Handling Uncertainty and Networked Structure in Robot Control (eBook)

Lucian Busoniu received the M.Sc. degree (valedictorian) from the Technical University of Cluj-Napoca, Romania, in 2003 and the Ph.D. degree (cum laude) from the Delft University of Technology, the Netherlands, in 2009. He has held research positions in the Netherlands and France, and is currently an associate professor with the Department of Automation at the Technical University of Cluj-Napoca. His fundamental interests include planning-based methods for nonlinear optimal control, reinforcement learning and dynamic programming with function approximation, and multiagent systems; while his practical focus is applying these techniques to robotics. He has coauthored a book and more than 50 papers and book chapters on these topics. He was the recipient of the 2009 Andrew P. Sage Award for the best paper in the IEEE Transactions on Systems, Man, and Cybernetics.  Levente Tamas received the M.Sc. (valedictorian) and the Ph.D. degree in electrical engineering from Technical University of Cluj-Napoca, Romania, in 2005 and 2010, respectively. He took part in several postdoctoral programs dealing with 3D perception and robotics, the most recent one spent at the Bern University of Applied Sciences, Switzerland. He is currently with the Department of Automation, Technical University of Cluj-Napoca, Romania. His research focuses on 3D perception and planning for autonomous mobile robots, and has resulted in several well ranked conference papers, journal articles, and book chapters in this field.

Contents 6
Contributors 14
Acronyms 17
Introduction 20
Part I Learning Control in Unknown Environments 28
1 Robot Learning for Persistent Autonomy 29
1.1 Persistent Autonomy 29
1.2 Robot Learning Architecture 30
1.3 Learning of Reactive Behavior 31
1.3.1 Autonomous Robotic Valve Turning 32
1.3.2 Related Work 32
1.3.3 Hierarchical Learning Architecture 34
1.3.4 Learning Methodology 34
1.3.5 Imitation Learning 36
1.3.6 Force/Motion Control Strategy 38
1.3.7 Learning of Reactive Behavior Using RFDM 39
1.3.8 Iterative Learning Control 42
1.4 Learning to Recover from Failures 43
1.4.1 Methodology 44
1.4.2 Fault Detection Module 45
1.4.3 Problem Formulation 45
1.4.4 Learning Methodology 46
1.4.5 Experiments 48
1.5 Conclusion 52
References 52
2 The Explore--Exploit Dilemma in Nonstationary Decision Making under Uncertainty 55
2.1 Introduction 55
2.1.1 Decision Making and Control under Uncertainty in Nonstationary Environments 58
2.2 Case Study 1: Model-Based Reinforcement Learning for Nonstationary Environments 59
2.2.1 Gaussian Process Regression and Clustering 60
2.2.2 GP-NBC-MBRL Solution to Nonstationary MDPs 63
2.2.3 Example Experiment 63
2.2.4 Summary of Case Study 1 65
2.3 Case Study 2: Monitoring Spatiotemporally Evolving Processes using Unattended Ground Sensors and Data-Ferrying UAS 66
2.3.1 Connecting the FoW Functional to Data 67
2.3.2 Problem Definition 68
2.3.3 Solution Methods 70
2.3.4 Simulation Results 73
2.3.5 Summary of Case Study 2 76
References 76
3 Learning Complex Behaviors via Sequential Composition and Passivity-Based Control 79
3.1 Introduction 80
3.2 Sequential Composition 82
3.3 Passivity-Based Control 83
3.3.1 Interconnection and Damping Assignment Passivity-Based Control 84
3.3.2 Algebraic IDA-PBC 85
3.4 Estimating the Domain of Attraction 86
3.5 Learning via Composition 89
3.5.1 Actor--Critic 89
3.5.2 Algebraic Interconnection and Damping Assignment Actor--Critic 90
3.5.3 Sequential Composition Reinforcement Learning Algorithm 90
3.6 An Example Simulation 93
3.7 Conclusions 98
References 99
4 Visuospatial Skill Learning 101
4.1 Introduction 101
4.2 Related Work 103
4.3 Introduction to Visuospatial Skill Learning 106
4.3.1 Terminology 107
4.3.2 Problem Statement 107
4.3.3 Methodology 108
4.4 Implementation of VSL 109
4.4.1 Coordinate Transformation 110
4.4.2 Image Processing 110
4.4.3 Trajectory Generation 112
4.4.4 Grasp Synthesis 114
4.5 Experimental Results 114
4.5.1 Simulated Experiments 114
4.5.2 Real-World Experiments 117
4.6 Conclusions 123
References 123
Part II Dealing with Sensing Uncertainty 126
5 Observer Design for Robotic Systems via Takagi--Sugeno Models and Linear Matrix Inequalities 127
5.1 Introduction 127
5.2 Preliminaries 129
5.2.1 Descriptor Models of Robotic Systems 129
5.2.2 Takagi--Sugeno Models 131
5.2.3 Linear Matrix Inequalities 135
5.3 Observer Design for TS Descriptor Models 137
5.4 Simulation Example 146
5.5 Summary 150
References 151
6 Homography Estimation Between Omnidirectional Cameras Without Point Correspondences 153
6.1 Introduction 153
6.2 Planar Homography for Central Omnidirectional Cameras 155
6.3 Homography Estimation 157
6.3.1 Construction of a System of Equations 158
6.3.2 Normalization and Initialization 159
6.4 Omnidirectional Camera Models 160
6.4.1 The General Catadioptric Camera Model 160
6.4.2 Scaramuzza's Omnidirectional Camera Model 162
6.5 Experimental Results 163
6.6 Relative Pose from Homography 167
6.7 Conclusions 173
References 173
7 Dynamic 3D Environment Perception and Reconstruction Using a Mobile Rotating Multi-beam Lidar Scanner 176
7.1 Introduction 177
7.2 3D People Surveillance 179
7.2.1 Foreground-Background Separation 180
7.2.2 Pedestrian Detection and Multi-target Tracking 181
7.2.3 Evaluation 183
7.3 Real Time Vehicle Detection for Autonomous Cars 185
7.3.1 Object Extraction by Point Cloud Segmentation 187
7.3.2 Object Level Feature Extraction and Vehicle Recognition 189
7.3.3 Evaluation of Real-Time Vehicle Detection 192
7.4 Large Scale Urban Scene Analysis and Reconstruction 193
7.4.1 Multiframe Point Cloud Processing Framework 194
7.4.2 Experiments 199
7.5 Conclusion 201
References 202
8 RoboSherlock: Unstructured Information Processing Framework for Robotic Perception 204
8.1 Introduction 205
8.2 Related Work and Motivation 207
8.3 Overview of RoboSherlock 208
8.4 Conceptual Framework 211
8.4.1 Common Analysis Structure (CAS) 211
8.4.2 Analysis Engines in RoboSherlock 212
8.4.3 Object Perception Type System 215
8.4.4 Integrating Perception Capabilities into RoboSherlock 216
8.5 Tracking and Entity Resolution 217
8.6 Information Fusion 219
8.7 Experiments and Results 222
8.7.1 Illustrative Example 223
8.7.2 Entity Resolution 224
8.7.3 Information Fusion 226
8.8 Conclusion and Future Work 229
References 229
9 Navigation Under Uncertainty Based on Active SLAM Concepts 232
9.1 Introduction 232
9.1.1 SLAM 234
9.1.2 Active Mapping 234
9.1.3 Active Localization 234
9.1.4 Active SLAM 235
9.2 High Level View of General Active SLAM Algorithms 236
9.3 Uncertainty Criteria 237
9.4 Main Paradigms of Active SLAM 239
9.4.1 A First Approach: Local Search Using Optimality Criteria 239
9.4.2 A Second Look: An Information Gain Approach 241
9.4.3 A Third Strategy: Considering Multiple Steps Ahead 243
9.5 Navigation Under Uncertainty: An Active SLAM Related Application 243
9.5.1 Path Planning in the Belief Space 244
9.6 Our Approach: Fast Minimum Uncertainty Search Over a Pose Graph Representation 245
9.6.1 Metric Calculation 246
9.6.2 Increasing Traversability 246
9.6.3 Decision Points 246
9.6.4 Decision Graph 247
9.6.5 Searching over the Decision Graph 248
9.7 Experiments 249
9.7.1 Graph Reduction 250
9.7.2 H0: Are the Minimum Uncertainty Path and the Shortest Necessarily Equal? 250
9.7.3 Timing Comparisons 251
9.8 Discussion 252
References 254
10 Interactive Segmentation of Textured and Textureless Objects 259
10.1 Introduction and Motivation 260
10.2 Overview of Interactive Segmentation Processing Steps 263
10.3 Segmentation of Cluttered Tabletop Scene 263
10.4 Push Point Selection and Validation 264
10.4.1 Contact Points from Concave Corners 265
10.4.2 Push Direction and Execution 265
10.5 Feature Extraction and Tracking 266
10.6 Feature Trajectory Clustering 267
10.6.1 Randomized Feature Trajectory Clustering 268
10.6.2 Trajectory Clustering Analysis 271
10.6.3 Exhaustive Graph-Based Trajectory Clustering 273
10.7 Stopping Criteria and Finalizing Object Models 274
10.7.1 Verification of Correctness of Segmentation 275
10.7.2 Dense Model Reconstruction 276
10.8 Results 278
10.8.1 Random Versus Corner-Based Pushing 278
10.8.2 Trajectory Clustering 279
10.8.3 System Integration and Validation 280
10.9 Conclusions 282
References 282
Part III Control of Networked and Interconnected Robots 285
11 Vision-Based Quadcopter Navigation in Structured Environments 286
11.1 Introduction 287
11.2 Quadcopter Structure and Control 288
11.3 Quadcopter Hardware and Software 289
11.4 Methodological and Theoretical Background 290
11.4.1 Feature Detection 291
11.4.2 Feature Tracking 293
11.5 Approach 299
11.5.1 Software Architecture 299
11.5.2 Quadcopter Initialization 301
11.5.3 Perspective Vision 301
11.5.4 VP Tracking 303
11.5.5 Control 303
11.6 Experiments and Results 305
11.6.1 VP Motion Model Results 306
11.6.2 Nonlinear Estimator Results 306
11.6.3 Indoor and Outdoor Results 307
11.6.4 Control Results 308
11.7 Summary and Perspectives 310
References 310
12 Bilateral Teleoperation in the Presence of Jitter: Communication Performance Evaluation and Control 312
12.1 Introduction 312
12.2 Communication Performance Evaluation for Wireless Teleoperation 314
12.2.1 The Wireless Communication Medium 314
12.2.2 Implementation of Application Layer Measurements 316
12.2.3 Experimental Measurements 317
12.3 Control of Bilateral Teleoperation Systems in the Presence of Jitter 323
12.3.1 Control Approaches to Assure the Stability in Bilateral Teleoperation Systems 323
12.3.2 Bilateral Control Scheme to Deal with Jitter Effects 326
12.3.3 Control Experiments 327
12.4 Conclusions 330
References 331
13 Decentralized Formation Control in Fleets of Nonholonomic Robots with a Clustered Pattern 333
13.1 Introduction 334
13.2 Problem Formulation and Preliminaries 336
13.2.1 Robot Dynamics and Tracking Error 337
13.2.2 Network Topology and Agreement Dynamics 339
13.3 Solving the Consensus and Tracking Problems 341
13.3.1 Linear Consensus for Networks with a Cluster Pattern 341
13.3.2 Tracking for Nonholonomic Systems 344
13.4 Overall Controller Design 345
13.5 Simulation Results 348
13.5.1 Small-Scale Example: Ellipse Formation 348
13.5.2 Larger-Scale Example: Three-Leaf Clover Formation 349
13.6 Conclusions and Perspectives 351
References 351
14 Hybrid Consensus-Based Formation Control of Nonholonomic Mobile Robots 354
14.1 Introduction 354
14.2 Background on Hybrid Automata 357
14.3 Hybrid Consensus-Based Formation Control of Holonomic Robots 359
14.3.1 Regulation Controller Design 360
14.3.2 Consensus-Based Formation Controller Design 361
14.3.3 Hybrid Consensus-Based Regulation and Formation Controller Design 363
14.4 Hybrid Consensus-Based Formation Control of Non-holonomic Robots 364
14.4.1 Nonholonomic Mobile Robot Equations of Motion 365
14.4.2 Regulation Controller of Mobile Robots 366
14.4.3 Consensus-Based Formation Control of Nonholonomic Mobile Robots 368
14.4.4 Hybrid Consensus-Based Formation Control 371
14.5 Simulation Results 373
14.5.1 Omnidirectional Robots 374
14.5.2 Nonholonomic Mobile Robots 376
14.6 Conclusions and Future Work 378
References 379
15 A Multi Agent System for Precision Agriculture 380
15.1 Introduction 381
15.2 General Architecture 382
15.3 Methodology 384
15.3.1 Model Identification 384
15.3.2 Low-Level PID Cascade Control 388
15.3.3 High-Level Model-Based Predictive Control 392
15.4 Experimental Results 397
15.4.1 Formation Control of UGVs 398
15.4.2 Path Following for the Quadrotor 399
15.4.3 Quadrotor as Flying Sensor for Ground Agents 401
15.5 Conclusions 403
References 404
Index 406