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

Core Expertise & Philosophy

• Russ Tedrick is a roboticist, professor at MIT, and Vice President of Robotics Research at Toyota Research Institute (TRI).
• He specializes in control systems for underactuated, stochastic, and difficult-to-model dynamical systems, such as the human body.
• Tedrick advocates for "letting physics do the work" by designing control systems that leverage passive dynamics rather than fighting them with high-power actuation.
• He distinguishes between mechanics (passive dynamics, gravity, springs/dampers) and control (active energy input via motors) but emphasizes their interdependence in nature and robotics.

Passive Dynamics & Efficiency

• Tedrick identifies Steve Collins and Andy Ruina's 3D passive dynamic walkers at Cornell as the most beautiful robot motion he has seen; these robots walk solely on gravity down a ramp with no motors or batteries.
• The mechanics of these walkers are modeled after "rimless wheels" (spokes) where energy loss from impact balances energy gain from stepping, creating a stable periodic motion.
• Experiments by George Lauder (Harvard) and Mike Tranavilo (MIT) showed rainbow trout surf on vortices behind rocks to save energy; remarkably, a dead trout suspended in such a vortex would swim upstream due to mechanical resonance with the water flow.
• Tedrick notes that biological efficiency is a co-evolutionary result of mechanical adaptation (e.g., bone remodeling after injury) and motor control, not just neural intelligence.

The DARPA Robotics Challenge (DRC)

• The DRC was launched post-Fukushima to develop disaster-response robots capable of driving vehicles, turning valves, and cutting through walls.
• Tedrick's MIT team competed in the simulation track before winning a spot to control a Boston Dynamics Atlas humanoid for the physical challenge.
• The competition highlighted the difficulty of semi-autonomy under degraded network conditions, requiring robots to function with limited human tele-operation.
• Tedrick's team faced a critical failure when extracting the robot from a Polaris vehicle: a human error in their checklist left a locomotion controller active while a balancing controller was engaged, causing the robot to fall into a configuration it could not recover from (butt on seat, feet in air) because it lacked state estimation for that specific contact topology.
• The winning team (KIST) succeeded largely due to having two robots, allowing for an aggressive testing regimen that uncovered rare failure modes Tedrick's single-robot team could not replicate.

Simulation & Software Engineering (Drake)

• Tedrick leads the development of "Drake," an open-source C++/Python framework for multi-body dynamics, optimization, and control.
• Drake is designed to bridge the gap between complex simulation and rigorous control theory, offering tools for linear/quadratic programming and system modeling with strict verification.
• He argues that current rigid-body simulation assumptions lead to "differential inclusions" and paradoxes (e.g., indeterminate forces in a 4-legged table) that make contact physics difficult to control and simulate.
• Toyota Research Institute (TRI) is using Drake to tackle home robotics, specifically the "dishwasher unloading" problem, which requires handling infinite corner cases and variable object geometries.

Underactuated Robotics

• Tedrick defines "underactuated" systems as those with fewer actuators than degrees of freedom (e.g., a human jumping, where the center of mass is unactuated once airborne).
• He argues that the world is inherently underactuated and that robots must learn to use passive dynamics and optimization rather than attempting to fully control every variable.
• Manipulation tasks like buttoning a shirt are inherently underactuated because the state space (fabric deformation) is infinite and high-dimensional, requiring "latent" or compressed state representations rather than full physical modeling.

Fleet Learning & Soft Robotics

• TRI is pursuing "fleet learning," where thousands of home robots (e.g., dishwashers) collectively share data to accelerate learning and identify corner cases faster than any single robot could.
• Tedrick champions soft robotics, arguing that rigid bodies are unnatural and that softness transforms contact from a discontinuous, hard-to-simulate event into a continuous, distributed one.
• Soft robotics facilitates the integration of tactile sensors (often embedded in the "skin"), providing rich feedback when vision is occluded during manipulation.

Human-Robot Interaction & Safety

• Tedrick predicts a future where humans and robots merge ("robot people"), citing neural interfaces and implants as a trend that will make the distinction between human and machine blur.
• He suggests that fear of robots (e.g., from Black Mirror) is culturally conditioned by sci-fi tropes rather than inherent to the technology.
• He advocates for "Baymax-style" robots—soft, contact-rich, and approachable—to facilitate safe physical interaction and emotional connection, contrasting this with the "Terminator" aesthetic.

Education & Personal Practices

• Tedrick teaches a course on underactuated robotics and emphasizes "rigorous thinking" over the lazy application of deep learning, urging students to understand first principles (physics, math) before applying data-driven methods.
• He runs approximately 12 miles to and from work barefoot or in minimal shoes to gain proprioceptive feedback that forces a softer, more efficient gait and prevents long-term injury.
• Tedrick recommends deep reading over skimming the internet, citing Mortimer Adler's How to Read a Book as a guide to mastering foundational texts rather than chasing the "fear of missing out" on new research papers.
• He views the "struggle" of hard work and deep learning as intrinsically rewarding and the source of happiness, rather than a burden.

Future Outlook

• Tedrick is optimistic about home robotics and manipulation, predicting that robots will become ubiquitous in logistics and elder care within 10–20 years.
• He believes the field of robotics is currently in its "early days," characterized by failures and rough prototypes, but essential for future societal advancement.
• He expects robots to eventually compete in physical games (like soccer) by 2050, though they may play differently than humans, exploiting their mechanical advantages rather than mimicking human biology.