In this episode, I sit down with guests Eric Jang and Pete Florence, two research scientists at Google working on deep learning and robotic manipulation. We discuss progress in robotics in the last 50 years, and the roles software and hardware have played in this development. On the algorithmic side, we discuss how much of imitation learning and reinforcement learning is needed to obtain general purpose robots and why policy evaluation is hard in the real-world. Finally, Eric and Pete tell us what they look forward the most to in robotics in the next 50 years.