Learning Reward Models and Using Foundational Models in Robot Learning

Watch a comprehensive lecture on reinforcement learning reward functions where Glen Berseth delves into the distinctions between extrinsic and intrinsic rewards in robotics. Explore the complexities of training reward functions in real-world scenarios, illustrated through practical examples like teaching an agent to cook in a cluttered kitchen environment. Learn about innovative concepts like "nexting" and the "unreal agent" approach to developing useful policy representations. Discover how foundational models are being integrated into reward systems, both as tools for data augmentation and dataset labeling, as well as their potential direct application as reward functions. Gain insights into the challenges and opportunities of implementing these advanced approaches in robotic learning systems.