Zili Zou

Ditch the human supervisors: a multimodal agent can guide robotic reinforcement learning with corrective waypoints and spatial constraints, boosting sample efficiency beyond human-in-the-loop methods.