Physical AI is moving fast.

But Russ Tedrake says the biggest shift may not just be better robots. It may be the way robotics itself is changing.

In this episode of Automated, Brian Heater speaks with Russ Tedrake, Toyota Professor at MIT and founder of a stealth physical AI startup, about why this moment in robotics feels different from past hype cycles.

Russ explains how machine learning has moved ahead of our theoretical understanding, and why that changes the role of robotics engineers. Instead of designing everything from first principles, teams are increasingly building systems they do not fully understand yet, then studying their behavior like scientists.

Brian and Russ also discuss the long arc of robot locomotion, from passive dynamic walkers to today’s humanoid robots. Russ reflects on why bipedal walking was always the dream, why humanoid hardware has become surprisingly turnkey, and why the next exciting question is what AI can do with a powerful general-purpose body.

The conversation also digs into one of the biggest debates in robotics right now: data. Russ argues that the robotics data problem is often framed the wrong way. Robots do not need to learn everything from scratch. Instead, he says the field can build on powerful video and multimodal models that already contain world knowledge, then train those models to output robot actions.

Russ also explains the difference between large behavior models and vision-language-action models, why multitask pre-training may help with robustness, and why real-world deployment is the next major milestone for the field.

Finally, Russ talks about launching a new physical AI company, why he believes robotics may have escape velocity this time, and why the future of work has to be central to the conversation. His goal is not just more capable robots. It is building systems that amplify people rather than replace them.