Russ Tedrake: Underactuated Robotics, Control, Dynamics and Touch | Lex Fridman Podcast #114

Passive dynamic walkers and trout surfing vortices show that if you design mechanics well and stop trying to cancel out the world with motors and control, you can get stable, natural, and efficient motion with minimal actuation.

Most real systems, including humans, have more degrees of freedom than actuators; optimal control and trajectory optimization let you exploit these dynamics rather than pretend everything is fully actuated and simple.

Making and breaking contact introduces discontinuities and combinatorial explosion in possible contact configurations, which breaks many classic smooth-control assumptions and makes both simulation and planning for manipulation fundamentally challenging.

While deep learning enables fast progress, Tedrake argues that clear mathematical models, analysis, and classical control tools are irreplaceable for safety-critical, complex robotic systems and to avoid over-relying on black-box function approximators.

The DARPA Robotics Challenge revealed that you cannot simply run a single robot more often; you need systematic testing, unit and integration tests, good simulators, and active methods to expose rare but catastrophic failure modes.

Soft robotics changes contact mechanics (from point to patch contact) and makes it easier to handle fragile or complex objects, while embedded tactile sensing provides crucial information exactly where cameras are occluded—at the hand–world interface.

In an era of information overload, Tedrake emphasizes reading a few great books and papers deeply, teaching and explaining concepts to others, and building and breaking real systems as the path to genuine expertise.