Sertac Karaman: Robots That Fly and Robots That Drive | Lex Fridman Podcast #97

Sertac Karaman discusses the fascinating world of autonomous robotics, covering both flying and driving systems. The conversation begins with a comparison of the technical challenges between aerial and ground-based robots. While flying robots require sophisticated control systems to maintain stability in three-dimensional space, autonomous vehicles face equally complex but different problems related to safety, prediction, and interaction with unpredictable human drivers and pedestrians.

Karaman explains how simulation has become indispensable in robotics development. By creating detailed virtual environments, researchers can test thousands of scenarios, validate algorithms, and train learning systems without the expense and safety risks of real-world testing. This approach accelerates development cycles and helps identify edge cases before deployment.

The discussion explores how game theory applies to autonomous vehicle development. When multiple autonomous agents share the same road, their behaviors create strategic interactions that cannot be solved through simple rule-based systems. Understanding these game-theoretic principles helps develop vehicles that can predict and respond to other agents' intentions.

Karaman provides insights into Optimus Ride's strategy for autonomous vehicles. Rather than attempting full autonomy everywhere, the company focuses on controlled environments and geofenced areas where variables are limited and predictable. This approach allows for rapid deployment of reliable autonomous shuttles in specific locations like campuses and transit hubs.

The conversation touches on the broader autonomous vehicle landscape, comparing approaches from companies like Waymo and Tesla. Karaman discusses why different companies have chosen different paths, from Waymo's focus on hardware and sensor suites to Tesla's emphasis on camera-based vision systems. He addresses the ongoing debate about whether Lidar has become a crutch that prevents progress on vision-based autonomous driving systems.

A significant portion of the discussion examines achieving seemingly impossible goals in robotics. Karaman emphasizes the importance of iterative learning and incremental progress rather than expecting breakthrough moments. He shares perspectives on how breaking down massive challenges into smaller, manageable problems makes the impossible seem achievable.

Toward the end of the episode, Karaman discusses fast autonomous flight and the algorithms that enable quadrotors and other flying robots to navigate complex environments with remarkable agility. The conversation concludes with reflection on what Karaman considers the most beautiful idea in robotics, touching on elegant mathematical frameworks that enable machines to understand and interact with their environment efficiently.