Lecture
This lecture by the instructor focuses on the boundary between hard coding and learning in robotics, emphasizing the importance of co-designing robotic hands and manipulation approaches to exploit environmental constraints. Through examples and experiments, the lecture showcases the development of a more dexterous robotic hand, optimized for in-hand manipulation by exploiting constraints from the environment. The lecture also delves into the concept of funnels as a low-dimensional representation for machine learning in manipulation tasks, highlighting the need to understand the interaction between the hand, environment, and object. By demonstrating the sequencing of funnels to exploit constraints, the lecture argues that manipulation, grasping, and in-hand manipulation are interconnected processes that rely on the design of the hand and the environment.