Feedback Design Principles for Efficient and Reliable Robot Learning
Explore a cutting-edge lecture from the 2023 Summer Robotics Colloquium featuring Tyler Westenbroek from UT Austin. Delve into the principles of feedback design for efficient and reliable robot learning as Westenbroek discusses the fusion of classical control techniques with modern AI and machine learning approaches. Discover how embedding feedback control design into reinforcement learning setups can leverage known structures in approximate dynamics models while maintaining flexibility to learn from unmodeled dynamics. Learn about principled reward shaping approaches, co-designing feedback controllers with policy gradient algorithms, and how these solutions lead to inherent robustness guarantees while significantly reducing the amount of real-world data required. Gain insights into new directions incorporating perception, human-robot interaction, and safety analysis in the field of robotics.