A survivability framework for autonomous systems

Concordantly, understanding how survivability arises in biological systems may provide valuable insights for applying similar principles to artificial systems in order to enhance their a

AUTONOMOUS S YSTEMS

QUEK BOONKIAT

NATIONAL UNIVERSITY OF SINGAPORE

2008

AUTONOMOUS S YSTEMS

QUEK BOON KIAT

(B.Eng (Hons.), NUS)

A THESIS SUBMITTED FOR THE DEGREE OF

Autonomous systems are being envisioned for enhancing human capabilities especially

in harsh or dangerous operating conditions Despite considerable advances, achievingautonomous operation remains a major challenge Current approaches predomi-nantly rely on the integration of disparate components or functional modules such

as perception, navigation, actuation and decision making, into complex, stand-alonesystems While notable progress has been achieved, few systems remain operationalwhen subject to unexpected situations or mishaps Apart from physical limitations

in sensors and hardware, the difficulty is exacerbated by the need to control andcoordinate the operation of multiple components or capabilities that constitutes thewhole system Addressing these limitations requires system architectures that canfulfil the operational and functional requirements of the intended application, whileleveraging on the systems’ capabilities despite their inherent imperfections

Many biological organisms in nature (e.g mammals) possess innate, autonomous

capabilities that enable them to survive Concordantly, understanding how survivability

arises in biological systems may provide valuable insights for applying similar principles

to artificial systems in order to enhance their autonomous operation Leveraging

on the survivability paradigm, this thesis introduces an architectural framework to

address the autonomous operation of artificial systems by exploiting the effectiveness

of survival-oriented processes occurring within biological organisms This is directed atenhancing the decision making capabilities of artificial systems in order to increase their

level of autonomy In this framework, the concept of survivability is introduced as an

important operational requirement, forming an integral part of a system’s architectureand governing its design process

The Survivability Framework incorporates the understanding of physiological, chological and cognitive processes underlying the ability of biological systems to survive

psy-when subject to unexpected challenges, especially those involving potential or immediate

danger It introduces the notions of needs and emotions as state representations whose

interactions form the mechanisms that motivate and regulate the behaviour of an

artificial system Within this context, survivability is defined as the ability of a system

to fulfil its needs These needs represent the gaps between current and desired system states, whereas emotions serve as indicators for monitoring the fulfilment of these needs Based on these representations, survivability then emerges from determining suitable

from which a selection of system architectures can be instantiated according to therequirements of the intended applications

The design and deployment of autonomous systems is realized with the steps defined

by the Survivability Framework This process is illustrated in the design walk-through,

simulation and implementation of an autonomous surveillance vehicle The feasibility

of the proposed framework is demonstrated in several use cases that highlight the

interactions between needs, emotions and actions in controlling and regulating the

behaviour of the autonomous surveillance vehicle

This thesis is an outcome of asking some very big questions What makes an entityautonomous? What is missing in the current understanding of how autonomy can beachieved in artificial entities? What approach should one adopt in any attempt toincrease the autonomy of an artificial system, such as a mobile robot, or an autonomouslyguided vehicle?

While these questions have been asked for decades since the advent of moderncomputing, many of them are still unanswered The difficulty with achieving autonomy

boils down to a lack of understanding of what being autonomous means to an artificial

system, such as a mobile robot This is still the case today, as queries on the nature

of autonomy remains centre-stage in artificial intelligence and robotics In particular,autonomy is often discussed within the framework of Agent Theory (AT) Agent Theory

is itself, an attempt at understanding autonomy by prescribing some of its knownrequirements, for instance, percepts, and actions, to artificial entities that may possiblyenable them to react in an autonomous manner to the environment (and increasingly, totheir own internal states) Despite its comprehensiveness, AT is just one among manyother possibilities in which this could be accomplished

By observing biological entities in nature, one expresses amazement at the multitude

of seamless activities occurring within these entities that gives them the ability

to act in an autonomous manner, and react to the environment in ways that cansafeguard their survival Most of the activities endowing a biological organism withits autonomous behaviours are closely related to those that safeguard its survival From

such observations, the survivability of an organism becomes an important and necessary

condition for the autonomy of a system This is the tenet that the thesis is based on

To understand the elements of survivability, a correlation between the needs of a

biological entity and its survival is formed Many of the activities that safeguard its

survival are driven by the presence of needs that the organism fulfils, via a constant

adaptation of its behaviour to meet these needs The different intensities of such

to these principles, they in return form the basic premises upon which an architecturalframework is constructed The aim is to mirror the processes occurring in nature, andapply them in the design and deployment of autonomous systems, in a biologically-inspired manner

With the approach laid out above, this thesis presents what can be termed a

survivability paradigm, i.e the notion that survivability is an intrinsic capability of

a system that makes it autonomous, and that autonomy can be enhanced therefore,

by addressing the survivability of a system Following this paradigm, a Survivability

factors that motivate and regulate behaviour in an artificial system Organization

of needs, emotions and actions into various relationships gives rise to a reference

architectures for an autonomous system The implementation of the framework in theform of these architectures is shown in a collection of case studies involving both realand simulated robots

The results obtained in this thesis are encouraging as a foray into the

multi-disciplinary field of cognitive and affective robotics Though this thesis is adopting such

an approach, its objective is not to mimic exactly the cognitive processes operating withinbiological systems; rather, the attempt is directed at understanding such processes,learning from them and determining the ways in which these principles can be applied

to the engineering of artificial systems

As the increased adoption of cognitive models of intelligence in future becomes a cleartrend, it is hoped that this thesis can serve as a contribution in closing the explanatorygap between cognitive and affective models of human intelligence and behaviour, withthe system engineering perspective of artificial systems design

Boon Kiat, Quek

July 2007

“Hofstadter’s Law states that:

It always takes longer than you expect, even when you take into account

Hofstadter’s Law.”

– Douglas Hofstadter, in his 1979 magnum opus,

Gödel, Escher, Bach: An Eternal Golden Braid

This is the single most difficult piece of writing I have ever done As the conclusion

of many years of education, this thesis bears testimony to many cycles of inspiration,discovery, frustration, disappointments, and inspiration all over again, over the course ofseveral years The journey was not lonesome, however It is to all the following esteemedpersons that I have met along the way who have helped me in some way or another inthe pursuit of this thesis, that I am grateful for

I would like to express heartfelt gratitude to my supervisor, Dr Lim Khiang Wee,without whose encouragements, knowledge and advice, and the intellectual freedom andlatitude that he has allowed me, I would not have been able to undertake this research.Similarly, the greatest influence over the daily direction of my research was perhapsprovided by my co-supervisor, Dr Javier Ibañez-Guzmán, who continued to guide metirelessly with an equal mix of inspiration and motivation, despite his relocation back toFrance For guiding me into the research path as an outcome of the successful completion

of an honours year dissertation, are Dr Prahlad Vadakkepat and Dr Xu Jian Xin, both ofwhom are members of my thesis advisory committee It is my great privilege to be undertheir continued guidance in the course of my candidature, where our interactions are notmerely those between student and supervisor, but more like those between friends.The patience and support of my family is a constant stream of reassurance For that Iwish to thank my father and mother, Mr Quek Kim Meng and Madam Ang Li Kian, fortheir care and nurture, and my brothers, Ben and Ronnie Quek, for showing constant

interest in, and endless amusement at the robot in this thesis Thanks also goes to mycousins Pauline & Irene Quek, for their support and love.

Heartfelt appreciation goes to my fellow sufferers, namely fellow Ph.D candidates:Kiew Choon Meng, Lee Hui Mien, Joe Yen Yen, Gan Hiong Yap, Mustafa ShabbirKurbanhusen, Daniel Teo Tat Joo, Jason Teo Chek Sing, Tey Ju Nie, Fua Cheng Heng,Goh Chwee Sim, Lim Wei Ying, Wong Shek Yoon, for sharing this journey together, andespecially Ang Su Fen, for her constant support and nurture, for sharing the final stages

of her candidature together, and being that beacon of light in the darkest moments.Despite the difficulty of it all, we have had a wonderful four years together

Sincere gratitude to a special few, who are forever loyal, patient, and always believing

in me: Dr Teh Huey Fang, Tan Hui Tong, Dr Florencia Edith Wiria, Wendy Tan GeakHuang, Chua Chen Yun, Wei Xiao Qin, Cynthia Quah, Jovice Ng Boon Sing, Tang Ying,Ada Lo, and especially my godsister, Joanna Koh Ying Ting My fellow buddies, withwhom I had shared the significant years of my tertiary education, for all the crazy things

we used to do together, not just due to the need for an outlet from work, but for ourimpulsive whims: Donny Soh Cheng Lock, Ng Chu Ming, Geoffrey Koh Yeow Nam, HanBoon Kiat, and Yeow Wai Leong My long-time friends, who have shown support andbelieve in what I do: William Sze and Song Chenying, Tan Chin Wee, Loy Hui Chienand his wife Xinyue, Lim Boon Pang, Ng Kwan Jee, Varian Lim and Emily Loh, Loh TzuLiang, Walter Lim and Soh Sook Leng, Lin Wei Ling, Lam Lap, Lee Seong Per, RhysWong, Mark Sin, Mahadev Appan, and Yee-wah Lee Jefferson

Many friends and colleagues with whom I have had the privilege of working with,have assisted me greatly, be it over suggestions on where to look, or the occasionalbrainstorming over a cup of tea From SIMTech: Dr Lim Ser Yong, Dr Lin Wei, Dr.Andrew Shacklock, Dr Ong Jiun Keat, Mr Ng Teck Chew, and Ms Jeannette Lim ShoukGaem From the Mechatronics and Automation Laboratory, NUS: Mr Chan Kit Wai,

Dr Tang Kok Zuea, Tan Chee Siong, Jim Tan Shin Jiuh, Liu Xin, and Dr Xiao Peng.Heartfelt appreciation to members of staff from the NUS Graduate School for IntegrativeSciences and Engineering, for all the administrative assistance and advice throughout

my whole candidature

Finally, I wish to thank the Agency for Science, Technology and Research (A*STAR)for scholarship support and funding throughout the entire duration of the candidature,and my sureties, Ang Choon Ngiap and Quek Kim Hoe (my godfather), without whomall of this would not have been possible

Figure 1.2(a):http://world.honda.com/news/2005/photo/c051213/index.html

Figure 1.2(b):http://http://www.ubergizmo.com/photos/2006/9/hitachi-emiew.jpgFigure 1.2(c):http://www.sony.net/SonyInfo/QRIO/top_nf.html

Figure 1.2(d):http://www.irobot.com/index.cfm

Figure 2.1(a):http://marsrovers.jpl.nasa.gov

Figure 2.1(b):http://www-rocq.inria.fr/imara/site2/gallery/2004_antibes/Figure 2.1(c):http://www.defenselink.mil/news/May2004/n05172004_200405173.htmlFigure 2.1(d):http://www.darpa.mil/grandchallenge/

Figure 2.2(a),(b),(c),(d):http://www.darpa.mil/grandchallenge04/

Figure 3.2: sprites are courtesy ofhttp://www.starwars.com/databank/droid/

P ART I C ONCEPTUAL F RAMEWORK AND F OUNDATIONS 1

1.1 Background 2

1.2 Understanding autonomous systems 4

1.2.1 Defining autonomous systems 4

1.2.2 Potential applications 5

1.2.3 Motivation for autonomous systems 7

1.3 Challenges in autonomous systems 10

1.3.1 Operational and functional challenges 11

1.3.2 Situational awareness and decision making 12

1.3.3 System engineering and integration 13

1.3.4 The Survivability Paradigm 13

1.4 Statement of the problem 14

1.4.1 Thesis statement 14

1.4.2 Problem context 15

1.4.3 Issues addressed 15

1.5 Research objectives 15

1.6 Approach 16

1.7 Contributions of this thesis 18

1.8 Thesis organization 18

Chapter 2 Towards Autonomous Systems 20 2.1 Introduction 20

2.2 State-of-the-art in autonomous systems 21

2.2.1 Service and assistive robotics 21

2.2.2 Field robotics 22

2.2.3 Defence and homeland security 23

2.2.4 Emerging developments 27

2.3 Issues in the development of autonomous systems 29

2.3.1 Operational requirements 29

2.3.2 Functional capabilities 30

2.3.3 Computational and implementation challenges 38

2.4 Formulating the problem of autonomous operation 42

2.4.1 Action-selection 43

2.4.2 Mission and task planning 44

2.4.3 Motion planning 45

2.4.4 Autonomous operation 46

2.4.5 Critical assessment 46

2.5 Architectures for autonomous systems 46

2.5.1 Hierarchial top-down architectures 47

2.5.2 Behaviour-based and reactive architectures 47

2.5.3 Layered and distributed architectures 48

2.5.4 Hybrid and biologically-inspired architectures 48

2.5.5 Critical assessment 52

2.6 Conclusion 53

Chapter 3 Foundations of Survivability in Autonomous Systems 55 3.1 Introduction 55

3.2 Survivability in autonomous systems 56

3.3 Survivability in information systems 57

3.4 Survivability in biological systems 59

3.5 Affect and emotion-related aspects of survivability 60

3.5.1 Emotion-related survivability processes 60

3.5.2 Influences of emotions on intelligence and behaviour 61

3.6 Motivation as a means of achieving survivability 62

3.6.1 Motivation in biological systems 62

3.6.2 Human motivation and needs theories 64

3.7 Cognitive basis for survivability 65

3.7.1 Elements of human decision making 66

3.7.2 Human decision making and information processing 67

3.7.3 Situational awareness, attention and memory management 72

3.8 Incorporating survivability into autonomous systems 74

3.8.1 Survivability as a means of achieving autonomy 74

3.8.2 Survivability as an operational requirement 75

3.8.3 Survivability as a process of needs-fulfilment 75

3.8.4 Extension of needs to autonomous systems 76

3.8.5 Needs as the basis for affect, motivation and behaviour 79

3.8.6 Emotions as a regulatory mechanism for behaviour 79

3.8.7 Survivability specification as a structure for the framework 80

3.9 Conclusion 81

Chapter 4 The Survivability Framework 84 4.1 Introduction 84

4.2 Defining architectural frameworks 85

4.3 Underlying principles of the framework 85

4.4 Overview of the framework 86

4.5 Functional perspectives 87

4.5.1 The ENVIRONMENTperspective 88

4.5.2 The PREFERENCESperspective 88

4.5.3 The MECHANISMSperspective 89

4.5.4 The CAPABILITIESperspective 89

4.5.5 Functional perspectives as the structure of the framework 89

4.5.6 Functional perspectives as a survivability specification 90

4.6 Needs of an autonomous system 91

4.6.1 Identifying the needs of an autonomous system 91

4.6.2 Representation of operational requirements 92

4.6.3 State-space formulation of needs 92

4.6.4 Determination of needs 96

4.7 Emotions of an autonomous system 100

4.7.1 Functions of affective states and emotions 101

4.7.2 Types of affective states and emotions 102

4.7.3 Formulation of emotions 104

4.7.4 Determination of emotions 107

4.8 Actions and generation of behaviour 108

4.8.1 Correlation between needs, actions and emotions 109

4.8.2 Selection of desired actions 110

4.8.3 Needs fulfilment as a closed loop system 111

4.8.4 Managing unanticipated scenarios 111

4.9 Reference architecture 112

4.9.1 Implementing the Survivability Reference Architecture 113

4.9.2 Action-selection policy 117

4.9.3 Comparison with existing architectural approaches 118

4.9.4 Survivability metrics 121

4.10 Conclusion 122

P ART II F RAMEWORK R EALIZATION AND D EMONSTRATION 124 Chapter 5 Framework Realization 125 5.1 Introduction 125

5.2 Overview of the realization process 126

5.3 Motivating scenario 126

5.3.1 Scenario description 126

5.3.2 Concept of operations 127

5.4 ENVIRONMENT 128

5.4.1 Identifying tasks and operating conditions 128

5.4.2 Designing sensing and perception systems 129

5.5 PREFERENCES 130

5.5.1 Identifying the needs of a system 130

5.5.2 Determining needs and their fulfilment levels 131

5.5.3 Identifying the emotions of a system 131

5.6 CAPABILITIES 132

5.6.1 Designing actions and behaviours to fulfil needs 133

5.6.2 Needs decomposition 134

5.6.3 Determining actions from decomposed needs 134

5.6.4 Implementing actions 136

5.6.5 Designing relationships between needs and actions 137

5.6.6 Designing relationships between emotions and actions 138

5.7 MECHANISMS 141

5.7.1 Architectural design considerations 141

5.7.2 Architectural realization for autonomous surveillance vehicle 142

5.8 Conclusion 145

Chapter 6 Framework Demonstration 147 6.1 Introduction 147

6.2 Experimental overview 148

6.2.1 Defining use cases 148

6.2.2 Claims and premises to be validated 149

6.2.3 List of use cases 150

6.3 Navigation and exploration 150

6.3.1 Experimental setup 150

6.3.2 Formulation 153

6.3.3 Results 155

6.4 Target tracking and following 158

6.4.1 Experimental setup 158

6.4.2 Formulation 158

6.4.3 Results 162

6.5 Surveillance and intrusion detection 164

6.5.1 Experimental setup 165

6.5.2 Formulation 167

6.5.3 Results 172

6.6 Conclusion 182

Chapter 7 Conclusions 183 7.1 Introduction 183

7.2 Summary of results and contributions 183

7.3 Future research directions 187

Appendix A Emotional Processes in Biological Systems 189

A.1 Neuroanatomy of emotion-related survivability processes 189

A.1.1 Emotions and the limbic system 189

A.1.2 Emotion regulation 190

A.1.3 Emotion and memory 190

A.1.4 Emotion and rewards 191

Appendix B Survivability Metrics 192 B.1 Metrics for assessing the survivability of a system 192

B.1.1 Balanced Needs Aggregate (BNA) 192

B.1.2 Root Mean Square Needs Aggregate (RMSA) 192

B.1.3 Priority-weighted Needs Aggregate (PWNA) 193

B.1.4 Weakest-link Needs Aggregate (WLNA) 193

B.1.5 Inverted-priority Needs Aggregate (IPNA) 193

B.1.6 Analytical example 193

Appendix C Systems and Demonstrators 195 C.1 Feature-based Perception Demonstrator 195

C.1.1 Experimental system overview 195

C.1.2 Representation of the environment 198

C.1.3 Representation of features 199

C.1.4 Ground plane detection 200

C.1.5 Obstacle detection 202

C.1.6 Free-space detection 203

C.1.7 Discontinuity detection 207

C.1.8 Feature fusion 207

C.1.9 Results 209

C.1.10 Discussion 212

C.2 Omnidirectional Camera System 213

C.2.1 Hardware configuration 213

C.2.2 Algorithms 214

C.2.3 Results 215

C.3 SURV 2AGV: a surveillance and survivability test-bed 219

C.3.1 Concept of operations 219

C.3.2 Hardware components 220

C.4 Urban Search and Rescue Simulator 223

Appendix D Supporting Case Studies 224

D.1 Introduction 224

D.2 Interpreting visual information for autonomous navigation 224

D.2.1 Experimental setup 224

D.2.2 Formulation 226

D.2.3 Results 227

D.3 Reactive navigation in indoor and outdoor environments 235

D.3.1 Experimental setup 235

D.3.2 Formulation 235

D.3.3 Results 238

D.4 Vehicle following and platooning 244

D.4.1 Experimental setup 244

D.4.2 Formulation 244

D.4.3 Results 246

D.5 Conclusion 247

Appendix E Publications 252 E.1 List of articles published during the candidature 252

List of Figures

1.1 Hardware demonstrators – past and present 6

1.2 Examples of service and assistive robots 8

1.3 Technology roadmap for development of intelligent systems 10

1.4 An illustration of the requirements faced by an autonomous system 12

1.5 Reaching into the survivability characteristics of multiple disciplines from an autonomous systems perspective 17

2.1 Unmanned ground vehicles 25

2.2 Snapshots from the 2004 DARPA Grand Challenge 26

2.3 Functional capabilities of an autonomously guided vehicle 31

2.4 The LAAS Architecture for Autonomy 49

2.5 The DAMN Architecture 50

2.6 The Autonomous Robot Architecture (AuRA) 50

2.7 The 4D-RCS Reference Architecture 51

2.8 The EMIB Computational Architecture 52

3.1 Many activities performed by biological systems involve those that safe-guard their survival 63

3.2 A biological systems’ analogy for processes necessary to ensure the sur-vival of autonomous systems 63

3.3 Elements of the human decision making process 67

3.4 A model of human information processing 68

3.5 A three-level model of human behaviour 69

3.6 An integrated model for adaptive decision making 70

3.7 The CogAff schematic framework 70

3.8 Information-processing model of decision making 73

3.9 A Need indicates the existence of a gap between current and desired results 76

3.10 Mapping from Maslow’s Hierarchy of Needs to Autonomous System

Sur-vivability Needs 77

3.11 Relationship between positive and negative affect, and needs fulfilment 80

4.1 The functional perspectives of the Survivability Framework 88

4.2 An illustration of the process of needs-fulfilment 111

4.3 Reference Architecture for realization of the Survivability Framework 113

4.4 A layered behaviour-based realization of the Survivability Reference Archi-tecture 115

4.5 A Subsumption-type realization of the Survivability Reference Architecture 115 4.6 An information processing model of the Survivability Reference Architecture.116 4.7 A cognitive model realization of the Survivability Reference Architecture 116

4.8 A connectionistic realization of the Survivability Framework 117

5.1 Phases in the realization of the Survivability Framework 127

5.2 Tasks specifications for autonomous surveillance vehicle 129

5.3 Classification of actions for an autonomous vehicle 137

5.4 Realization of the Survivability Reference Architecture with subsumption of actions for an autonomous surveillance vehicle 143

5.5 Realization of the Survivability Reference Architecture with command fusion for an autonomous surveillance vehicle 143

5.6 Realization of the Survivability Reference Architecture with deliberative and reactive levels for an autonomous surveillance vehicle 144

6.1 Use Case 6.1: Implementation of connectionistic architecture for simu-lated robot 152

6.2 Use Case 6.1: Plots of changes in needs, actions (behaviours) and emotions 156 6.3 Use Case 6.1: Simulated P2DX robot in the USAR Simulator 157

6.4 Use Case 6.1: Overhead and perspective views of the simulated NIST Red Arena The overhead view shows the path traversed by the AGV 157

6.5 Use Case 6.2: Implementation of connectionistic architecture for the SURV 2AGV test-bed 160

6.6 Use Case 6.2: Merged images showing the path travelled by the target and SURV 2AGV Labelled points A to H indicate the path taken by the robot while it attempted to track and follow a moving target 163

6.7 Use Case 6.2: Changes in needs, emotions and action activations 163

6.8 Use Case 6.2: Images depicting the trajectory travelled by the SURV 2AGV 164

6.9 Use Case 6.3: System Architecture for Autonomous Surveillance Vehicle 165

6.10 Use Case 6.3: Simulated SIMTech level 2 laboratories environment 167

6.11 Use Case 6.3: Simulated surveillance vehicle operating within indoor laboratory environments 167

6.12 Use Case 6.3: Target detection and tracking based on fuzzy colour segmentation 168

6.13 Use Case 6.3-A: Images depicting the operation of the surveillance robot 173

6.14 Use Case 6.3-A: Trajectory of vehicle through the simulated environment 173

6.15 Use Case 6.3-A: Changes in needs, emotions, and how they affect activa-tions for different acactiva-tions 174

6.16 Use Case 6.3-B: Images depicting the operation of the surveillance robot 176

6.17 Use Case 6.3-B: Trajectory of vehicle through the simulated environment 176

6.18 Use Case 6.3-B: Changes in needs, emotions, and how they affect activa-tions for different acactiva-tions 177

6.19 Use Case 6.3-C: Images depicting the operation of the surveillance robot 179

6.20 Use Case 6.3-C: Trajectory of vehicle through the simulated environment 179

6.21 Use Case 6.3-C: Changes in needs, emotions, and how they affect activa-tions for different acactiva-tions 180

C.1 Overview of the feature detection system 196

C.2 The feature detection software demonstrator 197

C.3 Layered list of attributes on the same grid 199

C.4 Hierarchical world modelling 199

C.5 Camera frame and feature map 201

C.6 Results of experiment conducted to determine hue-saturation histograms for reference images and associated feature maps 205

C.7 Membership functions for hue and saturation 206

C.8 Results of feature detection experiments 208

C.9 Map of Nanyang Technological University showing routes (1) to (3) in Table C.2 211

C.10 Detection of ground plane 212

C.11 Omnidirectional Camera System 213 C.12 Omnidirectional Camera System: Discontinuity / Edge detection algorithm 215

C.13 Omnidirectional Camera System: Images illustrating the many stages of

the algorithms 216

C.14 Omnidirectional Camera System: Snapshot 1 of experiment within clut-tered indoor environment 217

C.15 Omnidirectional Camera System: Snapshot 2 of experiment within clut-tered indoor environment 218

C.16 Omnidirectional Camera System: Snapshot 3 of experiment within clut-tered indoor environment 218

C.17 The Surv2AGV experimental prototype 219

C.18 Hardware configuration for SURV 2AGV 221

C.19 Wheel odometer and encoder 221

C.20 Infrared sensors and sensor fusion board 222

C.21 Hardware architecture for SURV 2AGV 222

C.22 Virtual NIST Arenas as rendered within the USARSim Simulator 223

D.1 The Digiclops trinocular stereovision system and experimental platform 228

D.2 Use Case D.1-A: Camera images of points-of-interest obtained from trinoc-ular stereovision system 229

D.3 Use Case D.1-A: Plot of θ values at which needs fulfilment levels are maximized, over time 230

D.4 Use Case D.1-A: Variation of needs intensity and fulfilment within cam-era’s field-of-view 231

D.5 Use Case D.1-B: Camera images of points-of-interest obtained from trinoc-ular stereovision system 232

D.6 Use Case D.1-B: Plot of θ values at which needs fulfilment levels are maximized, over time 233

D.7 Use Case D.1-B: Variation of needs intensity and fulfilment within cam-era’s field-of-view 234

D.8 Use Case D.2: Front and side views of small AGV demonstrator for experiments in reactive navigation A Digiclops trinocular stereovision system is mounted atop the vehicle with a customized camera stand 237

D.9 Use Case D.2: Snapshots of the traversal of AGV through different situations 238

D.10 Use Case D.2-A: Camera images of points-of-interest obtained from trinoc-ular stereovision system 240

D.11 Use Case D.2-A: Plot of θ values at which needs fulfilment levels are

maximized, over time 241D.12 Use Case D.2-B: Camera images of points-of-interest obtained from trinoc-ular stereovision system 242

D.13 Use Case D.2-B: Plot of θ values at which needs fulfilment levels are

maximized, over time 243

D.14 Use Case D.3: Changes in needs fulfilment for all robots in a vehicle

following and platooning task 248D.15 Use Case D.3: Snapshots of the team of six robots performing vehicle-

platooning task, with full observability of all needs 249

D.16 Use Case D.3: Snapshots of the team of six robots performing platooning task, with no observability of the need for safety, nsafe 250D.17 Use Case D.3: Snapshots of the team of six robots performing vehicle-platooning task, with no observability of the need for accomplishment,

vehicle-naccomp 251

List of Tables

3.1 Maslow’s Hierarchy of Needs (extended) 653.2 An ontological comparison of information processing models 713.3 Comparison between different information processing levels 714.1 Adapting the survivability specification for autonomous systems 90

4.2 Operational requirements and the needs of an autonomous systems 93 4.3 Example of needs and their measurement 101 4.4 Core emotions in humans / animals and correlation with needs 103

4.5 Examples of arbitration schemes for selection of actions 1194.6 Comparing architectural approaches using the Survivability Framework 1204.7 Comparison of Survivability Framework and task-oriented approaches 1215.1 Tasks specifications and the needs that are implicated in an autonomouslyguided surveillance vehicle 1325.2 Determining relationships between needs and emotions 1335.3 Decomposition of needs for an autonomous vehicle 1355.4 Determining possible actions from decomposed needs 1365.5 Designing relationships between actions and the needs they fulfil 1395.6 Designing relationships between emotions and actions 1406.1 Use Case 6.1: Navigation and exploration 151

6.2 Use Case 6.1: Correlation between needs fulfilment, emotions and actions 152

6.3 Use Case 6.2: Target tracking and following 159

6.4 Use Case 6.2: Correlation between needs, emotions and actions 160

6.5 Use Case 6.3: Surveillance and intrusion detection 1666.6 Use Case 6.3: Relationships between needs-fulfilment and emotions 1716.7 Use Case 6.3: Relationships between actions and the needs they fulfil 171

6.8 Use Case 6.3: Relationships between actions and emotions 171B.1 Analytical example of survivability metrics 194C.1 Parameter values used for stereo-matching 197C.2 Summary of outdoor experiments 210C.3 Processing timings for the test cases 210C.4 Processing timings for the software demonstrator 210C.5 The SURV 2AGV: Technical specifications 220D.1 Use Case D.1: Interpreting visual information for autonomous navigation 225D.2 Use Case D.2: Reactive navigation in indoor and outdoor environments 236D.3 Use Case D.3: Vehicle following and platooning 245

List of Abbreviations

ADAS Advanced Driver Assistance System

AGV Autonomously Guided Vehicle, Autonomous Ground Vehicle

ARV Armed Robotic Vehicle

CICAS Cooperative Intersection Collision Avoidance Systems

CRASAR Centre for Robot Assisted Search And Rescue

CVIS Cooperative Vehicle-Infrastructure Systems

DARPA Defence Advance Research Projects Agency

EURON European Robotics Research Network

EOD Explosive Ordnance Disposal

FCS Future Combat Systems

GPS Global Positioning System

JRP Joint Robotics Program

MULE Multifunction Utility Logistics Equipment

SCADA Supervisory Control And Data Acquisition

UGV Unmanned Ground Vehicle

UAV Unmanned Aerial Vehicle

VII Vehicle Intelligence Infrastructure

WiMAX Worldwide Interoperability for Microwave Access

WAVE Wireless Access in Vehicular Environments

List of Symbols

Z the set of all integers

R the set of all real numbers

k x k the Euclidean norm of the vector x

|A| the cardinality of the set A.

˙x, ¨x the first, and second derivative of x

ri the ith system (e.g a robot or device)

M : A→ B a function M that maps from the domain set A to the range set B

N , Nr the set of all needs; the needs of a system r

E, Er the set of all emotions; the emotions of a system r

A, Ar the set of all actions; the actions of a system r

nr, ¯nr the needs vector of a system r, and its fulfilment

Q, QN an optimization problem; one based on needs-fulfilment

FP(n) a function that determines the priority of a need n

Cb

a a correlation matrix between elements in a and b

Pr{x} the probability of event x occurring

C HAPTER 1

Introduction

“In the fifties, it was predicted that in 5 years robots would be everywhere.

In the sixties, it was predicted that in 10 years robots would be everywhere.

In the seventies, it was predicted that in 20 years robots would be everywhere.

In the eighties, it was predicted that in 40 years robots would be everywhere ”

– Marvin Minsky, 1969 Turing Award Recipient,

Robotics and Artificial Intelligence Pioneer, MIT A.I Laboratory.

1.1 Background

Since the dawn of time, mankind has been seeking to extend his physical and intellectualcapabilities via artificial means This has evolved from the early domestication ofanimals as tools for farming and hunting, to the mechanical augmentation of man’sadvantage over other species through the generation of power and the invention oflanguage and writing With increasing understanding of nature’s laws, man hasharnessed the knowledge of science and applied it in every aspect of his life Today,technological progress in information and communication systems have resulted inescalating computational capacity and a proliferation of artificial systems with manyembedded processing capabilities, bringing closer the reality of truly intelligent andautonomous systems that may prove capable of complex tasks in real world scenariosbeyond the laboratory

Existing systems have demonstrated varying levels of autonomy, the most notablebeing the deployment of autonomously guided vehicles (AGVs) in both indoor andoutdoor environments Autonomously guided vehicles are being envisioned as a means

of extending human capabilities and providing access to environments which arehazardous or involve considerable danger (Braybrook, 2004) Applications for which

these systems have been conceived include urban search and rescue operations (USAR)(Murphy, 2004), planetary exploration (Singh et al., 2000), and security operationssuch as explosive ordnance disposal (EOD), reconnaissance, surveillance and intrusiondetection (Carrolla et al., 2005; Rose et al., 2002) While considerable advances havebeen attained, few autonomous systems are capable of operating in real or uncontrolledconditions, or when subject to unexpected situations and environments beyond thosethey were designed for (Carlson and Murphy, 2005; Willmott, 2005) For the effectivedeployment of these systems, high levels of reliability, measurable standards of safetyand the ability to withstand unanticipated situations, are necessary (Pradalier et al.,2005; Zimmer, 1996) However, such capabilities continue to be elusive, and considerabletechnological challenges remain to be addressed.

This thesis provides a framework to support the design of autonomous systems toenhance their effectiveness when deployed in complex environments Using this frame-

work, survivability is introduced as one of the necessary operational requirements of an

autonomous system, and taken into consideration as part of its system architecture and

design This Survivability Framework is formulated to incorporate the understanding of

physiological, psychological and cognitive processes underlying the ability of biologicalsystems in surviving harsh environments, especially when subject to danger

The focus in the rest of this thesis is on the description of the Survivability Framework

and its underlying justifications This framework is directed at behaviour generation

in autonomous systems that takes survivability into account As an architecturalframework, it identifies a reference architecture for integrating environment models,robot capabilities, decision making mechanisms, and robot behaviours to form a completeautonomous system The contributions of this thesis are supported by both simulationand experimental evidence to demonstrate the feasibility of the framework

This chapter begins with a definition of autonomous systems, their potential cations and research motivations Despite the strong interests in autonomous systems,many challenges exist in their deployment; these are discussed within the context ofthe operational and functional requirements of autonomous operation Understandingthese challenges lead to the formal statement of the problem The thesis statementestablishes the focus of this thesis, followed by a summary of issues addressed in thisthesis, together with its objectives An outline of the approach taken, with a list ofcontributions are stated at the end of the chapter

appli-1.2 Understanding autonomous systems

An appreciation of the nature of autonomous systems, and the challenges involved

in their design and deployment is discussed in this section First a definition ofautonomous systems is presented Potential applications and the motivations forautonomous systems are identified, to highlight the lack of convergence between theneed for autonomous systems, and the operational, technical and theoretical challenges

in designing and building such systems This sets the stage for the portrayal of theproblem statement

1.2.1 Defining autonomous systems

The Merriam-Webster Dictionary defines autonomy as the quality or state of being

qualities of freedom and independence suggest that in order to be autonomous, an entity needs to be capable of making choices and reasoning about the world in a self-directed

fashion A central aspect of autonomy is the capability to select one’s goals and themeans of realizing them (Savage, 2003) However, for an entity to logically reason about

the choices to make and the behaviour to adopt, first it needs to become self-aware, or at least, be aware of alternative courses of action An entity that selects a behaviour based

on some pre-programmed policy cannot be said to be autonomous in much the same way that the Earth, circling the sun in its orbit, cannot be considered autonomous since it

would not be able to ensure its own survival by changing its orbit to avoid a meteor withwhich it is bound for a collision course (Blackburn, 2002)

An artificial entity can be considered autonomous when it is capable of accomplishing

a task or activity in an independent manner; it must possess capabilities allowing it

to act and interact with its immediate environment However, this is possible only

if such an entity can perceive its environment with its own sensors, and understandthe relationships between features in the environment and its own context Given thisknowledge, the entity may then determine the actions to execute The whole process

is reckoned as an embodiment of machine capabilities to handle the complexities ofthe real-world – an adept description of an autonomous system Being essentially

a system built upon systems, an autonomous system is realized from the systemic

integration of sensing, perception, world models, decision making, communication, andactuation, embodying machine capabilities to handle the complexity of the real world.Autonomous systems are artificial entities that have the capacity to form and adapt

their behaviour while operating in the environment (Luck and d’Inverno, 2000; Steels,

1995) A fully autonomous system is capable of interacting independently and effectively

with its environment, via its own sensors and effectors to achieve specific tasks (Russeland Norvig, 1995) This may be achieved by designing autonomous systems to emulatecertain human cognitive abilities, such as sensing, perception, and understanding of theenvironment, for the purpose of reasoning, decision making and execution of tasks

1.2.2 Potential applications

The strong research interest in autonomous systems is driven by their potentialfor deployment in different environments, especially those which are inaccessible orhazardous to human beings Their useability as extensions of the physical capability

of human beings or as means to perform tasks while alleviating human presence fromhazardous situations has been demonstrated in several applications Consequently, thedeployment of robots beyond the laboratory is progressing towards field applications ineveryday environments, or even in unexplored environments as in the case of planetaryexplorations

Major examples of autonomous systems centre on providing mobility for aerial,ground and underwater vehicles Technologies for autonomously guided vehicles such

as unmanned aerial, ground and underwater vehicles are primarily being developedfor defence and homeland security applications (Rose et al., 2002) The potential ofthese systems in extreme conditions has been demonstrated by the various roboticsprogrammes funded by the U.S Department of Defence under the Joint RoboticsProgramme (JRP) and Future Combat System (FCS) initiative (Albus et al., 2002;Lopez, 2005; Rose et al., 2002) Plans for the deployment of systems with robot-like capabilities to form a third of all operational ground units by the year 2015 issetting the thrust for the provision of a series of enabling technologies to make this apossibility (Bertozzi et al., 2006) As part of these endeavours, the Defence AdvancedResearch Projects Agency (DARPA) organizes events to spearhead the development

of enabling technologies for autonomous systems, most notably the DARPA GrandChallenge (Seetharaman et al., 2006) Results from the two grand challenges so far

is testimonial to the degree of advancement made on perception, decision making andmobility currently achievable

(a) (b) (c)

Figure 1.1: Hardware demonstrators – past and present. (a) The Packbot ExplosivesOrdnance Disposal (EOD) robot by iRobot, which has seen deployment in Iraq andAfghanistan (b) Autonomous harvester, developed under CMU Robotics Institute’s Demeterproject (c) The Xavier office delivery robot (Simmons et al., 1997) (d) The Athena SDMplanetary rover (Biesiadecki et al., 2001) (e) NAVLAB 11, the latest in CMU NavLab’s garage

of autonomous vehicles

Similar research efforts in Europe coordinated by the European Robotics ResearchNetwork (EURON) are striving for technological development in robotics to fulfil societalneeds in the foreseeable future (Dario et al., 2004), particularly with the expectedincrease in the size of the elderly population The United Nations Economic Commissionfor Europe’s 2002 World Robotics Survey has reported the increasing demand formedical, underwater and demolition robots in the workplace, where sales of thesesystems has grown by 2.5%, with another 25,000 units projected to be added to thestock (UNECE, 2002) The UNECE 2004 World Robotics Survey has reported 600,000household robots sold or deployed in that year, with more projected to come (UNECE,2004) It is an illustration of the spin-off effects of military investment in roboticsresearch that leads to the increasing adoption of robotics technology for civil applicationsand sold as consumer products

Autonomous systems have been envisioned for various other purposes, notablyplanetary rovers for space exploration (Chatila and Lacroix, 1997), field robots formining and demining (Scheding et al., 1999), robotic ushers welcoming visitors inmuseums and robotic assistants despatching documents within an office (Simmons et al.,1997) Some examples are shown in Figure 1.1, namely (a) the Packbot ExplosivesOrdnance Disposal (EOD) robot by iRobot, which has seen operational deployment inIraq and Afghanistan, (b) an autonomous harvester, developed under CMU RoboticsInstitute’s Demeter project, (c) Xaiver, the first successful office delivery robot, (d) theAthena SDM planetary rover (Biesiadecki et al., 2001), and (e) NAVLAB 11, part of thefamily of vehicles in Carnegie Mellon University’s NavLab demonstrating the potential

of autonomous navigation in urban environments

Many technological milestones have been reached in the service, assistive andentertainment robots sector, as illustrated in Figure 1.2 These are largely dominated

by efforts on humanoid robots by major Japanese corporations Honda’s ASIMO(Figure 1.2(a)) has reached a level of near-human dexterity, whereas Hitachi’s EMIEW(Figure 1.2(b)) exemplifies an autonomous system with high degrees of mobility for safeinteraction with human crowds Sony’s QRIO (Figure 1.2(c)) is an entertainment robotthat attests to the degree of acceptability that these robotic entities may one day reach.Finally, iRobot’s Roomba vacuum cleaner robots (Figure 1.2(d)) have demonstrated thecommercial success of service robots for the masses

Robotics technologies have been progressing beyond factory floor automation Theautomotive industry has begun incorporating advance perception systems for vehicleguidance and active safety (Dickmanns, 1998, 2002; Pradalier et al., 2005), and assistiverobotic devices for the elderly and physically-challenged are now available (Arkin et al.,2003) These assistive devices are designed not just to compensate for a reduction in thephysical and cognitive capabilities of human beings, but to enhance them as well

1.2.3 Motivation for autonomous systems

The above are some instances of notable developments in autonomous systems thatsignify the growing interest in the use of these systems in many domains To obtain abetter understanding of the functional and operational requirements of these systemswhen deployed in real environments, social and technological factors driving thisinterest are identified as follows:

(a) (b) (c) (d)

Figure 1.2: Examples of service and assistive robots Service and assistive robotics are

gradually becoming a part of everyday life (a) The Honda ASIMO humanoid robot (b)Hitachi’s EMIEW service robot (c) Sony’s QRIO entertainment robot (d) iRobot Roombavacuum cleaner robot

Social drivers

Demographic changes and advances in healthcare have meant an increase in the elderlypopulation to the possible extent that, within the next two decades, the age pyramidwould likely be reversed, i.e the number of elderly people would be higher thanthose of working age, especially in developed countries As the former would havereduced physical capabilities, initiatives for using autonomous systems as healthcareassistants are gaining momentum Energy shortages and ecological awareness areplacing increasing pressures on the overall reduction of pollutants and energy usage inall aspects of human activity Consequently, automobile manufacturers are looking intothe use of vehicle guidance systems and navigation tools to optimize the use of energywhile at the same time enhancing the safety of the passengers and other road users.Autonomous systems are being envisioned to address several of these needs Intelli-gent Transportation Systems (ITS) are currently in development to reduce the number

of privately-owned vehicles on the roads, and economize on the number of passengersconveyed per unit energy expended The use of electric-powered vehicles with next-generation, green power systems (such as fuel cells) would reduce the dependency onfossil fuels With the pressures of an aging population, domestic robotic assistants maylook after the ill or elderly within the household, surgery robots may assist surgeons inperforming delicate operations, and robotic toys capable of portraying different emotionsmay fulfil psychological or companionship needs of human beings (especially those whoare childless, and the elderly) as a soft form of rehabilitation

Heightened security concerns in response to the threats posed by terrorism and fighting efforts (Carrolla et al., 2005; Singh and Thayer, 2001), together with searchand rescue operations in the aftermath of natural and man-made disasters (Murphy,2004), are areas which are driving efforts for effective solutions based on autonomoussystems For instance, since 2003, more than 330 EOD robots have been shipped to Iraqand Afghanistan (Karlin, 2007) While the potential for military applications continue

war-to be a key driver for the development of auwar-tonomous systems, robots are increasinglybeing required for performing reconnaissance and surveillance in civilian settings, forinstance, as part of rescue and reparation efforts in the aftermath of hurricane Katrina

in 2005 where unmanned aerial vehicles (UAVs) were deployed in the face of severelimitations (Carlson and Murphy, 2005; Willmott, 2005)

et al., 2000) Progress in communication systems has led to greater mobility, withthe emergence of distributed or collaborative systems where simpler entities can evolve

and interact within intelligent spaces, sharing information gathered by their distributed

sensors (Jung and Zelinsky, 1999; Kogut et al., 2003) Increasingly, computation can

be performed on-board, using lightweight computational devices (Garcia and Valavanis,2006; Valavanis et al., 2005) The miniaturization of such devices paves the way forautonomous systems with greater mobility and smaller form-factors (Barnes et al.,2005; Bruch et al., 2005) Progress in sensor systems and computer vision have led

to an increase in machine perception capability (Brown et al., 2003; Dickmanns, 1998;Sukkarieh et al., 1999; Wagner et al., 2002), resulting in their pervasive use in currentautonomous platforms (Bertozzi et al., 2006; Kelly and Stenz, 1998; Thrun et al.,2006) The combination of these technologies is enabling the deployment of autonomoussystems in environments of increasing sophistication

Command Interpretation

Behaviour-based Intelligence Autonomous Task Planning Reasonable Thinking Human-like Intelligence

Today 1-2 years

Figure 1.3: Technology roadmap for development of intelligent systems European Robotics

Research Network (EURON) GROWTH ProRobot Project (Dario et al., 2004)

Convergence of social and technological drivers

Given the social and technological drivers delineated above, there are strong needs forthe development of autonomous systems with capabilities exceeding those which arecurrently available At the same time, there is enabling technology available in the form

of sensors, processing capability, software and analysis tools to pursue the development

of autonomous systems with useful capabilities These include the ability to perceive andinteract independently with the environment, exhibit higher levels of intelligence andadaptability, deeper levels of planning and task execution, and with stronger abilities

to interact effectively with other entities within the workspace Such capabilities canpotentially be brought about by modelling human capabilities in order to impart thesefeatures to autonomous systems

Despite the alignment of social and technological drivers, further research is expectedbefore fully autonomous capabilities can be attained, as indicated in a technologyroadmap for the development of intelligent systems in Figure 1.3 (Dario et al., 2004).While striving for autonomous systems with human-like intelligence, several interme-diate milestones have to be reached Furthermore, progress towards behaviour-basedintelligence in artificial systems is expected to be achieved only in the next decade

1.3 Challenges in autonomous systems

Achieving persistent and reliable operation of autonomous systems in real ments, particularly outdoors, has been found to be difficult (Carlson and Murphy, 2005),especially with limitations in perception, scene understanding, and the complexityarising from co-existence and co-operation with other entities including human beingswithin the same workspace (Coste-Manière and Simmons, 2000) Consequently, most

environ-are designed for specific applications where their viability is validated from the success

of demonstrators completing multiple trials within a given scenario With a departurefrom “laboratory conditions”, the operating set-point changes, and systems unable toaccount for such changes would cease operating

The difficulties faced by current systems were exemplified when none of the rithms developed and demonstrated for autonomous robots in various urban searchand rescue experiments or competitions were found to be usable in an actual situation(Murphy, 2004) In the aftermath of terrorist attacks on the World Trade Centre (NewYork, USA) on September 11th2001, attempts to deploy autonomous robotics technologyfor search and rescue recovery efforts had fallen back to the use of tethered, tele-

algo-operation remotely tele-operated systems These tele-operated systems are automatic,

as opposed to autonomous The complexity resulting from operating in real-world environments imposes a high demand on such systems to progress beyond automation and move towards autonomous operation Despite considerable progress in recent years,

this vision is still distant

1.3.1 Operational and functional challenges

Autonomous systems, such as autonomously guided vehicles (AGVs), comprise multiplesensors and perception systems, various means of locomotion, navigation, control andcommunication means, and other functional components depending on the tasks they aredesigned for Operation in unstructured or semi-structured environments for prolongedperiods of time poses numerous challenges in perception, mission-planning, decisionmaking, localization and mapping, mobility, co-existence and situation-awareness Inparticular, perception often requires computationally-intensive algorithms (Wagner

et al., 2002) and the ability to accommodate uncertainty in sensor data (Leal et al.,2000) Meanwhile, decision making has to be carried out under the influence ofsuch uncertainty and is further constrained by real-time requirements, preventingsystems from planning arbitrarily-deep into the future or making use of arbitrarily-pasthistorical data In the presence of uncertainty, living creatures may improvise and deriveoriginal ways of dealing with unexpected circumstances, whereas artificial systems donot yet possess such survival characteristics Furthermore, many applications requireinteraction with other entities (e.g humans and other vehicles) within the sameenvironment Sharing of workspace presents a dynamic environment to the system,making it difficult to acquire knowledge about the world and other entities e.g their

?What am

I?

What am I here for?

What will happen to me if I stay here?

What can’t I do

?

What a

m I capable o

Figure 1.4: An illustration of the requirements faced by an autonomous system in the form

of questions that a living entity would ask itself in order to complete its tasks

beliefs, behaviours, and intentions This is compounded by the need for interactionwith human operators, users or other autonomous vehicles in a collaborative manner.Safety is also an important issue especially when sharing workspace with living humanentities (A full treatment of these challenges can be found in section 2.3)

1.3.2 Situational awareness and decision making

The linear collection of functional capabilities is insufficient for operation in an tonomous manner In addition, purely reactive algorithms are unable to fathom thecomplexity of the environment An autonomous system needs to be aware of itsenvironment in order to understand its own state within it (situational awareness).Subsequently, it needs to decide its course of action as a function of its understanding ofits situation and the tasks that it has to accomplish, which requires it to act and interactwith its environment despite the possible occurrence of mishaps or perturbations Forthis purpose, various types of information are required: for instance, its position, andorientation, the location of obstacles, and traversability of the environment Figure 1.4illustrates the type of questions that an autonomous system may probably ask if it werecapable of reasoning as a living entity These address several outstanding issues inautonomy, for instance, decision making in the presence of uncertainty, the need to modeland to represent its environment as well as its ability to survive within the environment

au-in order to perform its tasks

1.3.3 System engineering and integration

Apart from the above difficulties, there is strong suggestion that the remaining hurdle

is one of system engineering, i.e the software architectures and algorithms thatintegrate available components to form a complex system, which is a challenge in its ownright (Joshi et al., 2002) Though the state-of-the-art in unmanned systems technologyand subsystems are already meeting most of the sensing and processing requirementsfor autonomous operation (Durrant-Whyte, 2001; Rose et al., 2002), achieving autonomyentails more than the integration of component systems There is a fallacy of composition

in presuming that the success of an autonomous system can emerge solely as aconsequence of the effectiveness of its constituent subsystems

System engineering determines the success of the autonomous system and warrantsthe design of effective control-command strategies and world representation schemes

to leverage on or maximize the use of available technology to sustain the autonomousoperation and survival of a system under adverse conditions However, before anunderstanding of the basic relationships between system architectures and behaviour

is obtained, – i.e the dynamics of the system-environment interaction, – any lation about potentially useful architectures for achieving higher level goals would bepremature (Pfeifer, 1995)

specu-1.3.4 The Survivability Paradigm

To address some of the above issues, an architectural framework is established in thisthesis to support the deployment of autonomous unmanned systems in real environ-ments Theoretical justifications for the underlying principles behind this frameworkare provided Central among these principles is the concept of autonomous system

survivability, which is introduced within the context of an architectural framework The

relationship between survivability and relevant concepts in the studies of affect (Ortony

et al., 2005) and cognitive science (namely, motivation and emotion) are identified This

is supported by studies in biological systems and animal behaviour which suggest thatautonomy entails the ability of a system to ensure its survival (Dean, 1998) The study

of biological systems, such as examining animal behaviour, can provide models that can

be operationalized within a robotic system (Arkin, 1998) With this tenet, the proposedframework prompts designers to develop system architectures directed at maintaining

the survivability of unmanned systems.

1.4 Statement of the problem

Reviewing existing work and platforms reveal that achieving persistent autonomousoperation in real environments remain a difficulty, due to several outstanding issues:

a There exist difficulties in fulfilling the operational requirements of autonomoussystems (section 2.3.1), which are conditions that must be met in order forautonomous systems to be effective in performing their tasks In addition,disparate difficulties exist for each of the functional components and capabilities

of an autonomous system (section 2.3.2)

b Notwithstanding these operational and functional challenges, problems exist inthe system integration of different components to optimize the overall capability

of the autonomous system System integration necessitates an effective systemarchitecture to control and coordinate the operation of a system’s functions

c There exists outstanding computational and implementation issues (section 2.3.3),which is the case even when all functional capabilities and separate componentsare available These are hard problems that are continually being addressed,especially in artificial intelligence and cognitive science

Many autonomous systems reported so far addresses the problem of autonomy asthe integration of disparate components and modules to form a complex system, forinstance, a linear collection of sensing, perception, processing and actuating componentsupon which some algorithms are applied to control the behaviour of the system whileinteracting with the environment and achieving its task Though the contributions aretechnically sound and indeed allow hardware platforms to be built and deployed, they

do not address the well-being of the system in a manner that can ensure its survival

1.4.1 Thesis statement

The research presented in this thesis aims to contribute to the autonomous capabilities

of artificial systems via an architectural framework The autonomous capabilities ofartificial entities can be augmented by addressing their survivability as part of thesystem design process, and its decision making The fundamental premise of thisresearch is stated in the following thesis statement:

Thesis statement : The autonomous capabilities of artificial entities can be augmented

by addressing their survivability in terms of the fulfilment of the system’s needs and the use of emotions to influence the choice or sequence of actions to be executed.

1.4.2 Problem context

This thesis addresses stand-alone systems which are intended to operate autonomouslywithout human intervention i.e the manner such systems operate is that there is no

loops (e.g thermostat and temperature controllers) in that they have multiple tasksnecessitating decision making between multiple possible courses of actions The choice

of autonomous vehicles as demonstration platforms in this thesis is motivated by theready availability of tangible real-world examples and situations in which they can bedeployed In addition, the issues in the development of these systems are being studied

by the robotics community and to some extent by the author, providing opportunities tobuild upon the understanding and experience accumulated so far

1.4.3 Issues addressed

The design of autonomous systems is a complex problem entailing multiple issues whichare closely coupled, making it difficult to examine them in isolation, necessitating anexamination of the correlations and interactions between them This thesis focuses onthe generation of behaviours and decision making processes that would enhance thecapabilities of autonomous systems To this end, the issues addressed are listed below:

a Representation: of a robot’s goals, and its well-being (i.e needs and emotions), in a

form that can motivate different behaviours

b Behaviour generation: the design of a repertoire of behaviours to accomplish tasks.

c Behaviour selection: the selection of behaviours to be executed at any time, given

the repertoire of behaviours that has been defined

d Architectural framework: the design of an architectural framework to support behaviour generation and selection based on the representation of its well-being.

1.5 Research objectives

The purpose of this research is to augment the decision making capabilities of

au-tonomous systems with the survivability paradigm To achieve this, the following

objectives are set:

a To incorporate the notions of survivability into the underlying architecture ofautonomous systems in a manner that can be included in the decision-processes ofsuch systems;