MIT's Russ Tedrake to Unveil Stealth Physical AI Startup

Large Behavior Models: The New Frontier

The focus of Tedrake's work at Toyota Research Institute — large behavior models — represents the current leading edge of the field. Drawing an analogy to large language models in text AI, large behavior models are trained on vast datasets of robot behavior, either teleoperated demonstrations or simulation-generated experience, and learn general-purpose policies that can be adapted to new tasks and environments with relatively few examples.

The success of this approach in early deployments has been striking. Robots trained on large behavior models can generalize across objects they have not seen before, adapt to new environments without task-specific reprogramming, and recover from failures in ways that traditional controllers cannot. The challenge is that training these models at scale requires both data — expensive to collect in the physical world — and substantial compute, and the resulting models must still be paired with robust control systems that can handle real-time physics.

The Startup Ecosystem Around Physical AI

Tedrake is entering a startup landscape that has become increasingly competitive. Covariant, Figure AI, Physical Intelligence, and several other well-funded companies are pursuing variations of the large behavior model approach to general-purpose robot intelligence. Physical Intelligence in particular has attracted significant attention and funding, and its leadership overlaps with the academic robotics community in ways that create both competitive and collaborative dynamics.

What Tedrake brings to this ecosystem is a specific combination of deep theoretical knowledge — he is one of the few people in the world who can move fluently between the mathematics of control theory and the practical engineering of robot hardware — and executive experience running a large research organization. His time at TRI gave him exposure to the industrial-scale infrastructure required to train, test, and deploy robotic systems at a level that purely academic experience cannot provide.

Boston as a Robotics Hub

The choice of the Robotics Summit in Boston for the unveiling is fitting. Boston has emerged as one of the global centers of robotics innovation, anchored by MIT, Boston Dynamics, and a dense ecosystem of startups and research institutions. The Summit draws engineers, investors, and executives from across the industry, making it the logical venue for a high-profile announcement from a figure of Tedrake's stature.

Details of the startup — its name, funding status, specific technical focus, and team — remain unreleased ahead of the May event. The robotics community will be watching closely. When someone with Tedrake's track record steps out of a major research institution to build a company, the specific bet they are making on the future of the technology tends to be worth understanding in detail.

This article is based on reporting by The Robot Report. Read the original article.