Modular transfer reinforcement learning in industrial robotics / von Christian Bitter (geb. Scheiderer). Wuppertal, [2025]
Inhalt
- Abstract
- Abbreviations
- Contents
- List of Figures
- List of Tables
- 1 Introduction
- 1.1 Motivation
- 1.2 Research Questions
- 1.3 Use Cases
- 1.3.1 Use Case 1: Wire-Loop Game
- 1.3.2 Use Case 2: Object Picking
- 1.3.3 Use Case 3: Clip Assembly in Aircraft Manufacturing
- 1.4 Structure
- 2 Foundations
- 2.1 Industrial Robotics
- 2.2 Deep Learning
- 2.3 Deep Reinforcement Learning
- 2.3.1 Reinforcement Learning Fundamentals
- 2.3.2 Deep Reinforcement Learning Algorithms for Continuous State-Action Spaces
- 2.3.3 Hierarchical Reinforcement Learning
- 2.3.4 Environment Parallelization
- 2.4 Transfer Reinforcement Learning
- 3 Related Work
- 3.1 Real-Time Asynchronous Reinforcement Learning
- 3.2 Sim2Real Transfer with <DR>
- 3.3 Robotic Movement Structure Exploitation
- 3.4 Cross-Robot Transfer Reinforcement Learning
- 3.5 Research Gaps and Focus
- 4 Framework and Baselines
- 4.1 Learning Framework
- 4.2 Use Case 1: Wire-Loop Game
- 4.3 Use Case 2: Object Picking
- 4.4 Use Case 3: Clip Assembly
- 4.5 Mapping Research Questions to Use Cases
- 5 Smooth Continuous Robot Control
- 5.1 Smoothness-Constrained Action Space Design
- 5.2 Asynchronous Learning Framework
- 5.3 Dynamics Rollout Module Development and Validation
- 5.4 Exploration of Design Choices
- 5.5 Asynchronous Wire-Loop Experiments
- 5.6 Summary
- 6 Sim2Real Transfer of Perception Modules
- 6.1 Pose Estimation Modules
- 6.2 Image Compression Modules
- 6.2.1 Unsupervised Image Compression
- 6.2.2 Explorative Experiments on Agent Attention
- 6.2.3 Semi-Supervised Image Compression
- 6.2.4 Semi-Supervised Compression with Domain Knowledge
- 6.3 Summary
- 7 Hierarchical and Backward Planning
- 7.1 Hierarchical Policy Transfer
- 7.1.1 Hierarchical Actor-Critic Learning Framework
- 7.1.2 Fetch Environment Variations
- 7.1.3 Experimental Evaluation
- 7.2 Assembly-by-Disassembly
- 7.3 Summary
- 8 Cross-Robot Execution Imitation
- 8.1 Robot-Task Environments
- 8.2 Cross-Robot Imitation Learning Framework
- 8.2.1 Cross-Robot Behavior Similarity
- 8.2.2 Cross-Robot Trajectory Mapping
- 8.2.3 Cross-Robot Imitation Learning
- 8.3 Experimental Evaluation
- 8.4 Summary
- 9 Critical Reflection and Outlook
- 9.1 Research Question 1
- 9.2 Research Question 2
- 9.3 Research Question 3
- 9.4 Research Question 4
- 9.5 Closing Remarks
- Bibliography