A talk by Marc Deisenroth on improving data efficiency in reinforcement learning for autonomous robots, covering probabilistic models, control constraints, and meta learning techniques.