC object is shown in Figure 7. It is modeled as an
FSA with three phases: Approach Target, Return to Base and End Mission. Figure 8 shows
the refinement FSA for Approach Target phase in Figure 7. Going another level down from
the Traverse Route phase, Figure 9 illustrates the functional block model for updating the
location while traversing the route. Figure 10 illustrates the refinement of RedA C Alert Mode
in Figure 8.
Assuming that a set of alternate routes and environment data are given through the
GUI, dynamic models are simulated and evaluated for each route. TIl, simulation process is
replicated and its output results are accumulated and then analyzed by the Analyzer (ref.
Figure 1).
For the object, we categorized the uncertainty into several types.

uncertainty of existence: the object may or may not exist.

uncertainty of location: an area of uncertainty of the object's location is available but
the exact location of the object is uncertain.

uncertainty of range: the exact detection range or firing range is not known.

uncertainty of mission: the exact mission type of an object is unknown.

uncertainty of fire power: the destruction capability of the object is uncertain.

T11 -, nondeterministic and stochastic characteristics were resolved by multiple simula-
tions using different samplings of the uncertainty factors. Tih planning problem becomes

one in optimization for an objective function representing the cost of traversing a route.
This cost is currently a function of elapsed time, remaining strength of the unit and the level
of success regarding achieving the goal. To reduce the total number of replications in the
simulation, we used two different output analysis methods: iterative and non-iterative. Tli
iterative method attempts to quantify significant pairwise differences among the alternatives'
means within a given confidence interval. T1i, method is referred to as "it(e ,! ., because he
algorithm iterates performing for every iteration, a set number of replications and analyzing
data to see if there are any significant differences among each route. Whenever a route is
found that is significantly worse than all other routes, this route is then eliminated. T11,
iteration continues until only two routes remain and a difference exists between the two of
T11, Non-iterative method is a method that avoids making an unnecessary number of
replications to resolve what may be an unimportant difference. When two alternatives are
close, we may not care if we erroneously choose one system (the one that may be slightly
worse) over the other (the one that is slightly better). Tlil-, given a correct selection prob-
ability P and the indifference amount D, the method calculates how many more replications
are necessary to make a selection with the probability of at least P, the expected score of
the selected alternative will be no smaller than by a margin of D. In our experiment, we
have chosen P = 0.95 and D = 13. A smaller D will produce more accurate results, but with
many more replications.
Recently, we have begun construction of a system, called IOOSE, to enable users to in-
teractively specify multimodels through a modeling window. Output is viewed via a scenario
window, similar to those shown in Figures 3-6. '-.OOSE (-.lul i iideling Object-Oriented
Simulation Environment) represents an implementation for a simulation system that is under
construction, and based on an extension to object oriented design (http: //www. cis .uf edu/
~fishwick/tr/tr96-026.html). '. OOSE is the next generation of SimPack (http: //www., which was initiated in 1990 for provid-
ing a general purpose toolkit of C and C++ libraries for discrete-event and continuous

5 Conclusions

We have discussed the method of simulation-based planning within the confines of an air
interdiction example. Our view is not that SBP replaces other forms of planning, but that
this new approach can be used in conjunction with existing, higher level planning approaches.
This way, given a set of alternatives to consider, SBP is able to extend the planning horizon in
three aspects: probabilistic uncertainty is handled through detailed and replicated simulation
of models rather than solving them analytically using probability theory; it extends the
level of reasoning to a finer level of granularity, producing plans that are closer to the
level of execution and discovering subtleties that may be missed by a higher level planner;
and finally, it breaks down the complexity of multiagent adversarial planning by employing
object-oriented multimodel simulation.
Once the simulation results have been produced, the data can be analyzed and interpreted
in several ways to choose the I" -," plan. For instance, we can choose the plan which has

not only a good mean score but also the minimum confidence interval width to ensure that
it is the safest plan possible. We may also decide to choose a plan that has the most
number of highest scores even though the confidence interval width may be large in order
to select a plan that has the best potential in spite of risks involved. We can even decide to
choose a plan at random (given that the scores are above some threshold) which will produce
nondeterministic planning. This is particularly useful for mission planning-opposing forces
should not be able to predict one's plan. In addition, similar to how simulation is used for
visualization, simulation can be easily used to perform visual playback of how a plan was
simulated to explain the planner's decision. This can be very useful for the military since
much of the military training is done through after action review.
Prior to the advent of fast low-cost personal computers, few researchers would consider
simulation of a fairly extensive experimental design to be a possible candidate for real-time
mission planning. However, as the speed of low-cost computers increases, the simulation-
based planning technique presents itself in a more attractive light. Our longer range goal
is to explicitly link several plan model levels together so that, for instance, a rule-based
level can be identified from a lower-level simulation. One of the authors (Kim) is studying
effective consistency measures which will rectify differences in rules produced empirically
(through knowledge acquisition) and rules generated automatically from multiple low-level

6 Acknowledgments

We would like to acknowledge Karen Alguire, Project '.I.11,. i, at Rome Laboratory, and
sponsorship under United States Air Force grant F30602-95-1-0031.

7 Biographies

Paul A. Fishwick is an Associate Professor in the Department of Computer and Informa-
tion Science and Engineering at the University of Florida. He received the BS in '. i, ii ii i, i
from the Pennsylvania State University, i1 in Applied Science from the College of William
and :. I iy, and PhD in Computer and Information Science from the University of Penn-
sylvania in 1986. He also has six years of industrial/government production and research
experience working at I'' "1,ort I'-' Shipbuilding and Dry Dock Co. (doing CAD/CA i.
parts definition research) and at NASA Langley Research Center (studying engineering data
base models for structural engineering). His research interests are in computer simulation
modeling and analysis methods for complex systems. He is a senior member of the IEEE and
the Society for Computer Simulation. He is also a member of the IEEE Society for Systems,
. i,~I and Cybernetics, AC' '. and AAAI. Dr. Fi-H'-i, 1: founded the comp.simulation Inter-
net news group (Simulation Digest) in 1987, which now serves over 15,000 subscribers. He
was chairman of the IEEE Computer Society technical committee on simulation (TCSIM)
for two years (1988-1990) and he is on the editorial boards of several journals including the
ACM Transactions on Modeling and Computer Simulation, IEEE Transactions on Systems,
Man and Cybernetics, T7, Transactions of the Society for Computer Simulation, Interna-

tional Journal of Computer Simulation, and the Journal of Systems Engineering. Dr. FIi-l-
wick's WWW home page is and his E-mail address
Gyooseok Kim received the B.S. degree in Electronics from Korean Air Force Academy
in 1984 and the '.i R degree in Computer Science from Korean i:'".,i i, ii! Defense College in
1989. He served in the Korean Air Force as a chief programmer of the Korean Air Defense
System. He is currently pursuing a doctoral degree in the Computer and Information Science
and Engineering department at the University of Florida. His research interests are focused
on knowledge acquisition and validation within qualitative and quantitative simulation. 'i.1
Kim's E-mail address is
Jin Joo Lee received the B.S. degree in Computer Science from Ewha Womans University,
Korea in 1988 and the '.1 'i degree in Computer Science from Brown University in 1991. After
receiving the :.I degree, she was a research engineer at Human Computers Inc., Korea until
1992. '11, received a PhD degree in the Computer and Information Science and Engineering
department at the University of Florida in 1996. Her research interests are in AI planning,
simulation and control. '.i- Lee's WWW home page is
and her E-mail address is


[AFM 1-7 54] Department of the Air Force. Air force Manual No. 1- 7: T/,. ,i! Air Forces
in Counter Air, Interdiction And Close Air Support Operations. Department of
the Air Force, 1954.

[CASTFOREM 96] CASTFOREM, Web Reference: http://hpOl.arc. iquest. cor/
mosaic/060 .html

[Czigler 94] ':. Czigler, S. Downes--. lirtin and D. Panagos. Fast Futures Contingency Sim-
ulation: A "What If' Tool for Exploring Alternative Plans, In Proceedings of the
1994 SCS Simulation :.ill!I Conference, San Diego, CA, 1994.

[Drew 92] D. '.1 Drew. Air Force Manual 1-1: Basic Aerospace Doctrine of the United States
Air Force, Volume II, Department of the Air Force, 1992.

[Fil, -. 72] R.E. Fil. -, P.E. Hart, and N.J. :'"i.!-i.1i Learning and Executing Generalized
Robot Plans, Artificial Intelligence,3,1972.

[Fi-lh, i, : 95] Paul A. Fi-lh, i, :. Simulation Model Design and Execution: Building Digital
Worlds, Prentice Hall, 1995.

[Janus 96] JANUS, Web Reference: http://hpOl. arc. iquest .com/war/janus. html

[Law 91] A. :.I Law and W. D. Kelton. Simulation Modeling and Analysis :.I Graw-Hill,

[Lee 93] J.J. Lee, W.D. Norris and P.A. Fi-lr -i, !. An Object-Oriented 'ii!, ii_,deling
Design for Integrating Simulation and Planning Tasks, In Journal of Systems
Engineering, 3, 220-235, 1993.

[Lee 94] J.J. Lee and P.A. Fi-lh, i, 1:. Real-Time Simulation-Based Planning for Computer
Generated Force Simulation, Simulation,299-315, 1994.

[Lee 95] J.J. Lee. and P. A. Fi-lli~" i, Simulation-Based Real-Time Decision Making for
Route Planning. In Proceedings of the 1995 Winter Simulation Conference, 1087-
1095, 1995

[Lee 96] J.J. Lee. A Simulation-Based Approach for Decision Making and Route Planning.
PhD T1 -i- University of Florida, 1996.

[Russell 95] S.J. Russell and P. Norvig. Artificial Intelligence A Modern Approach, Prentice-
Hall, 1995.

[Schoppers 87] '.1 Schoppers. Universal Plans for Reactive Robots in Unpredictable Do-
mains, In Int. Joint Conference on Artificial Intelligence,1987.

[Spick 87] '.1 Spick. An Illustrated Guide to Modern Attack Aircraft. An Arco :.il11 ,irjy Book,
Prentice Hall Press, York, 1987

[Salisbury 93] -. Salisbury and H. Tallis. Automated Planning and Replanning for Battle-
field Simulation, In Proceedings of the Third Conference on Computer Generated
Forces and Behavioral Representation,1993,243-254,Orlando,FL.

[Wargames 96] Wargame Catalog, Web Reference:

Executive 0


Replicator 1

Random Seed

Trial 2

Initialize For Route Ri, i 1..n}
a_ Route Simulation Module
Random #
Generator Eval n
________ |Evaluation Functio---

Evaluator 3
- stores sim.
- stores scores

set of scores
set of times

output data
make stat calls for data analysis

Analyzer 4
Mean ... etc.

Figure 1: Generic Top Level Architecture of a Route Planner

Figure 2: General Simulator 'i,!dule

[ ,:hilt g FoQr i.g

a, r h:r



, Gn ['.Io.I-i I "
M U f rIum I|Blue Force AircraftI
SLocation Information A TypelCoverage Information Coverage ONIOFF
Figure 3: A Typical Air earInterdiction Scenaranvaio

Figure 3: A Typical Air Interdiction Scenario

.fJ 3]


Quit Load File Save As

r rii : hI

Eiti:ng Forces

t n


Gl o. I-
*l ~riM rCiER
m15 .nlLu

Current File : Scenario X:111 Y :Z09

!F SwZaatEta L



Run Simulation I

1 ll

Ileanl Score Clif. inlerva
69.167723 3.5-1093B

Blue Force Aircraft

Location Information TypeCoverage Information Coverage ONIOFF
ear Canvas

I Help Window I

Figure 4: Two Possible Routes in the Figure 3


Quit Load File Save As urent File : Scenario3 X 111 Y :20
F| Enro--nrr r

S) gBHMunitions Factory E

E:hr g Foli,- / "

IRERTE noute3 Route2
MeanScorp Conf. tera

iBlue Force Aircraft

er Canvas jr .I.. J j
locatonn,-ormatio. Tylel*ovierge Information { Coverage ONO IF .|nImIlation,

SHelp Window

Figure 5: Deleting Routel in the Figure 4, and Inserting Route3


Figure 6: Deleting Route2 in the Figure 5, and Inserting Route4


new state

1 = true or 2 = true

1. mission failure
2. mission success

Figure 7: Blue Aircraft(BlueAC) object model

current state

6 = true -> mission failure

Figure 8: Approach Target for BlueAC

1. target in range
2. missile detected
3. red destroyed or avoided
4. red detected
5. target destroyed
6. blue destroyed or disabled
7. missile destroyed or avoided
8. time left of mission

BlueAC::Approach Target > Traverse Route


- speed
- current location
- delta-t
- angle
- wind factor
(force, angle)
- fuel level




new location

Figure 9: Traverse Route Function for BlueAC::Approach Target


S1. enough fire power
= true 2 = fa e 1 = false 2. enemy lost
& 3 false 4 = false or 5 = fals, 3. enemy destroyed
& 5 false 4. enough fuel
l=se 5. enough time
or 6. blue destroyed or disabled
a = enum
a. mission behavior
2 =tr = {passive, aggressive}

Evade 6 = true

Figure 10: Red Aircrat(RedA) Alert 'i-de for BlueAC object
Figure 10: Rted Aircraft (RedAC) Alert :. for BlueA C object

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs