Robot Learning by Visual Observation [Hardcover]

This book presents programming by demonstration for robot learning from observations with a focus on the trajectory level of task abstraction

• Discusses methods for optimization of task reproduction, such as reformulation of task planning as a constrained optimization problem
• Focuses on regression approaches, such as Gaussian mixture regression, spline regression, and locally weighted regression
• Concentrates on the use of vision sensors for capturing motions and actions during task demonstration by a human task expert

Preface xi

List of Abbreviations xv

1 Introduction 1

1.1 Robot Programming Methods 2

1.2 Programming by Demonstration 3

1.3 Historical Overview of Robot PbD 4

1.4 PbD System Architecture 6

1.4.1 Learning Interfaces 8

1.4.1.1 Sensor-Based Techniques 10

1.4.2 Task Representation and Modeling 13

1.4.2.1 Symbolic Level 14

1.4.2.2 Trajectory Level 16

1.4.3 Task Analysis and Planning 18

1.4.3.1 Symbolic Level 18

1.4.3.2 Trajectory Level 19

1.4.4 Program Generation and Task Execution 20

1.5 Applications 21

1.6 Research Challenges 25

1.6.1 Extracting the Teacher’s Intention from Observations 26

1.6.2 Robust Learning from Observations 27

1.6.2.1 Robust Encoding of Demonstrated Motions 27

1.6.2.2 Robust Reproduction of PbD Plans 29

1.6.3 Metrics for Evaluation of Learned Skills 29

1.6.4 Correspondence Problem 30

1.6.5 Role of the Teacl“%