Safe, Autonomous And Intelligent Vehicles (unmanned System Technologies)

This book covers the start-of-the-art research and development for the emerging area of autonomous and intelligent systems. In particular, the authors emphasize design and validation methodologies to address the grand challenges related to safety. This book offers a holistic view of a broad range of technical aspects (including perception, localization and navigation, motion control, etc.) and application domains (including automobile, aerospace, etc.), presents major challenges and discusses possible solutions. Provides a single-source guide to the practical challenges in designing autonomous and intelligent systems; Discusses the major challenges related to safety of next-generation autonomous and intelligent systems, given growing complexity and new applications; Describes new design and validation methodologies to address safety issues; Includes technical background to facilitate further research and development Preface 6 Contents 8 About the Editors 10 1 Introduction 15 2 Efficient Statistical Validation of Autonomous Driving Systems 19 2.1 Introduction 19 2.2 Background 21 2.2.1 Image Sensing 21 2.2.2 Image Processing 22 2.2.3 Visual Perception 24 2.3 Test Data Generation 25 2.3.1 Temperature Variation 25 2.3.2 Circuit Aging 28 2.3.3 Corner Case Generation 30 2.3.4 Numerical Experiments 34 2.3.4.1 Experimental Setup 34 2.3.4.2 Temperature Variation 35 2.3.4.3 Circuit Aging 36 2.4 Subset Simulation 37 2.4.1 Mathematical Formulation 37 2.4.2 Random Sampling 38 2.4.3 Summary 42 2.4.4 Numerical Experiments 42 2.4.4.1 Experimental Setup 42 2.4.4.2 Experimental Results 43 2.5 Conclusions 44 References 45 3 Cyberattack-Resilient Hybrid Controller Design with Application to UAS 47 3.1 Introduction 47 3.2 Problem Formulation 49 3.2.1 System and Cyberattack Models 49 3.2.2 Cyberattack Mitigation Problem 51 3.3 Hybrid Controller Design 52 3.4 Analytical Performance Verification 57 3.5 Extension to Infinite Time Horizon: Receding Horizon Controller 59 3.6 Illustrative Example 63 3.6.1 H2 Optimal Controller 63 3.6.2 H∞ Optimal Controller 63 3.6.3 UAS Model 64 3.6.4 Simulation Results 65 3.7 Conclusions 68 References 68 4 Control and Safety of Autonomous Vehicles with Learning-Enabled Components 71 4.1 Hamilton–Jacobi Reachability 72 4.1.1 Backward Reachable Set (BRS) 73 4.1.2 Application: Provably Safe Multi-Vehicle Trajectory Planning 76 4.1.3 Limitations of HJ Reachability 78 4.2 Learning-Based Model Refinement 79 4.2.1 Function Approximator-Based Model Learning 80 4.2.2 Goal-Driven Model Learning 81 4.3 Safety Analysis of Learned Models 82 4.3.1 Safety During Model Learning 83 4.3.2 Model Validation Before Deployment 83 4.4 Learning in Partially Observable environments 84 References 86 5 Adaptive Stress Testing of Safety-Critical Systems 90 5.1 Introduction 90 5.2 Related Work 91 5.3 Background 92 5.3.1 Definitions 92 5.3.2 Sequential Decision Process 93 5.3.3 Monte Carlo Tree Search 94 5.4 Adaptive Stress Testing 94 5.4.1 Full Observability 95 5.4.2 Partial Observability 96 5.5 Aircraft Collision Avoidance Application 101 5.5.1 Experimental Setup 102 5.5.2 Results 104 5.5.3 Performance Comparison 106 5.6 Conclusion 107 References 108 6 Provably-Correct Compositional Synthesis of VehicleSafety Systems 109 6.1 Introduction 109 6.2 Autonomous Driving Functions 110 6.2.1 Adaptive Cruise Control 111 6.2.2 Lane Keeping 111 6.2.3 Challenges in Composition 112 6.3 Composition of Invariant Sets Via Contracts 112 6.3.1 Contract Realizability Problem 116 6.3.2 Contract Refinement Heuristic 117 6.4 Contract Realizability Via Polyhedral Controlled-Invariant Sets 117 6.4.1 Computation of Polyhedral Controlled-Invariant Sets 118 6.4.2 Over-Approximation of Nonlinear Parametrizations 120 6.4.3 Removal of Nonlinearities via Convexification 121 6.4.3.1 Convex-Hull Computation With Monotone Functions 123 6.4.3.2 Convex-Hull Computation With Convex Projections 124 6.5 Design Flow for the Case Study 125 6.5.1 Constraints 126 6.5.2 Contracts 126 6.5.3 Realizability of LK Contract 127 6.5.4 Realizability of ACC Contract 128 6.5.5 Low-Fidelity Simulation Results 129 6.5.6 CarSim Simulation Results 129 6.6 Implementation in Mcity 130 6.7 Conclusions 132 References 132 7 Reachable Set Estimation and Verification for Neural Network Models of Nonlinear Dynamic Systems 135 7.1 Introduction 135 7.2 Neural Network Models of Nonlinear Dynamic Systems 137 7.3 Problem Formulation 139 7.4 Reachable Set Estimation for MLPs 140 7.5 Reachable Set Estimation for NARMA Models 144 7.6 Magnetic Levitation Systems (Maglev) 148 7.6.1 Brief Introduction 148 7.6.2 Neural Network Model 150 7.6.3 Reachable Set Estimation 151 7.7 Conclusions 153 References 154 8 Adaptation of Human Licensing Examinations to the Certification of Autonomous Systems 157 8.1 Introduction 157 8.1.1 Driving Licensing Exams 158 8.1.2 Aviation Licensing Exams 160 8.2 SRKE Taxonomy 163 8.3 Implications of Human Licensing on Autonomous Vehicle Certification 166 8.3.1 Vision Tests for AVs? 166 8.3.2 Knowledge Tests and Checkrides for AVs? 167 8.3.3 Graduated Licensing 168 8.3.4 Certifying Machine Learning Algorithms Is Unprecedented 169 8.4 Conclusion 171 References 173 9 Model-Based Software Synthesis for Safety-Critical Cyber-Physical Systems 175 9.1 Software Challenges in Safety-Critical Cyber-Physical Systems 175 9.2 Model-Based Software Synthesis Flow 178 9.3 Holistic Timing-Driven Synthesis 181 9.4 Multi-Objective Optimization 185 9.4.1 Fault-Tolerance 185 9.4.2 Security 190 9.5 Cross-Layer Codesign 193 9.6 Conclusion 194 References 195 10 Compositional Verification for Autonomous Systems with Deep Learning Components 199 10.1 Introduction 199 10.2 Compositional Verification 200 10.3 Analysis for Deep Neural Network Components 202 10.3.1 ACAS Xu Case Study 204 10.4 Example 205 10.4.1 Run-Time Monitoring and Control 207 10.5 Conclusion 207 References 208 Index 210 Front Matter ....Pages i-xiii Introduction (Huafeng Yu, Xin Li, Richard M. Murray, S. Ramesh, Claire J. Tomlin)....Pages 1-4 Efficient Statistical Validation of Autonomous Driving Systems (Handi Yu, Weijing Shi, Mohamed Baker Alawieh, Changhao Yan, Xuan Zeng, Xin Li et al.)....Pages 5-32 Cyberattack-Resilient Hybrid Controller Design with Application to UAS (Cheolhyeon Kwon, Inseok Hwang)....Pages 33-56 Control and Safety of Autonomous Vehicles with Learning-Enabled Components (Somil Bansal, Claire J. Tomlin)....Pages 57-75 Adaptive Stress Testing of Safety-Critical Systems (Ritchie Lee, Ole J. Mengshoel, Mykel J. Kochenderfer)....Pages 77-95 Provably-Correct Compositional Synthesis of Vehicle Safety Systems (Petter Nilsson, Necmiye Ozay)....Pages 97-122 Reachable Set Estimation and Verification for Neural Network Models of Nonlinear Dynamic Systems (Weiming Xiang, Diego Manzanas Lopez, Patrick Musau, Taylor T. Johnson)....Pages 123-144 Adaptation of Human Licensing Examinations to the Certification of Autonomous Systems (M. L. Cummings)....Pages 145-162 Model-Based Software Synthesis for Safety-Critical Cyber-Physical Systems (Bowen Zheng, Hengyi Liang, Zhilu Wang, Qi Zhu)....Pages 163-186 Compositional Verification for Autonomous Systems with Deep Learning Components (Corina S. Păsăreanu, Divya Gopinath, Huafeng Yu)....Pages 187-197 Back Matter ....Pages 199-204