Key Moments
Russ Tedrake: Underactuated Robotics, Control, Dynamics and Touch | Lex Fridman Podcast #114
Lex FridmanKey Moments
Robotics expert Russ Tedrake discusses elegant control, animal motion, bipedal robots, barefoot running, and the Darpa Robotics Challenge.
Key Insights
Passive dynamic walkers, powered solely by gravity, demonstrate the beauty and efficiency of natural robot motion, challenging conventional control approaches.
Nature, through evolution, optimizes for efficiency, as seen in fish 'surfing' vortices, highlighting the interplay between mechanics and control systems.
Bipedal robots, while complex, are valuable for navigating human-centric environments; their development pushes the boundaries of control theory.
Barefoot running emphasizes immediate feedback and refined gait for long-term injury prevention, mirroring a philosophy of letting physics guide movement.
The Darpa Robotics Challenge underscored the critical importance of robust software engineering, extensive testing, and managing complexity in safety-critical robotic systems.
Advancements in simulation, especially in handling contact mechanics and non-smooth dynamics, are crucial for developing sophisticated robotic manipulation and interaction.
THE BEAUTY OF PASSIVE DYNAMIC WALKERS
The discussion begins by exploring the concept of beautiful motion, highlighting passive dynamic walkers as prime examples. These robots, powered primarily by gravity and gravity alone, showcase an inherent grace and efficiency that rivals or surpasses human-like movement. The rimless wheel and compass gate walker models illustrate the fundamental principles of using physics to achieve stable motion, suggesting a paradigm shift from forceful control to harmonious interaction with physical forces. This contrasts with traditional robotic control, which often aims to impose the robot's desired dynamics by overpowering external forces.
NATURE'S EFFICIENCY AND THE DYNAMICS OF MOVEMENT
Tedrake draws parallels between robotic efficiency and natural systems, citing experiments with fish 'surfing' vortices in water tunnels. The observation that even a dead fish can be propelled upstream by resonating with these vortices underscores the power of mechanics and natural dynamics. This leads to a deeper consideration of how evolution optimizes for efficiency, not as an sole goal, but as a fundamental strategy for survival. The distinction between mechanics (passive dynamics, springs, dampers) and control (active energy input via motors) is explored, emphasizing their intricate interplay.
THE CASE FOR BIPEDAL ROBOTS AND HUMAN-CentIMPERFECT MOVEMENT
The evolution of bipedal locomotion in humans is examined, with Tedrake suggesting that while the exact evolutionary drivers are debated, the ability to navigate a world built for humans is a strong factor. From a robotics perspective, building bipedal robots is challenging and pushes control theory, but it's also pragmatic for operating in human-designed environments. The discussion touches upon the personal practice of barefoot running, which, for Tedrake, reinforces the importance of direct sensory feedback and adapting gait for long-term physical well-being, mirroring the idea of letting physics and feedback guide motion.
THE DARPA ROBOTICS CHALLENGE: LESSONS IN COMPLEXITY AND ROBUSTNESS
A significant portion of the conversation delves into the Darpa Robotics Challenge (DRC), a pivotal experience for Tedrake. He emphasizes the immense challenge of developing robust software and control systems for complex, semi-autonomous humanoid robots like Atlas. The DRC highlighted the difficulty of testing such systems adequately and the crucial role of rigorous software engineering. Failures, like the infamous robot falls, were attributed to a combination of factors, including unexpected environmental interactions, control system conflicts, and limitations in state estimation, stress-testing the robots' ability to handle unforeseen situations.
SIMULATION, CONTACT MECHANICS, AND ROBOTIC MANIPULATION
The development of simulation tools like Drake is discussed as essential for tackling complex robotic problems, particularly those involving contact. Tedrake explains that simulating contact is inherently challenging due to discontinuities and the indeterminacy arising from rigid body assumptions. This difficulty limits the success of robotic manipulation, as robots often rely on soft contact and an understanding of non-smooth dynamics. The team's work in simulating and controlling tasks like loading a dishwasher illustrates the ongoing effort to bridge the gap between simulation and real-world deployment, emphasizing the need for robust validation.
UNDERACTUATION, SOFT ROBOTICS, AND THE FUTURE OF HUMAN-ROBOT INTERACTION
Underactuated robotics, where a system has more degrees of freedom than actuators, is presented as a fundamental design principle, both in nature and in robots. Tedrake argues that embracing this limitation, rather than fighting it, leads to more elegant solutions. The concept of soft robotics is introduced as a way to better interact with the world, moving beyond rigid designs to incorporate tactile sensing and more compliant interactions. This is linked to the future of home robotics and the goal of helping people age in place, envisioning robots that are less intimidating and more seamlessly integrated into our lives.
THE PHILOSOPHY OF RIGOROUS THINKING AND LEARNING
Tedrake stresses the importance of rigorous thinking, especially in contrast to the ease of using deep learning without deep understanding. He advocates for a return to first principles, fundamental mathematics, and deep engagement with complex problems. The conversation touches on the value of structured learning, the power of well-chosen books and papers, and the benefits of teaching as a method of solidifying understanding. The journey from simple models to complex, real-world applications requires a disciplined approach to learning and problem-solving.
FEAR OF ROBOTS AND THE EVOLVING HUMAN-ROBOT RELATIONSHIP
The conversation addresses the common human fear of robots, attributing it to cultural narratives like 'The Terminator' and the anthropomorphism of robotic forms. Tedrake believes that while caution is necessary, the fear is often amplified by science fiction and a misunderstanding of current capabilities. He suggests that the future may involve greater integration, with humans and robots becoming more intertwined, rather than a futuristic conflict. The potential for robots to be perceived and utilized differently, perhaps as less 'other' and more collaborative, is explored, especially with softer, more tactile designs.
THE CHALLENGE OF MACHINE LEARNING AND STATE REPRESENTATION
Tedrake highlights the challenges in applying machine learning to complex robotic tasks, particularly concerning state representation. He argues that raw sensor data or the 'world state' from a Lagrangian perspective isn't always the most useful representation for control. Instead, relevant, compressed, or 'latent' states are needed to enable effective decision-making. Examples like chopping onions or buttoning a shirt illustrate how the relevant state for a task can drastically differ from a complete physical description, necessitating new ways of thinking about how AI perceives and interacts with the world.
DISCOVERING CORNER CASES AND THE POWER OF FLEET LEARNING
The difficulty of identifying and addressing 'corner cases' in robotic systems is discussed, where unexpected failures occur. Tedrake suggests that active experiment design, rare event simulation, and 'fleet learning' are crucial strategies. By collectively gathering data from many robots and focusing on instances where systems perform poorly, researchers can accelerate learning and improve robustness. This distributed approach, akin to how humans learn from collective experience, offers hope for developing more reliable robots, especially in safety-critical applications like autonomous driving and home assistance.
INSPIRATION FROM BOSTON DYNAMICS AND THE ROBOTICS ECOSYSTEM
The impact of companies like Boston Dynamics in inspiring public imagination and driving robotics innovation is acknowledged. However, Tedrake also notes the high failure rate of robotics startups, emphasizing the inherent difficulty of commercializing complex robotic systems. He views this as a natural part of an 'ecology,' where many companies attempt ambitious projects, leading to both spectacular successes and inevitable failures, ultimately pushing the field forward. The personal connection and emotional investment in building and working with robots like Atlas are also touched upon.
THE CORE OF ROBOTICS: CONTROL AND UNCERTAINTY
Defining a robot as a system combining computation with mechanical work that solves challenging problems, Tedrake elaborates on the central role of control. He distinguishes between different control problems for various robot types, focusing on manipulation and dynamic interactions as particularly rich areas. The inherent uncertainty and discontinuities in contact mechanics pose significant control challenges, contrasting with the smoother dynamics of systems like aircraft. This complexity necessitates robust control strategies that can manage unpredictable events and ensure task completion despite dynamic interactions.
EXCESSIVE OPTIMIZATION, CONTACT PROBLEMS, AND FUTURE DIRECTIONS
Tedrake reflects on the limitations of current robotic systems, particularly concerning contact and manipulation. He points out that while many manipulation tasks could be solved with current technologies, the truly challenging problems, like tying shoelaces, require deeper understanding of underactuated dynamics and soft interactions. The ideal future involves robots that embrace the world's softness, enabling safer and more effective manipulation. This pursuit of challenging problems, even if they are difficult to commercialize immediately, is what drives innovation in the field.
THE IMPORTANCE OF TOUCH AND EMBODIED LEARNING
The significance of touch in human-robot interaction and robotic manipulation is emphasized. Tedrake posits that tactile sensing is crucial for robots to interact safely and effectively, especially when working near humans. He advocates for 'soft robotics' that can change the mechanics of contact, making robots less afraid to engage with the world. This aligns with the idea that physical interaction, akin to how children explore the world through touch and manipulation, is fundamental for learning and developing sophisticated robotic capabilities, fostering deeper human-robot connections.
ADVICE FOR FUTURE GENERATIONS AND THE VALUE OF DEEP THINKING
Tedrake offers advice to young people interested in robotics and life, stressing the importance of deep, critical thinking, mathematical foundations, and hands-on building. He encourages embracing challenges, learning continuously, and understanding the 'why' behind phenomena. The conversation concludes with reflections on the intellectual journey, the early days at MIT, and the ongoing pursuit of knowledge. The value of finding a few deeply resonant books and papers, and dedicating oneself to understanding them thoroughly, is highlighted as a path to mastery in a world of information overload.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Common Questions
Passive dynamic walkers are robots that use gravity and their own mechanical design to walk gracefully down a ramp without motors or controllers. Simple models like a rimless wheel or a compass gate walker illustrate how they maintain motion through an energy balance of gaining energy from falling and losing it with each step.
Topics
Mentioned in this video
Russ Tedrake is a professor at MIT, and the institution is central to his work and the DARPA Robotics Challenge team he led.
Russ Tedrake is the Vice President of Robotics Research at TRI, which invests in simulation environments like Drake and works on home robotics to help people age in place.
Their team pushed the capabilities of Gazebo to scale for the DARPA Robotics Challenge, playing an important role in the virtual competition.
The team that won the DARPA Robotics Challenge, noted for their incredible engineering and disciplined testing approach using two robots.
Honda's humanoid robot, considered the best walking robot in the world in the late 90s/early 2000s, though noted for being less energy-efficient than passive walkers.
Boston Dynamics' quadruped robot initially known for cool videos but later featured in a viral video of patrolling, eliciting dystopian fears from the public. Praised for its new features.
iRobot's successful robotic vacuum cleaner, mentioned as an example of a robotics success story despite the difficulties in the industry.
Boston Dynamics' 400-pound humanoid robot, used in the DARPA Robotics Challenge. The MIT team adapted their control software for Atlas, experiencing both triumphs and spectacular failures.
A small car provided in the DARPA Robotics Challenge that Atlas had to drive and exit, which proved to be one of the most difficult obstacles due to size constraints and robot configuration.
A home robotics company mentioned as an example of a robotics company that failed, despite promising initial concepts.
Lex Fridman mentions owning and loving this autonomous drone, indicating a positive experience with consumer robotics.
His perspective on AI's future, where humans become 'robot people' due to increasing integration of technology, is highly favored.
Collaborated with Russ Tedrake and Aman Sinha on falsification algorithms for rare event simulation in robotics testing.
Worked at MIT on experiments with George Lauder, studying fish swimming behavior and energy efficiency behind rocks in a water tunnel.
Mentioned as a public figure who discusses the existential threats of artificial intelligence and robots.
Mentioned as one of the public figures who discusses the existential threats of artificial intelligence and robots.
Conducted experiments at Harvard studying fish swimming in water tunnels, specifically rainbow trout, to understand energy efficiency and how fish leverage vortices.
Mentioned as a builder of passive dynamic walkers with Andy Ruina at Cornell, creating graceful 3D walking machines powered only by gravity.
Initiated the original research on passive dynamic walkers, leading to the work by Collins and Ruina.
A sharp student who worked with Russ Tedrake to write a new, faster quadratic programming solver (FastQP) for the DRC, critical for coping with simulator speed changes.
Author of 'Fooled by Randomness' and 'The Black Swan', whose books are recommended for making the point about the impact of random events.
Quoted at the end of the podcast, discussing the human emphasis on human-based space exploration versus robotic exploration, and the public's excitement.
A theoretical physicist and computational neuroscientist at MIT, whose demanding intellectual culture profoundly impacted Russ Tedrake early in his MIT career.
Co-builder of passive dynamic walkers at Cornell with Steve Collins, whose work exemplified graceful, energy-efficient robot motion.
Author of a book on barefoot running, recommended by Russ Tedrake for its effective description of how to approach the practice.
Cited alongside Newton as an example of a scientist whose rigorous thinking and simple models were crucial for scientific progress.
A perceptual scientist at MIT who, after having children, shifted his focus from vision to tactile sensing due to observing how children interact with the world through touch.
A great science fiction writer known as 'the great explainer' for his non-fiction work, whose passion for clear communication and explanation resonates deeply.
Author of 'AI Superpowers', a book highly recommended for understanding the current landscape of AI.
Author of 'Animal Farm', a book that deeply influenced Lex Fridman with its themes of power and history.
An ultra-endurance athlete and motivational speaker known for pushing physical limits; Lex Fridman mentions talking to him daily and being inspired by his philosophy of suffering.
Led the Open Source Robotics team that developed the Gazebo simulator for the DARPA Robotics Challenge.
Author whose books, 'Sapiens' and 'Homo Deus', are recommended for their mind-opening philosophical insights.
Author of 'How to Read a Book', which Russ Tedrake found impactful for its romantic view of deep reading and intellectual dialogue with authors, especially in an era of overwhelming low-quality content.
Worked with Russ Tedrake and Matt O'Kelly on developing falsification algorithms for efficiently discovering corner cases in robotic simulations.
Cited as an example of rigorous scientific thinking, whose simple models (like F=ma) were transformative, in contrast to merely having complex neural network weights.
Provided the Atlas humanoid robots for the DARPA Robotics Challenge; highly praised for their innovative work in legged robotics and their ability to inspire the world.
Mentioned in the context of cognitive implants and brain-computer interfaces, representing a potential future where technology is deeply integrated with humans.
Its users/drivers are mentioned as another example of people participating in 'fleet learning' for autonomous systems, willing to tolerate imperfections for improvement.
Cited as an example of a robotics company that went through significant challenges to succeed with a single product, the Roomba.
A robotics company mentioned as one that failed, but had the idea of humanizing industrial robots, like Baxter, with eyes to improve worker experience.
A Japanese manga and anime series referenced as a cultural influence contributing to a more positive perception of robots in Japan, in contrast to the 'Terminator' influence in the U.S.
The soft, inflatable robot from 'Big Hero 6', cited as an ideal concept for home robots that are fundamentally safe and inspire positive human interaction.
A TV series known for its dystopian science fiction themes, specifically an episode that fueled public fear and anthropomorphism of robots like Boston Dynamics' Spot.
The movie featuring Baymax, a soft robot that Toyota Research Institute aims to emulate for home robotics, focusing on safety and positive interaction.
A science fiction film series cited as a cultural influence contributing to the fear of robots in the U.S., contrasting with Japan's 'Astro Boy' influence.
Lex Fridman was born in the Soviet Union, which allows him to deeply connect with the themes of 'Animal Farm' related to its history, suffering, and the nature of power.
Russ Tedrake mentions rowing on the Charles River as part of his commute, a magical time for thinking and decompressing.
The second book by Yuval Noah Harari, recommended for its mind-opening insights into the future.
A book by Kai-Fu Lee, recommended as a fantastic and essential read for those interested in AI topics.
The first book by Yuval Noah Harari, recommended for its mind-opening perspective on humanity.
A book by Mortimer Adler that Russ Tedrake recently read and loved, appreciating its emphasis on deep, deliberate engagement with books as a dialogue.
A book by Rodney Brooks, cited for its perspective that the future will see humans and robots integrated, rather than in conflict.
A book mentioned as being influential for its insights into randomness and highly impactful rare events, described as a 'mind opener'.
A book about barefoot running that Russ Tedrake read later in his journey, after already having started barefoot running based on conference talks.
A book by George Orwell that Lex Fridman has read a 'ridiculous number of times' and finds deeply connecting to history and the corrupting nature of power, inspiring more about life than anything else.
The simulation environment used in the virtual phase of the DARPA Robotics Challenge, run on the cloud, which presented challenges with real-time physics simulation.
Acknowledged for its incredible contribution to the robotics community, enabling rapid integration of different components, but also contrasted with Drake's more structured approach.
A robotics company that created a 'cute toy' robot but ultimately failed, highlighting the difficulty of making money in home robotics.
More from Lex Fridman
View all 505 summaries
154 minRick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492
23 minKhabib vs Lex: Training with Khabib | FULL EXCLUSIVE FOOTAGE
196 minOpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491
266 minState of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free