MIT robotics professor Russ Tedrake discusses physical AI, robot learning, deployment, and the human impact of automation with Automated host Bryant Heater in Boston.
An early automation project at a Ford assembly plant taught Tedrake that code can affect workers and production: shutting down paint booth fans pushed the temperature past the union’s 82-degree threshold. Later, at MIT, he used reinforcement learning to teach a small walking robot to move in about 20 minutes, building on passive dynamics and simulation.
Tedrake points to domain randomization in simulation, GPU computing, and neural-network policy gradient methods as key factors behind progress in locomotion. Robots trained across simulated stairs and bumps can handle varied real-world terrain, though researchers still have much to learn about why these methods work so well. He sees theory and careful experiments as useful for data efficiency, robustness, and practical deployment.
Tedrake uses “large behavior model” for models that take image sequences and produce actions, with vision-language-action models as one possible design. He argues robotics teams can build on the world knowledge of pretrained video and multimodal models, then use targeted data to teach them to connect that knowledge to robot actions. Simulation, teleoperation, and real-world examples can each help build that bridge.
Tedrake calls real-world deployment the next major milestone: capable robots could enable more deployments, which would generate data for further improvement. He wants automation to amplify workers and has been speaking with labor economists and potentially affected people about its effects on work, purpose, and aging in place. He says companies should help people understand and adapt to the tools.
Tedrake says leaders need clear priorities while staying ready to change course when a real breakthrough appears. He keeps hands-on by coding and relies on knowledgeable colleagues to help filter the flood of research. He sees a convergence of talent, investment, manufacturing capacity, and demand as giving robotics a strong opportunity, while emphasizing that success still depends on execution.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
MIT professor and Toyota Research Institute vice president Russ Tedrake explains how graphs of convex sets can combine trajectory optimization and discrete search for robot planning at the MIT Robotics Seminar.
Russ Tedrake, MIT associate professor and Toyota Research Institute vice president of robotics research, discusses feedback control and manipulation at a University of Toronto Robotics Institute seminar hosted by Florian.
