Marc Raibert: Boston Dynamics and the Future of Robotics | Lex Fridman Podcast #412

Introduction (00:00:00)

• Boston Dynamics' robots are known for their natural and beautiful movements.
• Marc Raibert, the founder and former CEO of Boston Dynamics, has been leading the creation of legged robots for over 40 years.
• Raibert has worked at Carnegie Mellon University, the MIT Leg Lab, and Boston Dynamics.
• He is currently the executive director of the Boston Dynamics AI Institute, which focuses on research and development of future generations of robots.

Early robots (00:01:43)

• Marc Raibert was always a builder from a young age.
• He was inspired to become a roboticist after seeing a disassembled robot arm in pieces in 1974.
• Raibert was initially in the Brain and Cognitive Sciences Department but found neurophysiology too abstract.
• He was drawn to the AI lab and the potential of robotics to understand intelligence.
• Raibert emphasizes the balance between function and creating something cool in robotics.
• He initially focused solely on function but later realized that aesthetics also play a role in the perception of robots.
• As a child, Raibert and his friend would disassemble fluorescent light components and turn them into rockets.
• Raibert reflects on the balance between function and creating something cool in robotics.
• He initially prioritized function over aesthetics but later realized that aesthetics also contribute to the perception of robots.
• Raibert discusses the tension between the Brain and Cognitive Science (BCS) approach and the robotics approach to understanding intelligence.
• BCS has evolved to bridge the gap between neuroscience and computer science.
• David Maher was a visitor in the AI Lab who created brain models appealing to biologists and computer scientists.
• Raibert highlights the cyclical nature of collaboration between BCS and robotics.

Legged robots (00:06:47)

• Marc Raibert, founder of Boston Dynamics, believes robots should be more aggressive and dynamic in their movements, like animals, rather than cautious and careful.
• Raibert's early work at the MIT Leg Lab focused on creating hopping robots that could balance and move dynamically, which laid the foundation for Boston Dynamics' subsequent advancements in robotics.
• The first hopping robot, created in 1982 with funding from DARPA, aimed to understand legged locomotion and animal movement rather than create practical machines.
• Raibert and Ben Brown collaborated to design and build a hopping robot that could start to work, overcoming challenges in making it work in 3D.
• Raibert emphasizes the importance of self-belief and finding the right collaborators when pursuing ambitious ideas, despite skepticism from others.
• Boston Dynamics began as a physics-based simulation company before transitioning into robotics through collaborations with Sony, including the development of the IBO Runner and programming tools for the humanoid robot, Cujo.
• Spot, Boston Dynamics' quadruped robot, has been well-received by the public, contrary to media portrayals of fear and terror.

Boston Dynamics (00:25:27)

• Boston Dynamics started with a surgical simulator that provided force feedback and 3D graphics.
• The simulator was designed to train surgeons and had a scoring system based on surgeons' inputs.
• Surgeons were competitive and enjoyed using the simulator, but they expected to be paid for teaching rather than pay for training.
• The company realized they lacked the resources to market the simulator to hospitals.
• After considering their options, Marc Raibert and Rob decided to pivot from surgical simulators and focus on building robots.

BigDog (00:28:45)

• Boston Dynamics transitioned from a simulation company to a robotics company, creating the quadruped robot "eyebrow runner" and later developing "Big Dog," a large quadruped robot funded by DARPA.
• Big Dog integrated power, computing, and sensing systems onto a single platform, allowing autonomous operation in challenging outdoor environments.
• Testing Big Dog in real-world scenarios provided insights into balancing and controlling robots in various environments.
• Early robots required skilled human operators, but advancements in controls enabled even amateurs to operate them effectively.
• Big Dog evolved into LS3, a larger robot capable of carrying heavy loads over long distances, while Spot, a later prototype, was fully electric and non-hydraulic, making it suitable for indoor environments.
• The inspiration for Spot came from a conversation with Larry Page, who envisioned a smaller, less intimidating robot for home use.

Hydraulic actuation (00:36:52)

• There are technical challenges in transitioning from hydraulic to electric actuation.
• Hydraulics offer great strength in a lightweight package.
• Boston Dynamics innovated hydraulic control valves, making them smaller, lighter, and more efficient.
• They developed a compact hydraulic power supply that integrates various components, producing 5 kilowatts of power.
• Hydraulics still has potential for further advancements.

Natural movement (00:38:44)

• Boston Dynamics' robots achieve natural-looking movements through advanced hardware and a dynamic approach that predicts future motion within a short time frame.
• For complex maneuvers like somersaults, calculations must be made further in advance to ensure a successful landing, requiring precise coordination of momentum, rotation, and foot placement.
• Robots provide insights into human movement and capabilities, as building them requires a deep understanding of the physics and algorithms involved in natural movements.
• Despite advancements, challenges remain in achieving human-like walking gracefulness, while running has come closer to resembling human motion.
• Marc Raibert, the founder of Boston Dynamics, emphasizes the importance of embracing the unknown aspects of robotics and incorporating compliance, or springiness, in robot design for efficient movement.
• Boston Dynamics' motto in its early days, "You have to run before you can walk," highlights the need for ambitious goals in robotics development.
• The Wildcat quadruped robot achieved a speed of 19 miles per hour on flat terrain, possibly making it the fastest quadruped in the world, but its racing go-kart engine drew complaints from nearby residents.

Leg Lab (00:44:31)

• Marc Raibert, the founder of Boston Dynamics, discusses the design and development of legged robots.
• Animals like cheetahs, pronghorns, and Olympic-level gymnasts exhibit graceful and efficient movements that inspire robot design.
• The number of joints and actuators in a robot's leg is determined by balancing efficiency, functionality, aesthetics, and practical considerations.
• Boston Dynamics' Big Dog robot was the first quadrupedal robot with knees but lacked a mechanism to store negative work during knee contractions.
• Boston Dynamics' robots, such as Spot, utilize energy-driven controls instead of solely relying on computer-controlled actuators.
• Passive Dynamics suggests that mechanical systems can generate efficient motion without constant computer control.
• Efficient robot design involves considering the mechanical properties of the robot and allowing the body to participate in the motion, rather than solely relying on computer control.

AI Institute (00:51:23)

• The Boston Dynamics AI Institute focuses on designing robots of the future.
• Marc Raibert believes intelligence has two parts: athletic and cognitive.
• Boston Dynamics has set the standard for athletic intelligence through mechanical design, real-time control, and energetics.
• Robots currently lack cognitive intelligence and require skilled people to program them, hindering their ability to fulfill human dreams.
• The AI Institute aims to combine athletic and cognitive intelligence in robots, enabling them to perform tasks like walking, understanding, and executing instructions.
• Achieving this goal is challenging and considered science fiction, but the institute adopts a long-term approach to solving these problems.

Athletic intelligence (00:54:41)

• Boston Dynamics' robots, such as Spot and Atlas, have demonstrated remarkable athletic intelligence, pushing the boundaries of what is considered possible for robots.
• Despite their physical prowess, robots still face challenges in terms of cognitive abilities, reliability, and cost-effectiveness, which hinder their commercial viability.
• Hyundai's collaboration with Boston Dynamics aims to leverage Hyundai's expertise in car manufacturing to address these challenges, potentially leading to more affordable and reliable robots.
• The institute employs a "Stepping Stones to Moonshots" approach, emphasizing incremental progress and tangible results to maintain motivation and provide feedback.
• One project involves analyzing human actions observed in videos, breaking them down into components, and mapping them to skills that robots can perform.
• The institute is initially focusing on simple bicycle repair tasks to develop these skills, as they encompass both complex and straightforward elements.
• Robots need to be able to navigate and operate effectively under uncertain and unspecified conditions.
• While machine learning is utilized in robotics, physical tasks differ from language tasks, and pixel values are not directly comparable to words.
• Boston Dynamics has an office in Zurich led by Marco Hutter, an expert in reinforcement learning for robots.
• Despite exploring machine learning approaches, the most impressive robotic performances still rely on traditional control methods like model predictive control.
• Boston Dynamics is pursuing a strategy of combining traditional controls with machine learning to achieve the optimal balance of athleticism and intelligence in their robots.

Building a team (01:02:35)

• Technical fearlessness: willingness to tackle unsolved problems, even if they seem daunting.
• Diligence:
  • Importance of persistence and hard work in achieving goals.
  • Overcoming challenges and setbacks through continuous effort.
  • Dedicating time and resources to research and development.
• Willingness to take on challenging problems without knowing the solution.
• Studying and learning from simplified versions or stepping stones.
• Overcoming the fear of the unknown and pursuing ambitious goals.
• Importance of persistence and hard work in achieving goals.
• Overcoming challenges and setbacks through continuous effort.
• Dedicating time and resources to research and development.

Videos (01:05:37)

• Diligence in robotics involves embracing broader problems and testing robots in all conditions to create robust solutions.
• Boston Dynamics' videos effectively demonstrate the scope, limits, and challenges of their robots by showcasing both failures and successes without unnecessary explanations or distractions.
• Simplicity is crucial when showcasing robots' capabilities, focusing on what's worth showing without adding extra production elements.
• Intrepidness is essential in robotics, requiring courage to persist through failures and keep trying until the desired results are achieved.
• Atlas, a humanoid robot developed by Boston Dynamics, required 109 attempts to successfully climb three big steps.
• The process involved robot learning with human assistance, and data from each attempt was used to improve performance.
• Atlas is designed to be robust enough to withstand the rigors of testing and frequent falls without breaking.
• Boston Dynamics emphasizes the importance of accepting failures and having a plan in place to repair and maintain robots to make progress in robotics.

Engineering (01:13:25)

• Engineers get to do work that they enjoy.
• Engineering allows individuals to have an impact on the world.
• Working in a team with diverse skills can be very satisfying.
• Engineers get paid for their work.
• Engineering combines science and art, allowing engineers to create new things that didn't exist before.
• Creating something totally new and bringing metal or machines to life is fun and magical.
• Designing efficient and effective systems can be very satisfying.

Dancing robots (01:16:53)

• Marc Raibert believes that the elegance of movement is one of the most beautiful aspects of robotics.
• Boston Dynamics is working on creating robots that can dance with humans.
• Currently, the robots' dancing is based on a fixed time base and they do not interact with humans or respond to their movements.
• Raibert envisions a future where robots can learn to dance from humans through intuitive communication that combines language and movement.
• He acknowledges the challenge of making robots perform alongside professional dancers due to the limited degrees of freedom in robots compared to humans.
• Raibert suggests that robots could excel in certain dance styles, such as robot animation, by performing unstable oscillations and movements faster than humans.

Hiring (01:21:40)

• To hire great engineers, create an environment where interesting engineering is happening.
• Boston Dynamics initially focused on the computing side and gradually achieved high levels of engineering.
• Hiring people who are passionate about building and making, even if they don't have professional degrees, is important.
• People who love repairing cars or building motorcycles in their garages often have the qualities needed for robotics.
• Videos have helped showcase robotics and inspire people, generating excitement and interest in the field.

Optimus robot (01:25:32)

• Marc Raibert, the founder of Boston Dynamics, admires Elon Musk's achievements but believes that Optimus, Tesla's humanoid robot, is not yet as advanced as Atlas, Boston Dynamics' humanoid robot.
• Raibert acknowledges the progress made by companies like Figure and EPtronic in robotics and primarily uses Spot robots as a platform for research and development at the AI Institute.
• While not overly concerned about competition in physical robotics, Raibert recognizes the increasing competition in cognitive AI.
• Raibert sees potential for quadruped and humanoid robots to become more affordable through mass production.
• Boston Dynamics welcomes competition in the quadruped robotics market as it validates the potential of quadrupeds and shifts the focus from their capabilities to selecting the best quadruped for a specific task.
• The primary challenge in robotics is identifying practical use cases that generate revenue, with warehouse and industrial applications currently being the main sources of income. However, there's potential for social robots, like Spot-type robots, to become widespread and profitable in the future.
• Creating successful social robots requires striking a balance between performance, safety, and cost while effectively communicating their benefits to the public.
• Overcoming negative perceptions about in-home robots with cameras is crucial for widespread acceptance, and proper presentation and clear boundaries can help establish trust and make people comfortable with the technology.
• Just as smartphones with cameras have become widely accepted, robots with cameras could potentially become commonplace in the future as people become more accustomed to the technology.

Future of robotics (01:34:02)

• Marc Raibert believes that computers are already smarter than humans in some aspects, but not in others.
• He thinks the conversation about AGI (Artificial General Intelligence) is puzzling because he doesn't understand why people assume that smarter computers would be a threat.
• Raibert argues that even if computers were much smarter than humans, they could still be available to us or under our control.
• He dismisses fears that superintelligent systems could harm humans in unpredictable ways, comparing it to the concerns about the first nuclear chain reaction.
• Raibert believes that the risks of new technologies should be balanced against the opportunities they offer, citing the example of cars.
• Marc Raibert's signature Hawaiian shirt is a symbol of his contrarian nature.
• He started wearing it regularly after someone told him they were old-fashioned.
• Raibert sees his contrarianism as a useful tool.
• Raibert finds negative motivation, such as people telling him something can't be done, to be more effective than positive motivation.
• He credits a former colleague, Ed Tovar, for teaching him the importance of asking "why not" when faced with challenges.