This was part of New Directions in Reinforcement Learning and Control

From Generative Models to Control: Representation-based Reinforcement Learning in Physical Systems

Na Li, Harvard