This book provides a concise survey and description of recent collision detection methods for robot manipulators. Beginning with a review of robot kinodynamic models and preliminaries on basic statistical learning methods, the book covers fundamental aspects of the collision detection problem, from collision types and collision detection performance criteria to model-free versus model-based methods, and the more recent data-driven learning-based approaches to collision detection. Special effort has been given to describing and evaluating existing methods with a unified set of notation, systematically categorizing these methods according to a basic set of criteria, and summarizing the advantages and disadvantages of each method. This book is the first to comprehensively organize the growing body of learning-based collision detection methods, ranging from basic supervised learning methods to more advanced approaches based on unsupervised learning and transfer learning techniques. Step-by-step implementation details and pseudocode descriptions are provided for key algorithms. Collision detection performance is measured with respect to both conventional criteria such as detection delay and the number of false alarms, as well as criteria that measure generalization capability for learning-based methods. Whether it be for research or commercial applications, in settings ranging from industrial factories to physical human–robot interaction experiments, this book can help the reader choose and successfully implement the most appropriate detection method that suits their robot system and application.
Author(s): Kyu Min Park, Frank C. Park
Series: Springer Tracts in Advanced Robotics, 155
Publisher: Springer
Year: 2023
Language: English
Pages: 132
City: Cham
Series Editor’s Foreword
Preface
Contents
Acronyms
List of Figures
List of Tables
1 Introduction
1.1 Collaborative Robot Manipulators and Safety
1.2 Human-Robot Collision Handling
1.3 Collision Detection for Robot Manipulators
1.4 Objectives
1.5 Outline
References
2 Fundamentals
2.1 Introduction
2.2 Robot Dynamics
2.2.1 Rigid Joint Robots
2.2.2 Flexible Joint Robots
2.3 Fundamentals of Collision Detection
2.3.1 Collision Types
2.3.2 Detection Sensitivity Adjustment
2.3.3 Collision Detection Performance Criteria
2.4 Related Learning Algorithms
2.4.1 Support Vector Machine Regression
2.4.2 One-Dimensional Convolutional Neural Network
2.4.3 Unsupervised Anomaly Detection
2.4.4 One-Class Support Vector Machine
2.4.5 Autoencoder-Based Anomaly Detection
References
3 Model-Free and Model-Based Methods
3.1 Introduction
3.2 Model-Free Methods
3.3 Basic Model-Based Methods
3.3.1 Energy Observer-Based Estimation of External Power
3.3.2 Direct Estimation of External Joint Torque
3.3.3 Velocity Observer-Based Estimation of External Joint Torque
3.3.4 Momentum Observer-Based Estimation of External Joint Torque
3.4 Dealing with Frictional Torque in the Joints
3.4.1 Friction Modeling and Identification
3.4.2 Learning Friction
3.4.3 Using Additional Sensors
3.5 Collision Detection Under Cyclic Motions
3.6 Alternative Model-Based Methods
3.7 Multiple Arms and End-Effector Types
3.8 Summary
References
4 Learning Robot Collisions
4.1 Introduction
4.2 Supervised Learning-Based Approaches
4.2.1 Training With Only Collision-Free Data
4.2.2 Training with both Collision and Collision-Free Data
4.2.3 Summary of the Supervised Learning-Based Detection Methods
4.3 Robot Collision Data
4.3.1 True Collision Index Labeling
4.3.2 Collision Scenarios
4.3.3 Monitoring Signal
4.3.4 Signal Normalization and Sampling
4.3.5 Test Data for Detection Performance Verification
4.4 Collision Detection Using SVMR and 1-D CNN
4.4.1 SVMR-Based Collision Detection
4.4.2 1-D CNN-Based Collision Detection
4.4.3 Collision Detection Performance Analysis
4.4.4 Generalization Capability Analysis
4.4.5 Discussion
References
5 Enhancing Collision Learning Practicality
5.1 Introduction
5.2 Learning Robot Collisions Without Actual Collisions
5.2.1 Using Unsupervised Anomaly Detection
5.2.2 Using Transfer Learning
5.3 Classification of Learning-Based Detection Methods
5.4 Collision Detection Using OC-SVM and Autoencoder
5.4.1 OC-SVM-Based Collision Detection
5.4.2 Autoencoder-Based Collision Detection
5.4.3 Collision Detection Performance Analysis
5.4.4 Generalization Capability Analysis
5.4.5 Discussion
References
6 Conclusion
6.1 Summary and Discussion
6.2 Future Work
Appendix A Collision Classification
A.1 SVM-Based Classification of Detected Collisions
Appendix B Alternative Supervised Learning-Based Detection Methods
B.1 Direct Estimation-Based Detection Methods
B.2 Model-Free Detection Methods
Appendix C Additional Generalization Capability Analysis
C.1 Generalization to Large Changes in the Dynamics Model
