Hierarchical Planning with Large Models for Robot Learning

Explore the fundamentals and advanced applications of hierarchical planning in robot learning in this 1 hour 51 minute lecture from Montreal Robotics. Gain insights into how hierarchical planning methods are implemented in cutting-edge research papers like Hi-robot and Gemini-Robotics. The lecture begins with essential background on hierarchical planning and reinforcement learning, using humanoid control as a practical case study, before delving into more sophisticated concepts. Learn about the exploration benefits of hierarchical approaches, effective problem decomposition strategies, methods for credit assignment, the challenges involved in developing robust low-level policies, and techniques for co-training policies. This comprehensive overview connects theoretical foundations with current research applications in the rapidly evolving field of robot learning.