
AI and Robotics: A Conversation with Chelsea Finn
Stanford professor Andrew Ng talks with fellow Stanford roboticist Chelsea Finn about what is and is not currently achievable in robot learning. Finn traces her path into the field before walking through the core technical problems: making robots robust to unpredictable environments, using reinforcement learning to let robots learn from trial and error the way game-playing agents do, and training in simulation before transferring skills to physical hardware. She describes day-to-day life running a robotics lab and what end-to-end learning systems look like when reinforcement learning and perception are combined directly. The conversation closes with her outlook on where robotics research is headed and concrete advice for students who want to enter the field. It plays as an informal but substantive exchange between two working researchers rather than a scripted talk, with real detail on current method limitations.