MULTIPLE UAV SIMULATION WITH MULTIRESOLUTION MULTISTAGE MODELS 
AND DECISION SUPPORT 
 
 
 
Except where reference is made to the work of others, the work described in this thesis is my 
own or was done in collaboration with my advisory committee. This thesis does not include 
proprietary or classified information. 
 
 
 
_____________________________________ 
Bo Li 
 
 
 
Certificate of Approval: 
 
 
 
_____________________________                 _____________________________ 
Min-Te Sun                                     Alvin Lim, Chair 
Assistant Professor                               Associate Professor 
Computer Science and                            Computer Science and 
Software Engineering                             Software Engineering 
 
 
 
_____________________________                 _____________________________ 
Cheryl Seals                                    George T. Flowers 
Assistant Professor                               Dean 
Computer Science and                            Graduate School 
Software Engineering 
 
MULTIPLE UAV SIMULATION WITH MULTIRESOLUTION MULTISTAGE MODELS 
AND DECISION SUPPORT 
 
 
Bo Li 
 
 
 
A Thesis 
Submitted to 
the Graduate Faculty of  
Auburn University 
in Partial Fulfillment of 
the Requirements for the 
Degree of 
Master of Science 
 
 
Auburn, Alabama 
August 9, 2008
 iii
MULTIPLE UAV SIMULATION WITH MULTIRESOLUTION MULTISTAGE MODELS 
AND DECISION SUPPORT 
 
 
Bo Li 
Permission is granted to Auburn University to make copies of this thesis at its discretion, 
upon request of individuals or institutions and at their expense. The author reserves all 
publication rights. 
 
 
_____________________________ 
  Signature of Author 
 
 
 _____________________________ 
  Date of Graduation 
 
 iv
VITA 
     Bo Li, son of Zhengyi Li and Xiaoli Bao, was born on June 1, 1979 in Beijing, P. R. 
China. He graduated from Beijing No. 22 Middle School in 1995. He entered Jin Lin 
University, Changchun, P. R. China and graduated summa cum laude with a Bachelor of 
Science degree in Computer Science in July, 2002. From January 2003 to August 2003, he 
worked at Beijing CTNA Technology CO., Ltd as a software programmer. From August 2003 
to November 2005, he worked at Panasonic Corporation of China as a system administrator 
and data analysis specialist. After that, he entered Graduate School of Auburn University 
majoring in Computer Science in January, 2006. During his graduate study at Auburn 
University, he performed well on both his academic study and research work. From January 
2006 to July 2006, he was involved in part time work at Distance Learning Department of 
Auburn University as a programmer. From 2006 to 2007, he was a Research Assistant, and 
from 2006 to 2008 he was a Teaching Assistant in the Computer Science Department of 
Auburn University. 
 
 v
THESIS ABSTRACT 
MULTIPLE UAV SIMULATION WITH MULTIRESOLUTION MULTISTAGE MODELS 
AND DECISION SUPPORT 
 
 
Bo Li 
Master of Science, August 9, 2008 
(Bachelor of Science, Ji Lin University, 2002) 
 
109 Typed Pages 
Directed by Alvin Lim 
 
Multi UAV simulation is a dynamic multi-simulation of multiple Unmanned Aerial 
Vehicles (UAV) activities in the battlefield with multiresolution multistage models. Decision 
support is implemented under this simulation. It provides an intelligent agent supported by 
decision making subsystem based on the decision making model. It deals with the uncertainty 
of the battlefield and presents a multiple simulations with Multiresolution Multistage 
(MRMSM) model. Unmanned Aerial Vehicle is one of the effective devices in the future 
battlefield and has many advantages. Single UAV has been used in the battlefield. Multiple 
UAVs have more advantages than a single UAV to attack a group of targets. A UAV team can 
 vi
accomplish tasks that a single one can not execute alone. The requirement to coordinate 
multi-UAVs continues to surface in both defense and civil applications. Multiple UAVs form 
teams and need to cooperate well in the battlefield to attack efficiently and effectively. 
Decision support subsystem generates two strategies that could be selected for the UAV team 
action. Bayesian Network is used for making decisions to select a strategy for UAV team in a 
specific situation. Based on the selected strategy, UAV team could attack the targets in an 
effective manner. In this simulation, there are multiple simulations working on different 
platforms at multiple stages. They exchange information in a timely manner with each other 
and are dynamically launched, updated or reconfigured, and stopped. These actions are 
supported by the decision making model. There are five main parts in this simulation, Matlab 
Simulink, Filter, Tactical Federate, High resolution and Visualization. Each part works on its 
own platform. These parts dynamically work with each other and communicate with each 
other and exchange data timely. This paper focuses on the Simulink, Filter, Tactical federate 
and communication between parties. The results show that each part can work independently 
and cooperate well with each other. UAV team can effectively attack battalions which are 
target clusters. The two-dimension visualization for low resolution and three-dimension 
visualization for high resolution show the entire simulation scenario at different stages. 
 vii
ACKNOWLEGEMENTS 
 
I would like to express special gratitude to my advisor Dr. Alvin Lim, professor, 
Department of Computer Science and Software Engineering at Auburn University, for his 
instruction, guidance, encouragement and patience in completion of the research and thesis. 
In particular, his suggestions, criticisms and general exchange of ideas greatly contributed to 
this thesis. 
Thanks, also, due to his advisory committee members, Dr. Yilmaz, Dr. Seals and 
Dr.Sun and the professors and staff members in the Department of Computer Science and 
Software Engineering at Auburn University for their kindness and help through these years. 
Finally, a sincere thank to my friend Haihong Liang. She gave her greatest support and 
encouragement to help me finish all the research work. Also, I thank my parents, who provide 
the inspiration, encouragement and support to me during these years. 
 
 viii
Style manual or journal used: Guide to Preparation and Submission of Thesis and 
Dissertation 
Computer software used: Microsoft Word 2003, PowerPoint 2003 
 ix
TABLE OF CONTENTS 
LIST OF FIGURES..................................................................................................................................xi 
CHAPTER 1: INTRODUCTION AND BACKGROUND............................................................. 13 
1.1 General Introduction ...................................................................................................13 
1.2 Multiresolution Multistage Model (MRMSM)...........................................................14 
1.3 Multisimulation...........................................................................................................15 
1.4 Multisimulation with MRMS Model ..........................................................................16 
1.5 Decision Making.........................................................................................................16 
1.6 Situation Awareness ....................................................................................................18 
1.7 Integration Mulisimulation with Decision Making Support Subsystem.....................18 
1.8 Unmanned Aerial Vehicle ...........................................................................................18 
1.9 Matlab Simulink..........................................................................................................20 
1.10 Filter..........................................................................................................................21 
1.10.1 Introduction.....................................................................................................21 
1.10.2 Contract Net protocol......................................................................................22 
1.10.3 Cluster Algorithm............................................................................................22 
1.11 Tactical Federate .......................................................................................................23 
1.12 High Resolution ........................................................................................................25 
1.12.1 Introduction.....................................................................................................25 
1.12.2 VR-Link, HLA Mak RTI, Stealth....................................................................26 
1.13 Strategy for UAV Team.............................................................................................27 
CHAPTER 2: MOTIVATIONS ............................................................................................................ 33 
CHAPTOR 3: OBJECTIVES................................................................................................................ 35 
CHAPTER 4: CHALLENGING PROBLEMS ................................................................................. 37 
4.1 The Problem of Deep Uncertainty ..............................................................................37 
4.2 Multiple Simulations at Multiple Stages.....................................................................37 
4.3 UAV Team Work with Strategy Support and Targets Clustering................................38 
CHAPTER 5: RELATED WORKS ..................................................................................................... 40 
5.1 Distributed Task Allocation ........................................................................................40 
5.2 Path Planning ..............................................................................................................42 
5.3 Consistency Maintenance ...........................................................................................43 
5.4 Multi-agents System ...................................................................................................44 
 x
5.5 Multi-agents Market-based System ............................................................................46 
5.6 Genetic Algorithm.......................................................................................................47 
5.7 Evidential Map-building .............................................................................................48 
5.8 Team Work ..................................................................................................................49 
CHAPTER 6: GENERAL APPROACHES ....................................................................................... 52 
6.1 Introduction.................................................................................................................52 
6.2 Simulation Scenario....................................................................................................52 
6.3 Low Resolution...........................................................................................................54 
6.4 Filter............................................................................................................................56 
6.5 Tactical Federate .........................................................................................................56 
6.6 High Resolutions.........................................................................................................58 
CHAPTER 7: DETAILS OF DESIGN AND SIMULATION........................................................ 61 
7.1 Introduction.................................................................................................................61 
7.2 Low Resolution...........................................................................................................62 
7.2.1 Initialization Model...........................................................................................63 
7.2.2 Communication Model .....................................................................................63 
7.3 Filter............................................................................................................................65 
7.3.1 Introduction.......................................................................................................65 
7.3.2 Targets Clustering .............................................................................................65 
7.3.3 Team Formation ................................................................................................66 
7.3.4 Consensus Model..............................................................................................67 
7.3.5 Communication Functions of Filter..................................................................68 
7.4 Tactical Federate .........................................................................................................69 
7.4.1 Strategy Selection .............................................................................................69 
7.4.2 Communication Functions of Tactical Federate ...............................................74 
CHAPTER 8: SIMULATION RESULTS COMPARISON ............................................................ 95 
CHAPTER 9: SUMMARY AND FUTURE WORKS................................................................... 102 
REFERENCES....................................................................................................................................... 104
 
 
 xi
LIST OF FIGURES 
Figure 1.1 Unmanned Aerial Vehicles (UAV). Predator.................................................................. 28 
Figure 1.2 Multiresolution Model ........................................................................................................ 29 
Figure 1.3 Mulitstage Model ................................................................................................................. 29 
Figure 1.4 Simulation Updating and Branching for Multisimulation........................................... 30 
Figure 1.5 Matlab Simulink................................................................................................................... 31 
Figure 1.6 Formation Flight of UAV Team and Flock of Birds (Song et al., 2005) .................. 32 
Figure 6.1 Low Resolution Visualization ........................................................................................... 59 
Figure 6.2 High Resolution Visualization........................................................................................... 60 
Figure 7.1 Low Resolution at Initialization Stage............................................................................. 75 
Figure 7.2 ?Save Plot Data? Block in Simulink ................................................................................ 76 
Figure 7.3 ?Save Plot Data? inside Structure..................................................................................... 77 
Figure 7.4 Time Unit of the S-Function is set to 0.1 second .......................................................... 78 
Figure 7.5 Simulink and Filter Communication Sequence Chart.................................................. 79 
Figure 7.6 Communication between Simulink and Tactical Federate .......................................... 80 
Figure 7.7 A Simple Targets Clustering Process ............................................................................... 81 
Figure 7.8 Targets are Clustered as Battalions................................................................................... 82 
Figure 7.9 Team Formation Process Based on Contract Net .......................................................... 83 
Figure 7.10 Team Formation Process Results.................................................................................... 84 
Figure 7.11 Filter and Tactical Federate Communication Sequence Chart.................................. 85 
Figure 7.12 Communication between Filter and Tactical Federate............................................... 86 
Figure 7.13 MSBNx Implementation .................................................................................................. 87 
Figure 7.14 Sample of Critical Target ................................................................................................. 88 
Figure 7.15 Assessment Table of ?Number of Target? Node.......................................................... 89 
Figure 7.16 Assessment Table of DSC Node ..................................................................................... 90 
 xii
Figure 7.17 Calculation Results of Bayesian Network .................................................................... 91 
Figure 7.18 Simulation using Fringe Point strategy (1) - Fringe and Fringe Points of  
Targets Area....................................................................................................................................... 92 
Figure 7.19 Simulation using Fringe Point Strategy (2).................................................................. 92 
Figure 7.20 Simulation using Fringe Point Strategy (3).................................................................. 93 
Figure 7.21 Simulation using Region Strategy.................................................................................. 93 
Figure 7.22 Program Sequence Chart of Tactical Federate Communication  
Functions............................................................................................................................................ 94 
Figure 8.1 UAV Destroy Target by Time Lapse ................................................................................ 99 
Figure 8.2 Target Destroy Rate by Using Different Strategies..................................................... 100 
Figure 8.3 Efficiency of UAV Destroy Target Using the Strategies............................................ 101 
 
 
 
 13
CHAPTER 1: INTRODUCTION AND BACKGROUND
1.1 General Introduction 
Unmanned Aerial Vehicles (UAV) (see Figure 1.1) is widely used in today?s battle field. 
Simulating multiple UAV team to attack a group of targets effectively is a problem. The goal 
of this simulation is to build a decision making system based on exploratory multisimulation 
technology integrated to the subsystem based on Naturalistic Decision Making theory. We 
simulate situation awareness and exploratory multisimulation to evaluate the efficiency and 
effectiveness of multiple actions at different levels of resolution. 
UAVs search and sense the area, perceive the situation, form teams and attack battalions based on 
strategies. The ability to use cooperative autonomous vehicles to perform a wartime mission is an 
important application in future military operations (Brown 1998, Davis 1995, McHale 2003, French 2003) 
[1]. This simulation focuses on UAV team work. There are multiple simulations and models in this 
simulation. They work respectively and cooperatively. Some simulations work on different platforms. They 
communicate with each other in a timely manner to exchange information and keep consistence between 
each other. Under the decision making model, there are multiple simulations with multiple stages. There 
are 2D visualization for low resolution and 3D visualization for high resolution. These visualizations are 
consistent with each other. 
 14
1.2 Multiresolution Multistage Model (MRMSM) 
Resolution is the degree of detail and precision used in the representation of real world 
aspects in a model or simulation [2]. Multi-Resolution Modeling involves the following: [3] 
(1) Building a single model with alternative user modes involving different levels of 
resolution for the same phenomena; 
(2) Building an integrated family of two or more mutually consistent models of the same 
phenomena at different levels of resolution; or 
(3) Both. 
There are two resolutions in this simulation, low resolution and high resolution. Since 
low resolution is more abstract than high resolution, it is at a high level and high resolution is 
at a low level. Multiresolution model (see Figure 1.2) composes multiple single models, or 
same model at different levels of resolution. The Multiresolution Multistage Model integrates 
multiresolution models with multistage models (see Figure 1.3) to help deal with the problem 
of multiple tasks that are at different resolution levels. Simulation entities need to be capable 
of simultaneously operating at different levels and maintain consistency at each level. 
Federates need to switch between the submodels. Models of entities are needed to be 
dynamically updated across the resolutions at run-time. Low resolution should exchange 
results with high resolution. Models have different stages and can switch as the resolution 
level of changes. Decision making subsystem need to make its decision and inject its 
judgments to the multiple subsystems. Since the uncertainty of the battlefield scenario, 
strategy decision will be made every period of time and dynamically update the simulations 
 15
which may make change of the resolution level. Figure 1.2 is Multiresolution model. There 
are three models, Model 1 (M1), Model 2 and Model 3 representing same entity at different 
levels of resolution. Figure 1.3 is the Multistage model in which there are two subset models. 
Only one subset model is active at one stage unless there are multiple simulations. The 
controller decides when and how the subset model switches from one to another. If the 
sub-model in one stage constitute of model of the same entity at different levels of resolution, 
the model is a Multiresolution Multistage model (MRMSM) [4], i.e., the UAV entity is at 
both low and high resolution at the same stage such as attacking. 
 
1.3 Multisimulation 
Multisimulation (or multisim, for short) is simulation of several aspects of reality in a 
study. It includes simulation with multiresolution model, multiaspect model and multistage 
model. In multisimulation model, decision supports model updating or changing. The 
associated parameters and experimental conditions are used in the subsequent stages of 
simulation [4]. In this simulation, a new model or simulation can be invoked at the 
interruption point. This work is based on the assessment of the previous simulation to deal 
with the uncertainty of the scenario. In each stage, there are multiple simulations work on 
different levels and cooperate with each other. Models and different simulations can be 
invoked or stopped during each stage. Multisimulation updating process is showed as Figure 
1.4. 
 
 16
1.4 Multisimulation with MRMS Model 
Different resolutions work independently and cooperatively such as low resolution and 
high resolution. There are many Course-of-Actions (COA) for a resolution. In this simulation, 
multiple high resolutions are implemented to analyze and assess COAs. The model and the 
components of the models are dynamically updated based on demands and assessment of 
previous work during the simulation. The UAV team in high resolutions can perform based 
on different strategies generated by low resolution strategy decision model which is in 
Tactical Federate. The multiple resolutions or simulations run on different machines or 
simulation platforms in parallel and corporation work. MRMS Models need to dynamically 
and timely exchange information with each other to make consistency between low resolution 
and high resolution. It needs run-time enforcement policy for multiple resolutions consistency 
when the models update. It also needs to flexibly update resolution levels on demand and 
change the behavioral models of each entity at run-time and at different resolution levels [4]. 
 
1.5 Decision Making 
The battlefield scenario is uncertain and changes dynamically. Different models and 
resolution needs to be consistent with each other. To make the UAV attack efficiently and 
effectively, proper strategy is needed to direct the UAV in attacking targets. Strategy 
generation and selection, models changing, and resolutions updating are all based on decision 
making. 
 17
Decision science contributes both to the understanding of decision making and to 
developing methods and tools to assist that decision making (Davis et al., 2005). Decision 
making is important to the selection of a course of action among variations. Every decision 
making process produces a final choice. It can be an action or an opinion. It begins when we 
need to do something but do not know how to do it.  
Rational choice theory, also known as rational action theory, is a framework for 
understanding and often formally models social and economic behavior. Although models of 
rational choice are diverse, all models assume that individuals choose the best action 
according to stable preference functions and constraints facing them. When human makes 
decision, there are many uncertain and complex conditions, and the decision needs to be 
made within limited time and resource [4]. 
The empirical work of Gary Klein (1997) on expert behavior in high-pressure 
environments resulted in a new school of thought in decision-making. This simulation uses 
Naturalistic Decision Making theory. The naturalistic decision making (NDM) framework 
emerged as a means of studying how people actually make decisions and perform cognitively 
complex functions in demanding situations. The NDM paradigm argues that people assess 
situations by using prior experience and knowledge. This simulation uses Bayesian Network 
to support decision making. 
 
 18
 
1.6 Situation Awareness 
Situation Awareness is an important technique for decision making which deals with 
complexity, complicate, dynamical and uncertain situations. It should perceive the situation 
before making a decision. If there is no perception or the perception fails, the decision 
making may not right and may cause the failure of the action. In this simulation, a perception 
model is implemented in the Filter to help UAV accurately perceive the situation. 
 
1.7 Integration Mulisimulation with Decision Making Support Subsystem 
Symbiotic Simulation interacts with the physical system in a mutually beneficial way. 
The simulation system not only performs ?what-if? experiments that control the physical 
system, but also accepts data and responses to physical system. In this simulation, 
Multisimulation cooperates with the decision making supporting subsystem. Models updating 
and switching of stages, resolutions switching of levels and stages are all supported by the 
decision making subsystem. 
 
1.8 Unmanned Aerial Vehicle 
An Unmanned Aerial Vehicle (UAV) (See Figure 1.1) has no onboard pilot. It flies 
autonomously based on pre-programmed flight plans or more complex dynamic automation 
 19
systems. Typically they are controlled by a remote operator or autonomous control logic. 
UAV has several main functions remote sensing, interaction, and transport. Remote sensing 
includes electromagnetic spectrum sensors, biological sensors, and chemical sensors. This 
simulation only simulated basic electromagnetic spectrum sensing function for each UAV 
perceiving the battlefield situation. Interaction includes attacking, taking samples, delivering 
a payload, and repairing a given structure. This simulation provides the attacking function for 
UAV and new functions such as interaction between UAVs and the controller, team formation, 
and team action with strategies. This simulation does implement the UAV transport payload. 
The UAVs in this simulation work cooperatively, exchange data, and communicate with each 
other in a timely manner. They can form a team to attack a group of targets based on some 
strategies. Their actions are evaluated and strategies are updated based on the evaluation. 
An Unmanned Aerial Vehicle is widely used in the battlefield and it has many 
advantages. Single UAVs have been used in the battlefield for years. While in some 
conditions such as attacking a group of targets, more vehicles need to work together to attack 
in an efficient and effective manner. In this simulation, UAVs are generated at Matlab 
simulink. Their initial task is sense or to perceive the area looking for the target. A 2D 
visualization is generated by Simulink to show this scenario. The data sensed by the UAV is 
sent to the Filter which is simulated by Visual C++.Net to analyze and integrate to get the full 
picture of the battlefield. When a group of targets is found, a team of UAVs is formed to 
attack them. A strategy for the attacking team is generated by Tactical Federate simulated by 
Visual C++.Net. It selects a suitable strategy to make the team work efficiently and 
effectively. It launches the high resolution with the team information, strategy, and targets 
 20
information. High resolution simulates the UAV team attacking a group of targets and shows 
the scenario in a 3D visualization. 
 
1.9 Matlab Simulink 
In this simulation, Matlab Simulink is used to implement the low resolution and 
generate a 2D visualization. 
Matlab is a high-level language and interactive environment that enables the user to 
perform computationally intensive tasks faster than with traditional programming languages 
such as C, C++, and FORTRAN. Simulink is integrated with Matlab and provides multiple 
visualization functions to programming. Simulink is an environment for multidomain 
simulation and Model-Based Design for dynamic and embedded systems. It provides an 
interactive graphical environment and a customizable set of block libraries that the user can 
design, simulate, implement, and test a variety of time-varying systems, including 
communications, controls, signal processing, video processing, and image processing. 
Simulink provides an environment where user can make model of physical system and 
controller as a block diagram. The block diagram is created by using a mouse to connect 
blocks and a keyboard to edit block parameters (see Figure 1.5). Matlab can generate 
visualization window for simulation and it provides multiple functions to control the 
visualization process. 
In Simulink, when the UAVs are launched, they unfold the battlefield by sensing it 
from west to east. They fly on the initialize routes. Simulink exchanges UAV and target 
 21
information with Filter in a timely manner. Based on the information, Filter clusters the 
targets as a battalion. When a battalion is found, UAVs form a team. All team members will 
fly to the center point of the battalion. After attacking, they disperse and attack the rest of the 
targets. This scenario is synchronized shown in the 2D visualization generated by Matlab. 
 
1.10 Filter 
1.10.1 Introduction 
Filter module is simulated by MS VC++.Net. It filters the data from Matlab. It 
communicates with Matlab Simulink and exchange data with it. When the UAVs are 
launched, every UAV?s private information such as position and perception of the field are 
sent to Filter. Based on the information, Filter integrates data and analyzes data to unfold an 
entire picture of the battlefield. It analyzes the targets by clustering them based on the cluster 
algorithm. Attacking a battalion is hard and dangerous for a single UAV. In this simulation, if 
there is a battalion, a team formation process is invoked, and one or more UAV teams are 
formed. A group of multi-UAVs in close formation looks like a flock of birds (see Figure 1.6) 
and relies on many technologies such as advanced control methodologies [16]. The team 
formation process is based on Contract Net protocol. It allows contracting as well as 
subcontracting. Since each UAV in the team could have different perceptions to other team 
members, a consensus of perception within the team is confirmed after the team formation. 
Filter sends the team and battalion information to Tactical Federate for further use. The Filter 
 22
also serves as a communication channel between Simulink and Tactical Federate. It passes 
UAVs and targets information from Tactical Federate to Simulink making their models 
consistent. 
Protocol used for communication between Simulink and Filter is UDP. Because Matlab 
integrates C within it and Filter is implemented by C, they can use UDP to communicate 
implemented by C. Filter and Tactical Federate communicate also based on UDP. 
 
1.10.2 Contract Net protocol 
Contract Net protocol is proposed by Smith (1980) to solve the problems in the 
distributed, multi-agents system such as tasks allocation. It is a fully automated negotiation 
protocol and work like the electronic marketplace for buying and selling goods. There are two 
types of agents, Initiator and Participant. At any time, any one agent can be an Initiator, 
Participant or both. In this simulation, the team leader serves as the Initiator and team 
members are the Participants. 
 
1.10.3 Cluster Algorithm 
Clustering is the classification of objects into different groups, or more precisely, the 
partitioning of a data set into subsets (clusters), so that the data in each subset (ideally) share 
some common trait - often proximity according to some defined distance measure. In this 
simulation, modified K-means algorithm is used. The K-means algorithm is an algorithm to 
cluster objects based on attributes into k partitions. It is similar to the 
expectation-maximization algorithm for mixtures of Gaussians in that they both attempt to 
find the centers of natural clusters in the data. It assumes that the object attributes form a 
vector space. The objective it tries to achieve is to minimize total intra-cluster variance, or, 
the squared error function , where is a chosen 
distance measure between a data point x
i
(j)
 and the cluster center c
j
. J is an indicator of the 
distance of the n data points from their respective cluster centers. The algorithm is composed 
of four steps:  
1. Place K points into the space represented by the objects that are being clustered. These 
points represent initial group centroids. 
2. Assign each object to the group that has the closest centroid. 
3. When all objects have been assigned, recalculate the positions of the K centroids. 
4. Repeat Steps 2 and 3 until the centroids no longer move. This produces a separation of the 
objects into groups from which the metric to be minimized can be calculated [55]. 
 
1.11 Tactical Federate 
Tactical Federate makes the decision of what action the UAV team will perform by 
selecting a strategy for the team. It engages the team federates and launches the high 
resolution. It communicates with Filter and High resolution. It forwards information 
exchanged between low resolution and high resolution to dynamically update them and to 
 23
 24
make them consistent. It is better to attack a battalion by UAV team than by a single UAV. If 
there is no strategy for the team, the UAVs will attack the targets randomly and may not be 
efficient or effective. A strategy is needed for this scenario. There are two strategies that can 
be used for a team in this simulation. Tactical Federate will select the proper strategy for the 
team at specific situations to make its attack efficient and effective. Bayesian Network is used 
for making decisions to select the strategy. A Bayesian network (or a belief network) is a 
probabilistic graphical model that represents a set of variables and their probabilistic 
dependencies. Based on the calculation of Bayesian Network, the probability of using each 
strategy is given and the strategy with the larger probability is selected as the strategy for 
team attack. The strategy is evaluated at each period of time during the attack by 
recalculation of the Bayesian Network. The strategy with the larger probability is selected for 
the UAV team which is acting at high resolution. 
Tactical Federate communicates with high resolution use HLA RTI provided by Mak. 
HLA RTI is based on UDP. Tactical Federate sends information to high resolution to build a 
scenario of the UAV team attacking a battalion. The high level resolution also sends back 
UAV team and battalion information to Tactical Federate for evaluating the current strategy 
and for updating the 2D visualization which is generated by Matlab. The information is sent 
back to Filter, and then to Simulink to update the visualization of low resolution. Tactical 
Federate also sends information to Low Resolution Visualization module (LRV) to generate a 
2D visualization. 
 
 25
1.12 High Resolution 
1.12.1 Introduction 
High resolution simulates the scenario of the UAV teams attacking battalions. It 
generates a 3D visualization to show this UAV team action. UAVs attack battalions based on 
the strategy generated at Tactical Federate. Since the battle field scenario is dynamic, and the 
UAV or target may be destroyed at any time, the strategy may not always be efficient and 
effective during the attacking. Strategy is needed to be evaluated and updated during the 
attacking. Since high resolution sends the battlefield information to Tactical Federate, 
strategy can be evaluated at each period of time based on the information. After evaluating 
the strategy, Tactical Federate sends back the proper strategy for this period of time. The UAV 
teams in high resolution will use the strategy sent from Tactical Federate. High resolution is 
simulated by VC++.Net using HLA Mak RTI and VR-link. It communicates with Tactical 
Federate using the HLA RTI which is a product provided by Mak. The visualization is 
generated by Stealth which is a product also provided by Mak. In this simulation, there are 
multiple high resolutions with their respective visualizations generated by Stealth. Different 
teams or one team using different strategies are simulated at their respective high resolutions. 
 
 26
 
1.12.2 VR-Link, HLA Mak RTI, Stealth 
VR-Link, HLA Mak RTI, Stealth are products provided by Mak and used for high 
resolution simulation which simulates UAV team attacking battalion and shows a 3D 
visualization. 
 
VR-Link 
VR-Link is a toolkit. It can quickly and easily use network simulators and virtual 
reality applications using either the U.S. DoD?s High Level Architecture (HLA) or the 
Distributed Interactive Simulation (DIS) protocol. 
 
HLA Mak RTI (Run Time Infrastructure) 
High Level Architecture (HLA) is a generally purpose architecture for distributed 
computer simulation systems. Using HLA, computer simulations can communicate to other 
computer simulations regardless of the computing platforms. Communication between 
simulations is managed by a run-time infrastructure (RTI). 
Mak RTI is the fastest way for High Level Architecture (HLA) compliant simulations to 
communicate. This efficient and easy-to-use tool eases concerns about the performance of the 
HLA by keeping CPU, bandwidth, and memory requirements to an absolute minimum. It is 
suitable for battlefield simulation. 
 
 
 27
Stealth 
Stealth is a 3D visualization tool that focuses on information. It provides the most data 
about user?s networked virtual world, and presents it in a clear and easily accessible way. 
 
1.13 Strategy for UAV Team 
This simulation makes the UAV team acting effectively and efficiently supported by 
using proper strategy in specific situations. The simulation strategy should be based on the 
experts make decision with their experiences that would improve efficiency and effectiveness 
of the simulation. There are three unique strategies that are used for attacking. One is Fringe 
Point strategy, one is region strategy and another is Destroy Communication strategy. To 
determine the effectiveness of alternative Course of Action (COA), the simulation is 
dynamically updated and branched with simultaneous execution of simulations, potentially at 
different levels of resolution. Strategies are dynamically evaluated and may be switched at 
each period of time to make the UAVs act efficiently and effectively through high resolution. 
 
Figure 1.1 Unmanned Aerial Vehicles (UAV). Predator 
 28
M1
M2
M3
Controller
 
 
 29
Figure 1.2 Multiresolution Model 
Figure 1.3 Mulitstage Model 
 
M1
M2
M3
Controller
SubM1
SubM2
 
 
 
 
 
 
 30
 
 
Figure 1.4 Simulation Updating and Branching for Multisimulation 
Sim
1
Monitor
Agent
Sim
1.1
Sim
1.2
Sim
1.11
Sim
1.12
Monitor
Agent
Sim
1.2
Sim
1.11k
Sim
1.12
Sim
1.2
Monitor
Agent
FP
FP
FP
Simulation updating and branching
 
 
 
 
 
 
 
 
 
 
 31
Figure 1.5 Matlab Simulink 
 
 
 
 
 
 
 
 
Figure 1.6 Formation Flight of UAV Team and Flock of Birds (Song et al., 2005) 
 
 
 
 
 
 
 
 32
 33
CHAPTER 2: MOTIVATIONS 
This simulation simulates the multiple UAVs act multiple works including team 
attacking. These works need to be done efficiently. It is a multisimulation, multresolution 
multistage system supported by the decision making subsystem and contains several 
subsystems, resolutions and simulations. These resolutions and subsystems with decision 
making make the UAVs team work efficiently and different works can be performed at 
different simulations. The motivation is to make the subsystems run independently on 
different platforms and work parallel and cooperatively. Another motivation is to integrate the 
simulations with the decision making system. Each sub-simulation needs to work 
cooperatively with each other and exchange information in a timely manner. The same model 
of entity needs to run at different levels of resolutions and consistency should be maintained. 
The models need to be updated dynamically at runtime, and injected into the simulation, or 
stopped. These actions are based on the decisions made by the decision making system. 
These are also motivations of this simulation. 
There is a 2D visualization for low resolution and a 3D visualization for high resolution. 
These visualizations should be consistent with each other. That requires exchange update 
information between the two resolutions and they are dynamically updated. 
The adaptive simulations need other support by the decision making system such as 
targets clustering and team formation. Targets are needed to be clustered as battalion based on 
 34
cluster algorithm. A single UAV can not attack a battalion by itself. A team is necessary for 
the UAV to attack a group of targets. Strategies are used for the UAV team work. Because of 
the uncertainty of the scenario, strategies are needed to be evaluated at each period of time 
and updated to select the proper strategy for specific situations. That makes simulation 
working efficiently and effectively through the simulation. A naturalistic decision making 
system is needed to make a decision for the UAV action such as strategy selection at different 
simulation stages and levels of resolutions. These are also the desirable applications and 
motivations of this simulation. 
 35
CHAPTOR 3: OBJECTIVES 
This simulation contains multiple simulations, resolutions and subsystems. They are 
supported by the decision making system. The object of the simulation include making the 
parties cooperate work with each other, each party can be supported by the decision making 
system, UAVs can achieves efficiency and effectiveness when they work. The object also 
includes implements two resolutions with multiple stages, each resolution has its 
visualization to illustrate the scenarios. We need to implement strategies for UAV team in 
different situations. These strategies are evaluated based on the simulation results. The result 
of the evaluation should be that UAV team uses the proper strategy in different situations and 
always works effectively. 
To achieve these objects, some algorithms need to be made such as clustering 
algorithms to cluster the targets, and consensus algorithm. The simulation scenarios  we 
need to achieved are: the UAVs are launched at low resolution starting searching the 
battlefield; they sense the area and exchange data with the controller; they cluster the targets 
they found; when the group of targets is a battalion, they will form a team led by the team 
leader and make a consensus within the team members; they attack the battalion based on a 
strategy which is selected by the decision making system, and the strategy will be updated 
based on the run-time situations. This will make them attack in an efficient and effective 
manner.  
 36
The low resolution is presented in a 2D visualization and high resolution is presented in 
a 3D visualization. The two visualizations are also consistent with each other. The 
consistency is based on the timely information exchange between the resolutions. 
 37
CHAPTER 4: CHALLENGING PROBLEMS 
4.1 The Problem of Deep Uncertainty 
This simulation simulates a battlefield scenario. In the most realistic scenario, the 
situations dynamically changes when the scenario unfolds. Some scenarios may not use any 
decision style. Dynamic adaptive in decision making is necessary to deal with emergent 
conditions and make the decision making process flexible. This requires implementing the 
naturalistic decision making algorithm. Actions should be evaluated at each period of time to 
select the right strategy. Models should be updated at run-time. Each model needs to be 
consistent with the others. To fulfill these requirements, the following issues needs to be 
resolved. 
 
4.2 Multiple Simulations at Multiple Stages 
This is a project having multiple simulations, resolutions and models in which different 
resolutions works at different platforms. Making the models and levels of resolution 
consistent with each other is a big challenge in this simulation. Low resolution is simulated 
by Matlab Simulink. Filter and Tactical Federate is simulated by VC.Net. They are simulated 
by different software by using different programming language. They could be run on 
 38
different machines or platform, and are required to cooperate with each other and exchange 
data through the network. There should be a policy to enforce consistency between the parties. 
This requires a communication channel to connect these parties to share the information. 
Decision making for UAVs in the battlefield requires perceiving situations accurately. This 
requires the UAVs to sense the battlefield in a timely manner and work collaboratively with 
each other. The decision making system should have the function to analyze the data 
perceived by each UAV to make a decision. 
 
4.3 UAV Team Work with Strategy Support and Targets Clustering 
It is dangerous to a single UAV to attack a group of targets. The UAV team is needed in 
this situation. A team can accomplish tasks that single one can not afford. When a team is 
formed, the problem is which targets it will attack. Targets need to be clustered as battalions 
and attacked respectively. If a battalion is found, the UAV team is formed and attack the 
battalion cooperatively with a strategy. The big challenge is coordination of multiple UAVs 
forming and working cooperatively in a system. Some issues arise when a team is formed, 
such as how many team members should be in a team, who is the team leader, and which 
battalion should this team attack. Each UAV has its own sensing range and can only detect 
the targets within its sensing range that cause the difference between UAVs perceptions. A 
consensus of perceptions is needed to be confirmed within each UAV team to attack 
accurately. To attack efficiently and effectively requires that decision making system choose a 
right strategy for the team. Naturalistic decision making algorithm needs to be implemented 
 39
to select the strategy. To make the right decision to select a strategy, the UAVs and targets 
information should be considered and analyzed. The information should include number of 
UAVs and targets, area of the battalion and the communication capability of the targets. The 
selected strategy is needed to be evaluated periodically, and dynamically updated when the 
current strategy is not the best one. This process should work throughout the simulation to 
make the overall simulation effective. 
 
 40
CHAPTER 5: RELATED WORKS 
5.1 Distributed Task Allocation 
Multi-UAV is a popular research field. Multiple tasks need to be allocated to UAVs. 
The challenging problem is allocation of the tasks to multiple UAVs. Lemaire et al., (2004) 
presented a distributed tasks allocation scheme in Multi-UAV Context. The tasks the UAVs 
must achieve are essentially ?goto? tasks, which are generated either directly by an operator 
or a central system that analyses all the information acquired by the various UAVs. It requires 
maintaining a communication link between the UAVs and the central station. The UAV 
sensing range could affect task allocation. Sujite et al., (2005) presented a negotiation scheme 
that efficiently allocates tasks for multiple UAVs. The paper presents the effect of sensor 
ranges on the task allocation and compares the results with that of greedy strategy and a 
variation of the negotiation mechanism. The result shows that the negotiation based task 
allocation strategy performs far better than greedy strategy. The result also shows that when 
the sensor radius is low, negotiation with targets information exchanged between 
neighborhood UAVs performs better than without the targets information exchange. 
When the UAVs can not communicate with each other and the UAVs sensors have 
limited range to detect the targets and neighboring UAVs, and assess target status, the task is 
difficult to allocate. Sujit et al., (2005) presented a team theory for Multi-UAV task allocation. 
 41
The team theoretic approach can efficiently allocate UAVs to the targets under this constraint. 
Centralized system has some problems such as scalability with the number of UAVs, 
network delays when UAVs communicate with the central location, and central location of 
the system is susceptible to malicious attacks. The distributed system can resolve the 
problems. Dionne et al., (2007) presented a new decentralized task allocation algorithm for 
UAVs. It uses intermittent and asynchronous communications. The algorithm provides a 
decentralized task consensus (DTC) across the UAVs. The decision to communicate is based 
on the outputs of the task allocation procedure to reduce the communication overhead. When 
the task allocation is decentralized such as each UAV allocate its own tasks, the information 
available is in general different from one UAV to another. To achieve cooperative behavior 
with decentralized task allocation approach, a decentralized task allocation algorithm resilient 
to inconsistent information across the team can be selected.  This decentralized task 
allocation algorithm encourages consensus by asynchronous intermittent communications, 
each UAV communicates when the error between its local and shared information exceeds a 
given threshold. The communication objective is to achieve a consensus in the outputs of the 
task allocation rather than in the input information. The algorithm increases efficiency in the 
communications. It simulates a team of unmanned almost-lighter-than-air-vehicles (ALTAVs) 
equipped with local sensors and a communication system. The results shows that compared to 
decentralized information consensus (DIC) algorithm, the DTC algorithm is demonstrated to 
require less communications to achieve a similar or better cooperative behavior. Dasgupta 
(2006) presented a distributed automatic target recognition system using multi-agents UAV 
swarms. It introduces a system COMSTAR-UAV (COoperative Multiagent System for 
 42
automatic TArget Recognition using Unmanned Aerial Vehicles). The system uses swarming 
techniques inspired from insect colonies to perform automatic target recognition in a 
distributed way. 
 
5.2 Path Planning 
Path planning is one of challenging problems in the multi-UAV simulation. Ahmadzadeh et al., 
(2006) presented a path planning algorithm for time critical cooperative surveillance. It uses a set of UAVs 
and they contain body fixed camera. An Integer Programming based strategy is developed. Integer 
Programming is applied over a receding planning horizon with terminal cost to reduce the computational 
effort of the planner and to incorporate feedback. It incorporates highly constrained motion and sensor 
capabilities of the vehicles in the mathematically programming formalism. Kanchanavally et al., (2006) 
presented a control scheme via feedback linearization for three dimensional cooperative path planning of a 
class of interconnected systems which is UAVs in this simulation. It showed that the feedback linearization 
technique along with a distance varied repulsive profile allows UAVs to converge to a known target 
location without colliding with other. It uses Lyapunov stability analysis and the results show the 
conditions are stable under the system. 
 
 43
5.3 Consistency Maintenance 
Multiple UAVs simulation could require multiple resolutions. Different resolutions are 
work respectively and need to be consistent; different stages in one resolution also need to be 
consistent. Maintaining the consistency of the resolutions is important to the simulation. 
Reynolds et al., (1997) presented a paper titled ?Consistency Maintenance in Multiresolution 
Simulations?. This paper described that inconsistency may occur despite the existence of 
valid models at each resolution level. Cross-Resolution Modeling (CRM) attempts to build 
effective Multiresolution simulations. A common solution approach is to dynamically change 
the resolution of an LRE (or HRE) to match the resolution of other encountered entities. This 
dynamic change is defined as aggregation (HREs to LRE) or disaggregation (LRE to HREs). 
The traditional approach to CRM aggregation-disaggregation causes chain disaggregation 
and puts an unacceptable burden on resources. Multiple Resolution Entities (MREs) is 
designed in this paper. MREs are the foundation of a design that incorporates maintaining 
internal consistency. Maintenance of core attributes is as an approach to maintaining internal 
consistency within an MRE. MREs make possible the isolation of the issues related to 
maintaining consistency, and they enable possible efficiencies through concepts such as core 
attributes. This paper describes the means that maintain the consistency of the resolution with 
dynamic aggregation-disaggregation and resolve the problems incurred by it such as chain 
disaggregation, network flooding, transition latency, and mapping problems. 
Liu et al., (2007) presented some key techniques on the parallel implementation of 
Multiresolution models (MRM). It explains the problems of interaction conflict in MRM 
 44
which includes consistency maintenance of different resolution models of the same entity, 
interaction conflict and the problem of morphism interactions. Concurrent interactions of 
multi-resolution models are considered seriously both in syntactically and semantically. The 
paper presented two new concepts, named morphism attribute and morphism event. 
Morphism attributes mean representation of the inherent attributes of an entity at deferent 
resolutions. Morphism events mean an essentially same event emitted by models of different 
resolutions, also called morphism interactions. To solve the inconsistent problems, it 
presented three methods, mutex locking, conflict detecting and model-oriented method. It is a 
progress in using Multiresolution models in simulation. It enhances the validity of simulation 
and engagement fairness in the battlefield simulation. 
 
5.4 Multi-agents System 
Multi-agent techniques have been used in multi-UAV simulation and improved the 
simulation. Barber et al., (2005) presented a run-time multi-agents system which includes the 
sophistic agents? interactions, uncertain environment conditions and dynamic domain 
requirements. It provided some tools for the design and analysis of agent systems and 
implementations. The design tools help quickly select agent functionality, assign it to model 
and evaluate the agents? technologies. Trace is an analysis tool and is provided to evaluate the 
difference between designed behaviors and implemented behaviors. Some technologies are 
used such as belief revision, information source selection, and action selection. The agents 
use these technologies to provide a reliable information network and to support agents? 
 45
action-selection in an UAV target tracking simulation [23]. Karim et al., (2005) presented an 
agent-based autonomous mission control system for unmanned aerial vehicle (UAV). It 
explores two different approaches for adding autonomous control to an existing UAV. The 
first adds a layer over the flight control system to control the mission. The second uses 
human metaphor of agency and implements an autonomous controller based on a model of 
human decision making. The experiments showed that they are both feasible and useful to a 
human cognitive modeling approach providing autonomous UAV control [24]. 
Jang et al., (2005) presented two scalable dynamic agent distribution mechanisms for 
large-scale UAV simulations. One mechanism aims at minimizing agent communication cost, 
the other mechanism attempts to move agents from overloaded agent platforms to lightly 
loaded platforms. The mechanisms are fully distributed algorithms and analyze only 
coarse-grain communication dependencies of agents instead of their fine-grain 
communication dependencies. In this system, each computer node has one agent platform, 
which manages scheduling, communication, and other middleware services for agents 
executing on the computer node.  
It simulated large-scale micro UAV simulations involving up to 10,000 agents using the 
agent distribution mechanisms. The results show that this multi-agent platform is adaptive to 
improve the scalability of the entire system [25]. 
 
 46
5.5 Multi-agents Market-based System 
Market based system is widely used for task allocation. It can support allocation of 
UAVs tasks in the multi-UAV system. Chen et al., (2006) presented a market-based approach 
as the main mechanism for autonomous collaboration in the Autonomous Collaborative 
Mission Systems (ACMS). ACMS is an extensible architecture and behavioral planning / 
collaborative approach to achieve the requirements for UAV in Future Combat System (FCS) 
Unit of Action (UoA). The market-based collaboration approach has a two-stage task 
specification and negotiation process that can accommodate different mission planning and 
task allocation strategies. It provides flexible collaboration among a variety of heterogeneous 
UAVs for a broad range of missions [29]. Dasgupta et al., (2006) presented a market based 
distributed task allocation across swarm units comprising a swarm-based system. The 
individual swarm units use the task selection mechanism to determine the order of executing 
tasks in the environment. It uses a computationally simple, yet efficient dynamic pricing 
algorithm to solve the distributed task selection problem for swarming in the market based 
model.  
The simulation result showed that the market based task selection strategy performs better than other 
heuristic based task selection strategies and can reduce the communication overhead [30]. 
 47
5.6 Genetic Algorithm 
Since the battlefield scenario is dynamic and evolutional, genetic algorithm is useful to 
deal with the battlefield situations to support the UAVs acitons. Price et al., (2006) presented 
genetic algorithm directed self-organized search and attack UAV swarms. The paper 
investigated the use of a self-organization (SO) framework for evolving UAV swarm behavior. 
A genetic algorithm (GA) is used for UAVs evolving behavior to successfully locate and 
destroy stationary targets. To make the Multi-UAV systems autonomous, it uses the 
organizational concepts used by colonial insects, wolf packs and economics. The 
self-organization realizes the groups of UAVs as a singular macro system rather than a set of 
interacting individuals. It provided a SO model for engagement between targets and UAVs, 
and it demonstrates an architecture for future modeling and SO UAV swarm behavior 
evolution. The SO model allows for the evolution of a multi-agent system. In most cases, it 
provides a cooperative and useful set of UAV behaviors [26]. Barlow and Oh (2006) 
presented an analysis for genetic programming controllers for unmanned aerial vehicles. It 
assures that prior to operation that an evolved controller will not damage the vehicle. A 
poorly performing controller might damage the vehicle. Evolved controllers must be robust to 
noise in the environment to operate the vehicle safely. The navigation controllers of UAVs 
use multi-objective genetic programming. The simulation results show that best evolved 
controller performs better than two hand-designed controllers and is robust to many sources 
of noise [27]. Lamont et al., (2007) presented a comprehensive mission planning system for 
swarms of autonomous aerial vehicles (UAV) using multi-objective evolutionary algorithm. 
 48
The system deals with the problems including path planning, vehicle routing, and hierarchical 
architecture-based swarms? behavior. The system has a path planner which is multi-objective 
evolutionary algorithm-based terrain-following and parallel, and an evolutionary 
algorithm-based vehicle router. It aims to minimize cost and risk generally associated with a 
three dimensional vehicle routing problem (VRP). The system finally is an extensible 
developmental path planning model integrated with swarm behavior and tested with a parallel 
UAV simulation [28]. 
 
5.7 Evidential Map-building 
Yang et al., (2005) presented two evidential map-building approaches for Multi-UAV 
cooperative search. One is based on Bayesian theory and the other is based on 
Dempster-Shafer theory. It utilizes the generated maps into the UAVs path planning 
procedure to cooperatively localize targets in the environment. The paper described the 
problem that a team of Uninhabited Air Vehicles (UAVs) seek to find targets in an unknown 
and uncertain environment. The decision on where UAVs search next is driven by the aims to 
increase the chance of locating targets and possibly avoiding threats. For this reason, UAVs 
need a good model of the environment and they should be able to construct and consistently 
update their environment models based on the sensory information. The efficiency and 
quality of the environment model consequently affect the UAVs search planning and 
decision-making activities. Since the UAVs sensor information of the real world may be 
inaccurate and uncertain, different vehicles need to combine the sensory information to 
 49
obtain the best knowledge of the environment. The paper use evidential reasoning techniques 
to fuse the sensors? information which is used as evidence. It shares the information to each 
other. The two evidential methods incrementally build cognitive maps to direct a group of 
UAVs to perform cooperative search tasks. Both of the methods are suitable for sensor 
measurements obtained in a sequential manner. They also allow the information to be 
incorporated with formula of finite horizon optimal control for the multi-UAV cooperative 
path planning problem in the search tasks. The simulation results show that these map 
building methods are effective to deal with the problem of UAVs cooperative search [31]. 
 
5.8 Team Work 
In Multi-UAV system, multiple UAVs cooperatively work to fulfill various tasks in the 
battlefield. Team work is a key point for UAVs to work effectively. Jeyaraman et al., (2005) 
presented using formal methods to model and verify cooperative unmanned aerial vehicle 
(UAV) teams. Since the multi-UAV coordination for cooperative control is a time-critical, 
zero-fault tolerant activity involving dynamic planning and real-time decision making, the 
formal methods can be used in this application area. It simulated multiple UAVs coordinated 
arriving at a specified target using Dubins? curves. The simulation is modeled using the 
Kripke models of ?possible worlds?. This formal model then used model checking technique 
to verify safety and reach ability. This model can be reused and be able to address some 
scalability issues [33]. 
 
 50
Gowtham and Kumar (2005) presented a simulation of multi-UAV flight formation. 
UAVs perform tasks such as surveillance, reconnaissance and target attack etc. Some 
algorithms have been developed for formation flying. It took into account the static and 
dynamic obstacles such as wind gusts that are encountered during the mission. It simulates 
arbitrary numbers of vehicles in a virtual reality environment. The simulation results show 
that the predetermined path is maintained by alleviating the deviations. In this way, it avoids 
static and dynamic obstacle collisions, inter vehicle collisions, and maintains the flight of 
formation [34]. Corner et al., (2004) presented a parallel simulation of UAV swarms. The 
framework includes support for discrete event simulations. This allows for quickly 
integrating higher fidelity or specialized support models as needed. Speedup is not achieved 
with the current implementation [15]. 
Song et al., (2005) presented an orthogonal transformation-based robust adaptive 
control method for close formation flight of multi-UAVs. The method deals with general case 
of uncertain dynamics due to vortex and external disturbances when UAVs are in close 
formation. It developed robust and adaptive control algorithms to maintain the desirable 
separation of wingman with the leading UAV or team leader. Compared with other existing 
methods, the formation control scheme is simpler in structure and easier to implement. The 
control scheme does not need detailed system dynamic information [16].  
McLain et al., (2001) presented the cooperative control of UAV rendezvous. The paper 
addresses the problem cooperative rendezvous in which multiple UAVs are to arrive at their 
targets simultaneously. It describes the development of a rendezvous manager state machine 
and a cooperative control decomposition approach [21]. 
 51
Goktogan et al., (2003) presented the system architecture of a real time Multi-UAV 
simulator. The simulator is a real demonstration of multiple UAVs conducting both 
decentralized data fusion and control. It has testing and validation mechanism. The 
mechanism includes off-line simulation of complex scenarios, hardware-in-the-loop tests, 
validation of real test results, and on-line mission control system demonstrations. It presented 
CommLihX, a communication framework. The system allows simulation modules to 
communicate over single or multiple virtual channels [14]. 
 
 52
CHAPTER 6: GENERAL APPROACHES 
6.1 Introduction 
In this chapter, general simulation scenario and general approaches are described. The 
models described in this chapter include the low resolution model which is implemented in 
Matlab, Filter, Tactical Federate and high resolution model. Their general functions and 
cooperative works are illustrated. Some of the algorithms implemented in this simulation are 
also introduced. 
 
6.2 Simulation Scenario 
This simulation simulates multiple UAVs at the battlefield scenario. The UAV has 
multiple tasks including search, tracking, forming, attacking, and evaluation. The tasks are 
allocated based on the rum-time analysis of the UAVs and targets information.  
The simulation begins at low resolution which is simulated by Matlab Simulink. There 
is a 2D visualization for low resolution generated by Matlab. When the UAV is launched, it 
starts searching the battlefield. Each UAV has its own view of the battlefield based on its 
sensing.  
 
 53
The UAVs and targets information are sent to the Filter for analysis. Based on a cluster 
algorithm presented in this article, targets are clustered. The Filter will decide whether the 
target is single or belongs to a battalion. If a battalion is detected, the UAVs need to form 
teams to attack the battalion. Based on the Contract Net protocol, a team leader and several 
team members are selected to do the task. One team is assigned to attack a particular battalion. 
Because of the different views of each UAV and mistakes may be made by the UAV when it 
perceives targets, the accuracy of those views need to be assured. This means that the team 
members within one team need to make a consensus of their views. When a consensus is 
confirmed within one team, each member gets the consensus model.  
The team member will fly to the battalion location to attack. To make the attack in an 
efficient and effective manner, different strategies are needed to be used in different 
conditions. The strategies are generated by Tactical Federate. Three basic strategies are 
designed; one is Fringe Point Strategy, one is Region Strategy and the other is Destroy 
Communication Strategy. The strategy selection for the team at particular condition is based 
on the analysis by Bayesian Network. After the team formation and strategy selection, teams 
are engaged and high resolution is launched which is implemented by Mak VR-Link, HLA, 
Mak RTI. Team information, strategy information and battalion information will be sent to 
the high resolution simulation. Based on the information, high resolution simulation is 
initialized for the multi-UAV teams attack battalions. A 3D visualization is implemented by 
Mak Stealth to show this scenario.  
The team performance at high resolution is evaluated periodically based on the team 
and battalion run-time information. The evaluation work is done at Tactical Federate. The 
 54
suitable strategy is selected, and high resolution is updated with it. The low resolution and 
high resolution need to be consistent with each other such as the active UAVs and targets 
information in both resolutions. The low resolution will update its models when it gets the 
information from high resolution. The two resolutions exchange information periodically. To 
be consistent, models exchange data with each others and are updated at run-time. When the 
UAVs in the high resolution destroy all the targets or they are all destroyed by the targets, 
high resolution stops. When the low resolution is informed that high resolution stops, it 
allocates new tasks to the rest of UAVs. If the battalion is destroyed, the UAV will disperse 
and go on search the area to find and attack other targets. If the UAV team is destroyed, the 
rest of UAVs will form new teams to attack the battalion. 
 
6.3 Low Resolution 
Low resolution models are simulated by Matlab Simulink. The simulation begins from 
low resolution. The UAVs and targets are generated and initialized in Simulink. This part 
simulates the UAV independent action. Different UAVs have different colors. Since the UAV 
has a limited sensing range, it can?t sense the target outside its sensing range. All targets are 
initialized having distances between them and UAVs larger than the UAV sensing rang that 
means all targets are outside all the UAVs sensing range. UAVs can?t see any target at first. 
The first task of the UAVs is to fly toward the east and sense the area to look for the targets. 
When the UAVs fly and sense, the area is unfolded, and targets can be found by the UAVs. 
The targets found by UAVs will be marked black. Low resolution sends all UAVs and targets 
 55
information to Filter. Based on the feedback from Filter, the targets are clustered and different 
clusters are marked by different colors in the low resolution visualization. Filter also sends 
back the new tasks allocated to the UAVs to low resolution based on the analysis of the UAVs 
and targets information. The UAVs in low resolution then performs the tasks assigned to 
them. 
When a team is required, Filter sends back the team members and task information to 
low resolution. Based on the information, the UAV teams are formed and fly as a swarm to 
attack specific battalion which is assigned by Filter. In low resolution visualization, the UAV 
team members will fly to the battalion center and stop. At this point, high resolution is 
launched to perform the attacking scenario. The high resolution periodically sends back 
UAVs and targets information to low resolution and it dynamically updates the corresponding 
models.  
When the high resolution stops, low resolution is informed. It assigns new tasks to the 
active UAVs based on the information from high resolution. If the battalion is destroyed, the 
UAVs will go on searching the rest of the area to find the targets and attack. If there are active 
targets in the battalion and the UAV teams are destroyed, new teams may be formed to attack 
the battalion. A 2D visualization is generated by Matlab to show the scenarios of low 
resolution (see Figure 5.1). 
 56
6.4 Filter 
The function of the Filter includes: exchange information with low resolution and 
Tactical Federate, clustering the targets, making consensus model, and form UAV teams. The 
data received by Filter from the low resolution includes the UAVs and found targets 
information. The UAV information includes the UAV position and status. The target 
information is the position of the target which is found or seen by a UAV. Based on the target 
information, the analysis is processed. The targets are clustered using the cluster algorithm. 
The cluster function is a function by the distance between the targets. Each cluster is deemed 
as a battalion. Because of the different perception of each UAV, a consensus model is used to 
help assure the accuracy of the perception. Each team member gets the consensus model after 
confirming the consensus within the team. Filter sends back information such as UAV teams, 
target clusters and consensus to low resolution to update it. Filter also sends this information 
to Tactical Federate to select a strategy for each team. Filter also received information such as 
UAV and target from high resolution through Tactical Federate and sends the received 
information to low resolution to update it. In this way, the two resolutions can be consistent 
with each other. 
 
6.5 Tactical Federate 
The Tactical Federate functions as a control and conjunction between several models. It 
makes the decision of which strategy to be used by a particular UAV team to attack a 
 57
particular battalion. This decision making model makes sure the UAVs attack in an efficient 
and effective manner. This decision making model is based on Bayesian Network algorithm. 
The UAV team, targets and the battlefield information are the input of the decision making 
model, and the output are the different strategies with their respective probabilities of being 
used. This probability shows the chances of using the strategy in the particular conditions. 
Larger probability means having more chances to get the task done by using the strategy in 
the current situation. The strategy with the larger probability is selected for the team.  
Tactical Federate engage the teams and launches the high resolution model after it 
selected the strategy for teams. It initializes the high resolution with teams, strategy and 
battalions information. During the high resolution simulation, teams and targets information 
are sent back to Tactical Federate as evidences to evaluate the strategies. Tactical Federate 
evaluates the strategy by using the information as the input of the decision making model, 
and gets the probability of each strategy at the current situation. Based on the evaluation, 
strategy with larger probability is selected and high resolution updates the UAV team action 
based on the selected strategy.  
The Tactical Federate also serves as part of a communication tunnel between low 
resolution and high resolution. In a timely manner, it sends all the UAVs and targets 
information to Filter, and then Filter sends it to low resolution. When low resolution gets the 
information, it updates the corresponding models at run-time. In this way, low resolution and 
high resolution are consistent with each other. Tactical Federate also sends all of the UAVs 
and targets information to a 2D visualization for visualizing the low resolution.  
 
 58
This visualization is initialized by the information from low resolution and Filter, and 
then updated by the information from high resolution when the high resolution executes. 
 
6.6 High Resolutions 
High resolution is the simulation for the UAV team action with strategy. When it is 
launched, it is initialized with the UAV team, strategy and battalions information. In this part 
of simulation, the UAVs fly in formation and attack the assigned battalion with a strategy. 
UAV team members fly directly to the battalion, and attack the battalion based on the selected 
strategy. High resolution has multiple simulations and simulates multiple UAV teams 
attacking different battalions with particular strategies. All UAVs and targets information is 
sent to Tactical Federate periodically to evaluate the current strategy. After the evaluation, a 
new strategy may be selected for the team or keep the current strategy. The communication 
between Tactical Federate and high resolution is implemented by HLA Mak RTI and 
VR-Link which provide run-time and timely communication between the two parties. A 3D 
visualization for high resolution is implemented by Mak Stealth. It provides a dynamic 3D 
animation environment with high quality pictures (see Figure 5.2).
 
 
 
 
 
 
 
 
 
 
Figure 6.1 Low Resolution Visualization
 59
 
 
 
 
 
 
Figure 6.2 High Resolution Visualization 
 
 60
 61
CHAPTER 7: DETAILS OF DESIGN AND SIMULATION 
7.1 Introduction 
This chapter describes the design and simulation details. Low resolution, Filter, and 
Tactical Federate are illustrated with design architecture and implement details. Techniques 
utilized in those parts are also illustrated in detail. 
In low resolution, the UVAs are initialized and looking for target. Every time unit, the 
sensing information need to be sent to the Filter to analyze. The transmission protocol such as 
TCP and UDP can be used. Because the communication is run-time and time sensitive, UDP 
is chosen in this project. Voiding overhead of checking of whether every package arrived, 
UDP is faster and more efficient. 
In the Filter, targets information needs to be analyzed. The targets need to cluster. 
There are several clustering algorithms we can select. K-means is one of the simplest 
unsupervised learning algorithms that solve the well known clustering problem [10]. Because 
the K-means algorithm is sensitive to the initial randomly selected cluster centers, we use 
modified K-means algorithm to initialize each target as a cluster center. The team formation 
process is done at Filter. Because each UAV is an individual unit, it can choose whether or 
not to join in a team. Team leader need to ask the UAV to join in its team.  
 
 62
Contract Net protocol is a good protocol for this scenario since it works like electronic 
market buying and selling goods. UAV can decide whether to join the team by itself based on 
this protocol. 
Tactical Federate is the strategy decision generator. The strategy is used for the UAV 
team which attacks a battalion. Because the battlefield scenario is dynamic that makes the 
evidences such as number of target variable. Some of the evidences are not independent and 
affect each other. The different strategies may not absolutely good or bad for a specific 
scenario. We need a model to analyze the evidences of the battlefield and select a proper 
strategy for the team. Bayesian Network can be used to learn casual relationships and hence 
can be used to understand a problem domain and predict the consequences of intervention 
[34]. It is a directional probability dependence graph, using the evidences to analyze and get 
the probabilities of the causes to the evidences. Because the model has both a causal and 
probabilities semantic, it can combine the prior knowledge and data [34]. Some of the 
evidences can not be used for a specific scenario such as efficiency of the team since at the 
beginning of the attraction there is no efficiency. Because the model of Bayesian Network 
encodes dependencies among all variables, it readily handles situations where some data 
entries are missing [34]. Bayesian Network supports this situation and can analyze using all 
or some of the evidences. 
7.2 Low Resolution 
Low resolution is simulated by Simulink. Two main models, initialization model and 
communication model are described. 
 63
7.2.1 Initialization Model 
When the simulation starts, the initialization model is launched. In this model, the simulation 
environment is initialized including the battlefield size, the number of UVAs and each UAV?s information, 
the number of target and each target?s information. UAV information includes x and y position, speed, 
sensing range, waypoint, and color. Way point is generated by the way point model which is the 
functionality model for generating UAV way point. It calculates way point based on the UAV current 
location and destination location. Target information includes x and y position. In this simulation, there are 
8 UAVs and 20 targets. All UAVs start from west of the battlefield and fly at the same speed (see Figure 
7.1). The initialized way point of each UAV goes straight from the starting point to the east of the 
battlefield. Each UAV has its unique color. The targets are set all over the battlefield. All of them are black. 
The targets are unseen or transparent when the distance between them and each UAV is larger than the 
UAV sensing range. Since all the targets are set to be out of the sensing range of all UAVs at initialization, 
all targets are unseen to the UAVs and unseen in the low resolution visualization. When UAVs unfolds the 
battlefield and the distance between the target and an UAV is less than the UAV?s sensing range, the target 
is sensed by the UAV and appears in the low resolution visualization marked black (see Figure 6.1). 
7.2.2 Communication Model 
The communication model is one of the most challenging models in this simulation. 
Low Resolution runs in Matlab Simulink environment, which uses the Matlab language. On 
the other side, the filter runs in Visual stdio.net environment written in C. Since UDP is the 
protocol of communication, it is needed to be implemented in Matlab environment. Although 
 64
there is no UDP program written in Matlab language to communicate to UDP written in C, 
Matlab has the friendly functionality for using C or C++ within its environment. The code 
written by C or C++ can be compiled in Matlab. In Matlab, the S-Function, User-definable 
block can be written in C program and be used after the C program is compiled. In this way, 
the UDP code written in C can be used in Matlab. The communication model is built in the 
?Save Plot Data? block (see Figure 7.2). The transmitting data block is ?trans? and the 
receiving block is ?recv? (see Figure 7.3). They are S-Functions containing a function 
mdlUpdate which will be executed in each time unit. The time unit in this project is set to 0.1 
second (see Figure 7.4). The UDP piece of program is added to the mdlUpdate function and is 
executed in each time unit to send and receive data message between Simulink and Filter. In 
the Filter, there is a corresponding piece of UDP program to communicate with the UDP 
program in Simulink. In this way, Matlab program can communicate to the program written 
in Visual stdio.Net by using the UDP program written in C. The program sequence chart of 
communication between Simulink and Filter is shown in Figure 7.5. The program execution 
snapshot is shown in Figure 7.6. 
 
 65
7.3 Filter 
7.3.1 Introduction 
Filter analyzes the targets, clusters targets as battalion, forms UAV teams, and makes 
consensus within each UAV team. It also communicates with low resolution and Tactical 
Federate to exchange information in each time unit. 
 
7.3.2 Targets Clustering 
Since the targets are all out of the each UAV?s vision range, there is no target sensed by 
UAV. During the unfolding of the battlefield, the UAVs fly across the battlefield and the 
target is sensed when the distance between it and the specific UAV is less than the UAV 
sensing range. Targets data such as x and y positions are sensed by the UAVs. When Filter 
gets the targets information sensed by low resolution, it clusters the targets by using the 
modified K-means algorithm.  
The clustering process starts by assigning each target to a cluster to create 20 clusters if 
there are 20 targets sensed. The distance between the clusters is the same as the distance 
between the targets. The process finds the closest pair of clusters and merges them into a new 
single cluster. There are 19 clusters at this point. It then computes the distance between the 
new cluster and other clusters to find the shortest distance and merge them to create a new 
cluster. It then repeats the previous processes until the cluster size reaches the maximum 
cluster size which is 4. It can also stop when all distances between the clusters are larger than 
a threshold which means all the clusters are far way from each other and do not need to be 
merged. The computation of the distance between clusters uses the Average Linkage to 
calculate the mean distance between two clusters members. The formula of the calculation is:  
D(r, s) = 
sr
nn
1
??
==
r s
n
i
n
j
sjri
xxdist
11
,
)(
 
A simple clustering process is shown in Figure 7.7. When the clustering process stops, each 
cluster is deemed as a battalion. Different clusters are marked by different colors (see Figure 
7.8). 
 
7.3.3 Team Formation 
When the targets have been clustered, there are several battalions on the battlefield. A 
single UAV is not suitable to attack a battalion and UAV team is needed to attack in an 
effective manner. The first UAV who finds the battalion will be the team leader. The team 
leader will decide the team size N based on the battalion size. The team leader will send CFP 
(Call for Proposals) information to N  UAVs which have the minimal distance between them 
and the battalion center. Each UAV has a willing attribute which is a random integer between 
1 and 100 showing the willingness to be in a team. The UAV which received the CFP checks 
the willing attribute to decide to be a team member. If the willing of the UAV is between 0 
and 70, it is willing to be a team member. That means there are 70% chances that a UAV can 
accept the CFP request. If the UAV is willing to be the team member, it accepts the proposal. 
 66
 67
It sends acceptance information to the team leader and makes a team contract with team 
leader. If the UAV is not willing to be in a team, it sends rejection information to the team 
leader. The team leader informs the team member by sending a Confirm message. This 
process is shown in Figure 7.9. If there are not enough team members on the team after 
receiving all feedbacks from the N UAVs, the team leader will send CFP to the other UAVs 
based on the distance between them and the cluster center. The process stops when there are 
N UAVs on the team, or the team leader has received all rest of UAVs feedbacks.  
The program execution snapshot is shown in Figure 7.10. The number of UAVs on a 
team can be three, four, or five since there are eight UAVs total. When the first team is 
formed, the rest of the UAVs form another team based on the same protocol. For example 
(see Figure 7.10), there are two teams, team 1 and team 2. Team 1 has three team members 
which are UAVs having ids 5, 4, 6. Team 2 has five team members which are UAVs having 
ids 1, 2, 3, 7, 8. Team leader in Team 1 is UAV 5; team leader in team 2 is UAV 1.  
 
7.3.4 Consensus Model 
Since there are different perceptions of each UAV within a team, a consensus is needed 
to be made within a team. After a team is formed, the consensus process is loaded. After this 
process, consensus is confirmed within the team and each team gets a consensus id. This id 
indicates that the consensus perception is perception of the UAV team member which has the 
UAV id equaling to consensus id. 
 
 68
 The team will attack the targets seen by that UAV. For example (see Figure 7.10), 
team 1 will attack the targets seen by UAV 5, team 2 will attack the targets seen by UAV 3. 
 
7.3.5 Communication Functions of Filter 
The Filter is a middle part between low resolution and Tactical Federate. It 
communicates to other parts based on UDP. Filter communicates to low resolution is 
described in the section 7.2. Both Filter and Tactical Federate are simulated by Visual 
stdio.Net. Filter is written in C and Tactical Federate is written in C++. The communication 
between them utilizes UDP socket program. The data transferred between them are in the 
structure DataTransfer which includes UAV, Target, Cluster, and Team information. This data 
is sent to Tactical Federate and high resolution. The program sequence chart of 
communication between Filter and Tactical Federate is shown in Figure 7.11 and the program 
execution snapshot is shown in Figure 7.12. 
 
DataTransfer structure 
struct datatransfer 
{  
 team uteam[2]; // all teams information 
 TARGET ut[TN]; // all targets information 
 utrans u[UN]; //all UAVs information 
 ctrnas c[TN]; // all clusters information 
 69
 int noc; // number of cluster 
}; 
 
7.4 Tactical Federate 
Tactical Federate has the following functionalities: strategy selection for UAV team, 
strategy evaluation, team engagement, launching high resolution, and communication to 
Filter, to high resolution, and to low resolution visualization. Strategy selection, strategy 
evaluation and communications to other parties work at run-time. Communication with other 
parties is required to exchange information and ensure consistency between the parties. The 
Bayesian Network is used to select the proper strategy for the UAV team in a specific 
scenario. The UAV team is engaged in Tactical Federate based on the team information 
generated by team formation process in Filter. 
 
7.4.1 Strategy Selection 
To make the UAV team attack battalion in an efficient and effective manner, a strategy 
is used to direct the UAV team attacking.  
When the UAV team is formed by Filter and it informs the Tactical Federate, Tactical 
Federate inputs the information as evidences to the strategy decision model to select a 
strategy for the team. When the UAV team is acting at high resolution, the feedback 
information from high resolution is used as inputting evidences to evaluate the current 
 70
strategy, and a proper strategy is selected for the UAV team at current conditions. Bayesian 
Network is used as the algorithm to select the strategy. We use the Microsoft Bayesian 
Network tools MSBNx to implement the algorithm in this simulation. MSBNx is a tool for 
doing that kind of cost-benefit reasoning for diagnosis and troubleshooting. Based on that, it 
can select an efficient and effective strategy for the UAV team.  
The MSBNx implemented in this project is shown in Figure 7.13. The top nodes are 
the input nodes as the evidences of this model. They are Area, Betweenness, Degree, 
CommRang (Communication Range), TargetNumber (number of target), UAVNumber 
(number of UAV), FSEfficiency (Fringe Point Strategy Efficiency), FSEffectiveness (Fringe 
Point Strategy Effectiveness), DCSEfficiency (Destroy Communication Range Strategy 
Efficiency), DCSEffectiveness (Destroy Communication Range Strategy Effectiveness). The 
bottom nodes are DCS (Destroy Communication Strategy) and FS (Fringe Point Strategy) 
which are the strategies we can choose for UAV team action (see Figure 7.13). Area is the 
battalion?s approximate area. Betweenness is the number of geodesic paths that pass through 
a node. It is the number of ?times? that any node needs a given node to reach any node by the 
shortest path. Degree is the number of ties to other nodes, row or column sums of adjacency 
matrix. Communication range is the ability of one node can communicate to other nodes. 
Since the efficiency and effectiveness are not generated until the UAV team is attacking 
battalion, the FSEfficiency, FSEffectiveness, DCSEfficiency and DCSEffectiveness have no 
value before launching high resolution. 
The fringe point strategy is that we suppose there is a fringe around targets area and set 
several points on it. When UAV team attack these targets in the area, each team member will 
 71
fly to a point on the fringe. Then, the UAV will attack target within its sensing range in the 
area. After attacking the target within UAV?s sensing range, it will attack other targets in this 
area. The region strategy is that UAV team leader divide the area into several regions. It will 
allocate each UAV team member a region. Each UAV will attack targets in the allocated 
region. The basic idea of destroy communication strategy is to destroy the communication 
path between targets. The path means the communication route between two targets. For a 
given target, if the degree is low and betweenness is high that means the target is a critical 
target for other targets to communicate, since there are less paths income and outcome to the 
target and this target is on many of the shortest paths between other targets (see Figure 7.14). 
In this situation, destroy communication strategy has a larger probability to be selected. If 
there is not a critical target in the situation, the fringe point strategy is more preferred since 
the UAV team attack based on the fringe point strategy has more effectiveness than ad-hoc 
attack in general situations. 
The Bayesian Network in this simulation is built by MSBNx as the following process. 
First, a diagram is made shown in Figure 7.13. It illustrates the relations of the nodes. If node 
A decides or is the input to node B, A is parent node of B and there is an arrow from A to B. 
The number of targets (TargetNumber) and Area nodes decide the spread of the targets 
(TargetSpread) in the area which means: ?large? spread, a small number of targets are 
dispersed in a large area; ?small? spread, a large number of targets are dispersed in a small 
area. The spread of targets (TargetSpread), the number of the UAV (UAVNumber) and 
Degree nodes decide TargetsOverPowered node. DCSEfficiency (Destroy Communication 
Range Strategy Efficiency) and DCSEffectiveness (Destroy Communication Range Strategy 
 72
Effectiveness) nodes decide DCSOverall node. FSEfficiency (Fringe Point Strategy 
Efficiency) and FSEffectiveness (Fringe Point Strategy Effectiveness) nodes decide 
FSOverall node. TargetsOverPowered, Betweenness and DCSOverall nodes decide DCS 
(Destroy Communication Strategy) node. TargetsOverPowered, Betweenness and FSOverall 
nodes decide FS (Fringe Point Strategy) node. 
Second, we set uncertainties, the probabilities of various situations of each node in the 
Assessment module. For example, we set the ?number of targets? node to four levels in its 
assessment table: low, medium, high, critical with the probability as low 0.15, medium 0.35, 
high 0.35 and critical 0.35 (see Figure 7.15). The set of uncertainty is done at each node in 
the system. The node having more inputs has the larger Assessment table since the more 
situations are introduced. For example, the bottom node DCS has three parent nodes: 
DCSOverall, TargetOverPowered and Betweenness. Because DCSOverall has 2 situations, 
TargetOverPowered has 5 situations and Betweenness has 4 situations, the DCS has 40 
(2*5*4=40) situations total in its assessment table (see Figure 7.16). The red bar means the 
probability to select this strategy in the situations which are based on the parent nodes 
conditions, and the yellow bar means the probability not to select this strategy in the 
situations. The sum of the red probability and yellow probability is 1 in each situation. 
The program loads this diagram at run-time, inputting required information and 
calculating the probability for each strategy based on the diagram. The calculation result is 
shown in Figure 7.17. The UAV team uses the strategy with the larger probability to attack 
the battalion. When the UAV team is attacking the battalion, the UAV team and battalion 
information are sent back to Tactical Federate to evaluate the current strategy in each period 
 73
of time. The efficiency and effectiveness for each strategy are generated at high resolution 
and sent back with the other information. The information is input to the decision making 
model using the Bayesian Network to calculate the probability of each strategy in the current 
situation. After each calculation, if the strategy with larger probability is different from the 
strategy currently used, the UAV team is informed to use the strategy with larger probability. 
If the strategy currently used has the larger probability, the UAV team still uses the current 
strategy to attack in the next period of time. The evaluation works at run-time and the 
strategy update is dynamical at high resolution. In this way, the UAV team can attack in an 
efficient and effective manner during the high resolution simulation. Eff = NoTar/T is used to 
calculate the efficiency (Eff). In this formula, NoTar is the number of targets destroyed by the 
team and T is the time the team spends to do the job. 
Results show that UAV team using different strategies based on the evaluation having 
almost same efficiency. That means the team can attack efficiently at different situations. 
In simulation, we implemented fringe point strategy with region strategy for UAV team. 
In the simulation with fringe point strategy, there is a virtual fringe of the targets area and 
there are four points on this fringe (see Figure 7.18). Each UAV fly to a point of the fringe at 
the beginning and attack the target close to the point. The UAVs firstly attack the targets close 
the fringe, and then attack targets in the inner of the fringe (see Figure 7.19 and Figure 7.20). 
Region strategy is based on the dividing the targets into several regions. Each UAV will 
deal with different regions in this simulation. A region is created by using a single point as a 
reference. Around the reference, the region is created. The algorithm to create region is: 
1. Pick a reference point; 
 74
2. Get next N amount of closest targets from the reference point. 
3. Aggregate these targets as a region. 
4. Repeat from step 1 to 3 until all targets have been put into a region. 
Based on the region strategy, the targets are divided into regions and each UAV is 
assigned to attack a region. When a UAV destroy all targets in its region, it will most likely 
move to the next closest region. In this simulation, we set N is 2 or 3 which means there are 
two or three targets in one region. In figure 7.21, there are nine targets and they are divided 
into four regions circled by different colors. There are four UAVs to attack these targets and 
each UAV is assigned to attack the targets in a region. When a UAV finishes its work in the 
region, if there are targets left in other regions, it moves to the closest region among those 
regions. 
 
7.4.2 Communication Functions of Tactical Federate 
Tactical Federate communicates to Filter, Team Federate, high resolution and low 
resolution visualization. The program sequence chart is shown in Figure 7.22. The 
communication between Tactical Federate and Filter, Tactical Federate and low resolution 
visualization are based on UDP. The communication between Tactical Federate and high 
resolution uses the VR-Link and HLA Mak RTI provided by Mak. All parties communicate 
and exchange data with each other in a timely manner during the simulation. This 
periodically data exchange ensures that the parties are consistent with each other. 
 
 
 
 
 
 
 
 
 
Figure 7.1 Low Resolution at Initialization Stage 
 
 
 
 75
 
 
 
 
 
 
 
 
 
 
 
Figure 7.2 ?Save Plot Data? Block in Simulink 
 
 76
 
 
 
 
 
 
 
 
 
 
 
Figure 7.3 ?Save Plot Data? inside Structure 
 
 
 77
 
 
 
 
 
 
 
 
 
 
 
Figure 7.4 Time Unit of the S-Function is set to 0.1 second
 78
 
 
 
 
 
 
 
 
 
 
 
Figure 7.5 Simulink and Filter Communication Sequence Chart 
 
 79
 
 
 
 
 
 
 
 
 
 
Figure 7.6 Communication between Simulink and Tactical Federate 
 
 
 
 
 
 
 
 
 
 
 80
 
 
 
 
 
 
 
 
 
 
 
T1 
T2 
T3 
  
 
Target 1 
Target 2 
Target 3 
 
 
 81
Figure 7.7 A Simple Targets Clustering Process 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 7.8 Targets are Clustered as Battalions 
 82
 
CFP
CFP
CFP
Team
Leader
UAV3
UAV1 UAV2
CFP
UAV4
 
 
 
Accept
Reject
Accept
Team
Leader
UAV3
UAV1 UAV2
Accept
UAV4
 
 
 
Confirm
Confirm
Team
Leader
UAV3
UAV1 UAV2
Confirm
UAV4
 
 
Figure 7.9 Team Formation Process Based on Contract Net 
 83
 
 
 
 
 
 
 
 
 
 
 
Figure 7.10 Team Formation Process Results 
 
 84
 
 
 
 
 
 
 
 
 
 
 
Figure 7.11 Filter and Tactical Federate Communication Sequence Chart 
 85
 
 
 
 
 
 
 
 
 
 
 
Figure 7.12 Communication between Filter and Tactical Federate 
 
 
 86
 
 
 
 
 
 
 
 
 
 
 
 
Figure 7.13 MSBNx Implementation
 87
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 88
 
 
 
 
3
5
8
9
8
7
9
7
4
8
8
4
6
7
3
5
5
4 3
4
4
6
4
7
4
8
8
4
Figure 7.14 Sample of Critical Target 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 7.15 Assessment Table of ?Number of Target? Node 
 
 89
 
 
 
 
 
 
 
 
 
 
Figure 7.16 Assessment Table of DSC Node 
 90
 
 
 
 
 
 
 
 
 
 
Figure 7.17 Calculation Results of Bayesian Network 
 91
 
 92
Figure 7.18 Simulation using Fringe Point strategy (1) - Fringe and Fringe Points of 
Figure 7.19 Simulation using Fringe Point Strategy (2) 
Targets Area 
 
 93
 
Figure 7.20 Simulation using Fringe Point Strategy (3) 
 
 
Figure 7.21 Simulation using Region Strategy 
 
 
 
 
 
 
Figure 7.22 Program Sequence Chart of  
Tactical Federate Communication Functions 
 
 94
 95
CHAPTER 8: SIMULATION RESULTS COMPARISON 
In this simulation, we use two strategies, fringe point strategy and region strategy, to 
support UAV team attack battalion. The UAVs have different performance when they use the 
two strategies in the same situations. 
Figure 8.1 shows the targets that are destroyed and the time lapse. The targets are 
destroyed by using region strategy is faster at the beginning and slower at the end than by 
using fringe point strategy. UAVs spend a little less time using fringe point strategy to destroy 
all the targets than using region strategy. Figure 8.2 shows the rate of UAV destroying targets 
using region strategy is much faster at the beginning and slower at the end compared to using 
fringe point strategy. The average target destroy rate of using fringe point strategy is 0.17 
targets/second and the rate of using region strategy is 0.14 targets/second. 
The reason of this is that using fringe point strategy, the UAVs firstly fly to the fringe 
points and then attack the closest target. While using region strategy, the targets areas are 
divided into four regions. Each UAV is assigned to attack one region at the beginning. The 
UAV do not need to fly to the fringe point and then attack, they perform in a region, a smaller 
area than the entire targets area. That makes UAVs spend less time to start to attack the 
targets. UAVs using fringe point strategy destroyed targets closest to the fringe and move to 
attack the targets in the inner region of the fringe. After about half of the simulation time, the 
rest of targets have smaller area compared to the initialized targets area. Because the updated 
 96
fringe is smaller than before, it takes less time for UAV fly to the new fringe points to attack. 
Compared to the fringe point strategy, by using region strategy, some regions are very large 
and it takes some time for UAV fly from one target to another. In the smaller region, the 
UAVs could destroy all targets in a short time, but in the larger region, the UAVs spend a lot 
of time flying between the targets such as the region marked black in Figure 7.21. Figure 8.1 
shows by using region strategy, the eighth target is destroyed at 44 seconds and the ninth 
target is destroyed at 66 seconds. Because the average target destroy rate by using region 
strategy is 0.14, in 20 seconds, there could be 2.8 targets to be destroyed, but in reality, there 
are no target destroyed during this time. The UAV that destroyed the last target, spent 41 
seconds fly from the previous destroyed target to the last one. That causes the region strategy 
to spend more time to destroy all the targets than the fringe point strategy. The average target 
destroy rate using region strategy is smaller than that using the fringe point strategy. If the 
last target is not count in, the average target destroy rate using fringe point strategy is 0.19 
and the rate using region strategy is 0.17. That
strategy is a little faster  this situation. 
The results shows that UAVs using region strategy destroy target faster at the beginning 
than at the end; UAVs using fringe point strategy do not perform much differently. In Figure 
8.1, using region strategy, the fourth target is destroyed at 13 second; using fringe point 
strategy, the fourth target is destroyed at 27 second. The average of target destroy rate of 
using fringe point strategy at 13 second is 0.17 targets/second; the rate of using region 
strategy is 0.31 targets/second. These results show that using region strategy has larger target 
destroy rate at the beginning of the simulation (13/66=20%, the simulation lasts 66 seconds). 
 means on the average, using fringe point 
 at destroying target than using region strategy in
 97
The overall average of target destroy rate of using fringe point strategy is a bit larger than 
using region strategy. For the battlefield scenario, although using region strategy on the 
average is er at the 
beginning. That has some advantages for UAVs at the battlefield. UAVs destroy targets fast at 
the be
 of time. Compared to region strategy, using 
fringe
inge point strategy 
to atta
 a little slower at destroying targets, the UAVs destroy targets fast
ginning could be safer for them than destroying them later. Because the targets in the 
real world could fire back to UAVs, destroying targets earlier could result in fewer threats to 
the UAVs and the UAVs will be safer for the rest
 point strategy has on the average a little larger target destroy rate and much smaller 
rate at the beginning of the simulation. UAV using this strategy are more dangerous. If the 
targets could fire back, there are larger probabilities for UAVs to be destroyed at the 
beginning than using region strategy. Suppose there are some UAVs destroyed at the 
beginning, there will be not as many UAVs to attack targets as before and it could take longer 
time for UAVs to fulfill the tasks and target destroy rate could dramatically drop down at the 
rest of time. In such a situation, it is not a good choice for UAVs to use fr
ck the targets.  
In this situation, using region strategy has more advantages and could fulfill the task 
with smaller number of UAVs destroyed compared to using fringe point strategy.  
Efficiency is number of targets destroyed in a period of time. Effectiveness is defined 
as T/U. T is the number of targets destroyed in a period of time and U is the number of UAVs 
destroyed plus 1 in that period of time. That means if no UAV is destroyed at that period of 
time, effectiveness will be the number of targets destroyed at that period. Because the target 
cannot fight back and UAV is not destroyed. Effectiveness equals to efficiency. In this 
 98
simulation, efficiency is shown in Figure 8.3. In the situation where targets could fire back, 
UAVs using fringe point strategy could have larger probabilities to be destroyed at the 
beginning. The efficiency of using the fringe point strategy could be smaller at beginning 
than it is shown in Figure 8.3. There could be less UAVs and many targets left after a while 
from beginning (a quarter of the total simulation time). Because UAVs destroy target faster 
using the region strategy than using fringe point strategy at the beginning, there will be more 
UAVs and fewer targets left after that period time. UAVs have larger probabilities not be 
destroyed, while UAVs using fringe point strategy are more vulnerable. The average 
efficiency of using region strategy could be larger than using fringe point strategy in this 
situation. 
 
 
 
 
 
 UAV Destroy Target Time by Using Different Strategies
0
20
30
50
70
Destroyed Target
Ti
e
 (se
c
d
)
10
40
60
123456789
m
o
n
Fringe Point
Region
 
Figure 8.1 UAV Destroy Target by Time Lapse 
 
 99
 
 
 
 
 
UAV Destroy Target Rate by Using Different Strategies
0.00
0.05
0.10
0.20
0.25
6 1218243036424854606
Tot
a
a
r
g
e
t
s
t
or
d b
y
ti
 (t
a
r
g
/s
ec
o
d
)
0.15
0.30
Time (second)
l
 t
s
 de
ye
 
m
e
e
ts
n
Fringe Point
Region
  
Figure 8.2 Target Destroy Rate by Using Different Strategies 
 
 100
 101
 
 
 
 
Efficiency of UAV Destroy Target Using Different Strategies
0
1
2
3
4
6 12182430364248546066
time (Second)
E
f
f
i
c
i
e
n
c
y
 (
t
a
r
ge
t
s
/
s
e
c
ond)
Fringe Point
Region
 
Figure 8.3 Efficiency of UAV Destroy Target Using the Strategies 
 
 102
CHAPTER 9: SUMMARY AND FUTURE WORKS 
With the populating of using UAV in battlefield, multiple UAVs could work together to 
fulfill multiple tasks. Targets are clustered as battalions based on each UAV?s perception. The 
UAV teams are implemented to attack battalions and have a consensus of perception within 
each team. The UAVs actions are important to make the attack efficient and effective. The 
decision making model using Bayesian Network to selects the proper strategy to support the 
UAV team in attacking in an efficient and effective manner in different situations. This 
simulation co  models in 
ultistage. Different parties work on different platforms and have a timely data exchange. 
The sub models can be evaluated, updated and upgraded dynamically at run-time. In this way, 
the consistencies between sub models and between resolutions are maintained. Multiple 
visualizations are utilized for both low resolution and high resolution. The animate and 
friendly multiple visualizations show the different stages of the resolutions of this simulation. 
That is convenient for the human controller and users to perceive the simulation. 
Future works address the simulation of active targets. The targets could move and fire 
UAVs. That will make the simulation more like real world battlefield. Selecting a strategy for 
a team based on the evaluation of current battlefield by decision model and run-timely 
changing the strategy for the team will be simulated to achieve better effectiveness. Other 
issue will address interaction with human controller at run-time. The key of the decision 
ntains multisimulation, multiresolution and decision making
m
 103
an control with decision model supports. Combination of human control 
achine computation will ensure more accurate decision. The GUI of high resolution 
could be a tree structure, and each leaf of the tree represents a simulation among the multiple 
simulations. The human controller or users could pick up a leaf node on the tree to perceive a 
simulation such as a UAV team attacks a battalion using a specific strategy in the 
visualization. 
 
control could be hum
and m
 104
[1] Co
2004. 
 a distributed 
9, NO. 12, 
DECE
REFERENCES 
rner, J.J.; Lamont, G.B. Parallel simulation of UAV swarm scenarios. Simulation Conference, 
Proceedings of the 2004 Winter. Volume 1, Issue , 5-8 Dec. 2004 Page(s): - 363 
[2] DoD. Modeling and Simulation (M&S) Glossary. DoD Publication 5000.59-M, Jan. 1998. 
[3] P.K. Davis, and R. J.Hillestad. Experiments in Multiresolution Modeling. RAND Report. MR1004. 
RAND Corp.1998. 
[4] Levent Yilmaz and Alvin Lim. Multisimulation with Multi-resolution, Multi-stage Models over 
Federated Simulation Infrastructures. October 2006. 
[5] Brown, D. A. 1998. Medusa?s Mirror: Stepping Forward to Look Back ?Future UAV Design 
Implications From the 21st Century Battlefield?. School of Advanced Military Studies; United States 
Army Command and General Staff College. 
[6] Davis, C. R. 1995. Airborne reconnaissance: The leveraging tool for our future strategy. Executive 
Research Project, National Defense University. The Industrial College of the Armed Forces. 
[7] McHale, J. 2003. Unmanned aircraft armed and dangerous. Military and Aerospace Electronics 14 
(11): 18?25.  
[8] French, M. 2003. Uavs advance since desert storm. Available online via <www.fcw.com/fcw/ 
articles/2003/0217/web-uav-02-19-03. asp>[accessed August 19, 2003]. 
[9] Reid G. Smith. The Contract Net Protocol: high-level communication and control in
problem solver. IEEE TRANSACTIONS ON COMPUTERS, 1104 VOL. C-2
MBER. 
[10] J. B. MacQueen (1967): "Some Methods for classification and Analysis of Multivariate 
Observations", Proceedings of 5-th Berkeley Symposium on Mathematical Statistics and Probability, 
Berkeley, University of California Press, 1:281-297 
[11] Lemaire, T.; Alami, R.; Lacroix, S.; A distributed tasks allocation scheme in multi-UAV context. 
Robotics and Automation, 2004. Proceedings. ICRA '04. 2004 IEEE International Conference on 
Volume 4, Apr 26-May 1, 2004 Page(s):3622 - 3627 Vol.4. 
 105
ation, 2003. Proceedings. ICRA apos;03. IEEE International Conference on 
, 14-19 Sept. 2003 Page(s): 2720 - 2726 vol.2 
[15] Joshua J. Corner, Gary B. Lamont. Parallel simulation of UAV swarm scenarios. Proceedings of 
the 36th conference on Winter simulation WSC '04. December 2004 
[16] Song, Y.D.; Yao Li; Liao, X.H.; Orthogonal transformation based robust adaptive close formation 
control of multi-UAVs. American Control Conference, 2005. Proceedings of the 2005. 8-10 June 2005 
Page(s):2983 - 2988 vol. 5. 
[17] Dionne, Dany; Rabbath, Camille A.; Multi-UAV Decentralized Task Allocation with Intermittent 
Communications: the DTC algorithm. American Control Conference, 2007. ACC '079-13 July 2007 
Page(s):5406 ? 5411. 
[18] Ahmadzadeh, A.; Jadbabaie, A.; Kumar, V.; Pappas, G.J.; Multi-UAV Cooperative Surveillance 
with Spatio-Temporal Specifications. Decision and Control, 2006 45th IEEE Conference on 13-15 
Dec. 2006 Page(s):5293 ? 5298. 
[19] Kanchanavally, S.; Ordonez, R.; Schumacher, C.J.; Path planning in three dimensional 
environment using feedback linearization. American Control Conference, 2006. 14-16 June 2006 
Page(s):6 pp. 
[20] Paul F. Reynolds, Anand Natrajan, Sudhir Srinivasan. Consistency maintenance in 
multiresolution simulation. July 1997, ACM Transactions on Modeling and Computer Simulation 
(TOMACS), Volume 7 Issue 3.  
[21] McLain, T.W.; Chandler, P.R.; Rasmussen, S.; Pachter, M.; Cooperative control of UAV 
rendezvous. American Control Conference, 2001. Proceedings of the 2001. Volume 3, 25-27 June 
2001 Page(s):2309 - 2314 vol.3. 
[22] Liu Baohong, Hao Jianguo, Huang Kedi. Parallel implementation of multi-resolution models in 
distributed simulation system. May 2007, Proceedings of the 18th conference on Proceedings of the 
18th IASTED International Conference: modelling and simulation MOAS'07.  
[12] Prithviraj Dasgupta. Distributed automatic target recognition using multi-agent UAV swarms. 
International Conference on Autonomous Agents. Proceedings of the fifth international joint 
conference on Autonomous agents and multiagent systems. SESSION: Robotics, Pages: 479 ? 481. 
2006 
[13] Sujit, P.B.; Ghose, D. Multiple UAV search using agent based negotiation scheme. American 
Control Conference, 2005. Proceedings of the 2005. Volume , Issue , 8-10 June 2005 Page(s): 2995 - 
3000 vol. 5. 
[14] Goktogan, A.H.; Nettleton, E.; Ridley, M.; Sukkarieh, S. Real time Multi-UAV Simulator. 
Robotics and Autom
Volume 2, Issue
 106
[23] K. S. Barber, N. Gujral, J. Ahn, D. DeAngelis, K. Fullam, D. C. Han, D. N. Lam, J. Park. Design, 
runtime, and analysis of multi-agent systems. July 2005, Proceedings of the fourth international joint 
conference on Autonomous agents and multiagent systems AAMAS '05. 
[24] Samin Karim, Clint Heinze. Experiences with the design and implementation of an agent-based 
autonomous UAV controller. Industrial ations. July 2005, Proceedings of the 
fourth international joint conference on Autonomous agents and multiagent systems AAMAS '05.  
. International 
Conference on Volume , Issue , April 18-21, 2005 Page(s): 85 - 90 
[26] Ian C. Price, Gary B. Lamont. Military applications: military analysis I: GA directed 
 WSC '06. 
nary robotics, adaptive behavior. July 2006, 
Proceedings of the 8th annual conference on Genetic and evolutionary computation GECCO '06. 
Computational Intelligence in Multicriteria Decision 
Making, IEEE Symposium on 1-5 April 2007 Page(s):10 ? 20. 
et-Based 
Collaborations for Autonomous Operations of Unmanned Air Vehicles. Distributed Intelligent 
73 ? 278.  
Sw l conference on Intelligent 
Agent Technology IAT '06. 
rch. American Control Conference, 2005. Proceedings of the 2005 June 8-10, 2005 
Page(s):116 ? 12. 
ehicle. Computational Intelligence and Security, 
2006 International Conference. 3-6 Nov. 2006. Volume: 2, On page(s): 1225-1228 
 track, aerospace applic
[25] Myeong-Wuk Jang; Gul Agha. Scalable agent distribution mechanisms for large-scale UAV 
simulations. Integration of Knowledge Intensive Multi-Agent Systems, 2005
self-organized search and attack UAV swarms. December 2006, Proceedings of the 38th conference 
on Winter simulation
[27] Gregory J. Barlow, Choong K. Oh. Robustness analysis of genetic programming controllers for 
unmanned aerial vehicles. Artificial life, evolutio
[28] Lamont, G.B.; Slear, J.N.; Melendez, K.; UAV Swarm Mission Planning and Routing using 
Multi-Objective Evolutionary Algorithms. 
[29] Liang Chen; Gregory, B.; Easley, M.; Peot, M.; Lee, J.; Altshuler, T.; Mark
Systems: Collective Intelligence and Its Applications, 2006. DIS 2006. IEEE Workshop on 15-16 June 
2006 Page(s):2
[30] Prithviraj Dasgupta, Matthew Hoeing. Market-Based Distributed Task Selection in Multi-agent 
arms. December 2006, Proceedings of the IEEE/WIC/ACM internationa
[31] Yanli Yang; Minai, A.A.; Polycarpou, M.M.; Evidential map-building approaches for multi-UAV 
cooperative sea
[32] Huahui Yi; Bifeng Song; Dongchao Ji; Tianxiang Yu. Experiment Research on Situation 
Awareness of the Operators for Unmanned Aerial V
 
 107
ol. 7 
[35] Gowtham, G.; Kumar, K.S.; Simulation of multi UAV flight formation. Digital Avionics Systems 
, J.K.; An overview of emerging results 
in cooperative UAV control. Decision and Control, 2004. CDC. 43rd IEEE Conference on Volume 1, 
[37] Zhongjian Li; Jinwen An; Dong Xu; HIL Simulation of High Altitude Long Endurance UAV's 
[38] Edrich, M.; Schmalenberger, R.; Combined DSSS/FHSS approach to interference rejection and 
p. 
-22 Oct. 2005 Page(s):F2H-15 - F2H-20 
 - 2-345 
6 April 2006 Page(s):5 pp. 
[33] Suresh Jeyaraman; Tsourdos, A.; Ratal Zbikowski; White, B.; Formal techniques for the 
modelling and validation of a co-operating UAV team that uses Dubins set for path planning. 
American Control Conference, 2005. Proceedings of the 2005. 8-10 June 2005. On page(s): 4690 - 
4695 v
[34] David Heckerman; A Tutorial on Learning with Bayesian Network. Microsoft research, 
Advanced technology division, 1995. MSR-TR-95-06. 
Conference, 2005. DASC 2005. The 24
th
 Volume 2, 30 Oct.-3 Nov. 2005 Page(s):6 pp. Vol. 2. 
[36] Ryan, A.; Zennaro, M.; Howell, A.; Sengupta, R.; Hedrick
14-17 Dec. 2004 Page(s):602 - 607 Vol.1 
Target Tracking/Flight Integrated Control. Intelligent Control and Automation, 2006. WCICA 2006. 
The Sixth World Congress on Volume 2, 21-23 June 2006 Page(s):6188 ? 6192. 
navigation support in UAV communications and control. Spread Spectrum Techniques and 
Applications, 2002 IEEE Seventh International Symposium on Volume 3, 2002 Page(s):687 - 691 
vol.3. 
[39] Borrelli, F.; Subramanian, D.; Raghunathan, A.U.; Biegler, L.T.; MILP and NLP Techniques for 
centralized trajectory planning of multiple unmanned air vehicles. American Control Conference, 
2006 14-16 June 2006 Page(s):6 p
[40] Augeri, C.; Neebel, D.; Baird, L.; de Freitas, A.; UAV Communications: Integrating a Real-World 
Scenario with Computer Architecture. Frontiers in Education, 2005. FIE '05. Proceedings 35th Annual 
Conference 19
[41] Si, Yang; Xiaomin, Li; Hui, Xie. UAV Servicing and Training System Based on VR and HLA. 
Electronic Measurement and Instruments, 2007. ICEMI '07. 8th International Conference on Aug. 16 
2007-July 18 2007 Page(s):2-341
[42] Ye, Z.; Bhattacharya, P.; Mohamadian, H.; Majlesein, H.; Yongmao Ye; Equational dynamic 
modeling and adaptive control of UAV. System of Systems Engineering, 2006 IEEE/SMC 
International Conference on 24-2
[43] Jiecai Luo; Some new optimal control problems in UAV cooperative control with information 
flow constraints. American Control Conference, 2003. Proceedings of the 2003 Volume 3,  4-6 June 
2003 Page(s):2181 - 2186 vol.3. 
 108
[46] Wigley, G.; Jasiunas, M.; A low cost, high performance reconfigurable computing based 
t of unmanned aerial vehicles in 
adversarial environments. American Control Conference, 2005. Proceedings of the 2005. 8-10 June 
.J.; Chandler, P.; UAV team decision and control using efficient 
collaborative estimation. American Control Conference, 2005. Proceedings of the 2005 on 8-10 June 
ithm 
based offline/online path planner for UAV navigation. Systems, Man and Cybernetics, Part B, IEEE 
 Control System of UAV Based on 
Man-in-loop Real-time Operation. Control Conference, 2007. CCC 2007. Chinese July 26 2007-June 
Algorithm to Optimize UAV Tight Formation Flight. Computational 
Engineering in Systems Applications, IMACS Multiconference on Volume 1, 4-6 Oct. 2006 
Varaiya, P.; Task planning and execution for UAV teams. Decision and 
Control, 2004. CDC. 43rd IEEE Conference on Volume 4, 14-17 Dec. 2004 Page(s):3804 - 3810 
.; Ariola, M.; Ciniglio, U.; Corraro, F.; Pironti, A.; Virgilio, M.; Algorithms for 3D 
UAV Path Generation and Tracking. Decision and Control, 2006 45th IEEE Conference on 13-15 Dec. 
ing. Collaborative 
Technologies and Systems, 2005. Proceedings of the 2005 International Symposium on 15-20 May 
005 Page(s):347 ? 352. 
[44] Dongli, Yuan; Jianguo, Yan; Xinmin, Wang; Qingbiao, Xi; A Study of Information Fusion for 
UAV Based on RBF Neural Network. Control and Automation, 2007. ICCA 2007. IEEE International 
Conference on May 30 2007-June 1 2007 Page(s):2839 ? 2842. 
[45] Cook, Kendra L. B.; The Silent Force Multiplier: The History and Role of UAVs in Warfare. 
Aerospace Conference, 2007 IEEE on 3-10 March 2007. Page(s):1 ? 7. 
unmanned aerial vehicle. Aerospace Conference, 2006 IEEE 4-11 March 2006 Page(s):13 pp. 
[47] Mehdi Alighanbari; How, J.P.; Cooperative task assignmen
2005 Page(s):4661 - 4666 vol. 7 
[48] Shima, T.; Rasmussen, S
2005 Page(s):4107 - 4112 vol. 6 
[49] Nikolos, I.K.; Valavanis, K.P.; Tsourveloudis, N.C.; Kostaras, A.N.; Evolutionary algor
Transactions on Volume 33, Issue 6, Dec. 2003 Page(s):898 ? 912 
[50] Fengqing, Zhang; Quancun, Kong; Distributed Remote
31 2007 Page(s):119 - 122 
[51] Yun-an Hu; Bin Zuo; Xiaodong Li; The Application of an Annealing Recurrent Neural Network 
for Extremum Seeking 
Page(s):613 - 620 
[52] Sousa, J.; Simsek, T.; 
Vo l . 4 .  
[53] Ambrosino, G
2006 Page(s):5275 ? 5280.  
[54] Lundell, M.; Tang, J.; Nygard, K.; Fuzzy Petri net for UAV decision mak
2
 109
[56] A Tutorial on Clustering Algorithms. 
[55] Yan Jin; Minai, A.A.; Polycarpou, M.M.; Cooperative real-time search and task allocation in 
UAV teams. Decision and Control, 2003. Proceedings. 42nd IEEE Conference on Volume 1, 9-12 Dec. 
2003 Page(s):7 - 12 Vol.1. 
http://home.dei.polimi.it/matteucc/Clustering/tutorial_html/kmeans.html