Lex Fridman - Robert Playter

1) Boston Dynamics CEO, Robert Playter, discusses the elegance and beauty in building robots that seamlessly integrate physics and intent. He references his work with the legendary engineer, Marc Raibert, and the lessons he learned, including pursuing curiosity and interests, doing something because you love it, and having the courage of your convictions to find big problems to tackle. Playter highlights the source of happiness in robotics and the physical interaction between humans and robots that creates an external connection, making it easier to identify with the work. He notes that people who pursue something they love can find greater fulfillment and engagement, which is what keeps people at Boston Dynamics.

2) Robert discusses the importance of simplifying problems in robotics and how the company tries to simplify to the core essence. He shares how the company started with a pogo stick and how solving the fundamental problems in balancing it could eventually lead to a galloping robot. He also emphasizes the need to break things and fix them to make progress and how it's not always easy to work at a principled level with both software and hardware. Playter believes that one should see breaking tools positively and encourages people to learn from working with machines.

3) Robert discusses the art and science behind feedback control and Dynamics in the development of legged and humanoid robots. He explains how the scientific principles of a spring Mass bouncing system are applied to develop core principles for legged robots, and the complexity involved in the development of humanoid robots. Playter emphasizes the importance of getting the physics right in achieving lifelike and elegant movement, which makes it more efficient and stable. Finally, Playter shares his thoughts on the challenges involved in developing humanoid robots like Atlas.

4) Robert discusses the challenges of achieving natural-looking gate and walking movements for their humanoid robots. They initially took a contract from the army to build a robot that could walk naturally in order to test it in a gas environment while wearing a suit. This required a lot of math and physics to overcome the challenge of making the robot land on an extended leg and immediately flex the knee upon making contact. The team had to simplify the human body into a simpler subsystem to generate these movements in earlier generations of robots. However, new techniques have allowed them to take the full physics of the robot into account to deal with singular configurations, such as the straight leg.

5) Robert discusses the challenges of creating humanoid robots, particularly with regards to maintaining balance and handling heavy objects. The humanoid's complex balance system and heavy legs make it a more difficult platform than quadruped robots, and explicit modeling is required to handle the shape and weight distribution of the objects it picks up.The ability to create new behaviors for the robot is becoming quicker, with the tools developed by Boston Dynamics enabling them to create new motions in a matter of days. While it is necessary to pay attention to the detailed shape of objects when grasping them, once they are picked up, only the mass and inertia matter, allowing for some generalization.

6) Robert discusses the strength and ability of their humanoid robot, Atlas, to jump onto a one-meter high box. He mentions that the next generation of Atlas will be much stronger than humans, to the point where a person would not be able to perform the same jump. Playter explains the testing process and how it has evolved from manual iteration to using model predictive control techniques. He explains that the robot is able to calculate its trajectory and make adjustments in real-time, allowing it to perform gymnastic tricks even in the air. Playter calls this evolution of robotics technology surreal, but also points out that there are still limitations due tounder actuated systems.

7) Robert discusses the challenges of working with fixed momentum in legged robotics, citing the difficulties of changing direction mid-air. He describes the company's process of "build it, break it, fix it" to improve robot designs and break down weak points. Playter also reflects on the DARPA Robotics Challenge in 2015, where Atlas, their first-generation humanoid robot, was tasked with crawling over stairs and rough terrain, identifying vowels, driving vehicles, and manipulating tools. Despite the challenge's tragic outcomes, Playter acknowledges that it pushed an entire industry forward and helped the company learn invaluable lessons in robotics.

8) Robert discusses the different layers of control loops involved in managing a robot, from high-speed actuator control to larger physics models and sensory inputs like vision. He emphasizes the importance of efficient and fast operation for real-time control, which becomes more feasible as computation power increases. The company uses physics-based simulation tools to test and develop algorithms and models to ensure the same code works in simulation and on the hardware of the robot. Playter notes that the challenge in developing these tools and efficiently running simulations was modeling interactions with the environment and developing simulations that could run fast while also capturing critical elements of this interaction.

9) Robert discusses the challenges of implementing dexterous manipulation in robots, which goes beyond foot-ground contact and involves modeling various types of surfaces to capture slipping and moving objects. Playter mentions that Atlas, Boston Dynamics' humanoid robot, has not been tested on a beach yet but has walked on various surfaces, including rocks and hills. He also talks about Big Dog, one of Boston Dynamics' iconic robots, and how loose rocks became a challenging walking surface that they had to develop a test for.

10) Robert discusses how people tend to anthropomorphize robots, attributing levels of intelligence that the machines do not possess. He acknowledges this tendency and incorporates it into the design of the robots, adding a level of personality and making them more entertaining. Playter also shares his personal interest in legged robots and communicating emotion through body movement, which he believes can enhance the interaction between humans and robots. He also emphasizes the importance of developing a way for the robot to communicate physically with humans, indicating its behavior or needs, as a way of making it more usable andreliable for customers.

11) Robert discusses the origins of Spot and the challenges of transitioning from a research and development company to a commercial company. During their time at Google, Larry Page asked them what their product was going to be, and they decided on a walking quadruped that was smaller and electrically actuated. While Page wanted a consumer-level product that was inexpensive, Boston Dynamics believed that building the best hardware was important for learning about the important parts of making robots work. Now that Boston Dynamics is building robots for the industrial sector, they can afford to make more expensive machines, which can help them learn how to simplify and eventually lead to a consumer-level product that is cheaper.

12) Playter discusses the transition from research to manufacturing and the challenges they have faced with reliability and cost. They have set up a manufacturing facility and are working on issues of cost-cutting by learning to cast parts, control the molding process, and build the same robot twice in a row. However, iterating on the design – as they find failures from reliability tests – has been challenging, and getting everyone on the same page to move fast while also reducing costs is a challenge they are facing during this migration. Despite these challenges, there is still room for simplification, and Boston Dynamics is already thinking about what the next generation of their robot, Spot, will look like.

13) Robert discusses the possible uses of purpose-built machines and the challenges the company faced in designing and manufacturing Spot's arm. Playter believes that the future of mobile robots lies in mobile manipulation, which involves getting a mobile robot to do useful manipulation autonomously. To make Spot more reliable and less expensive, Boston Dynamics plans to build a next-generation machine that specifically caters to the needs of factory robots with integrated sensors. Boston Dynamics also designed Spot's arm to experiment with the problems faced in mobile manipulation. The arm is almost as complex as the robot itself and can be attached as a payload. The company intends to offer the ability of autonomous action with the help of the arm through a tablet interface.

14) Robert discusses automating industrial tasks usingtheir robots. He mentions how they automated a task for Ontario PowerGeneration, where the robot was able to control the breaker switch withouthuman intervention. Playter notes that they love working with customers whopresent challenging problems and challenges motivate them into solving the nextlogical step. However, the challenging part is finding applications broadenough to sell thousands of robots. Playter concludes by stating that theybelieve they can make money in the industrial monitoring space and becommercially successful if the robot can prevent costly production losses.

15) Robert discusses the company's focus on developing robots for industrial use in preventing unintended shutdowns. Using athermal camera, acoustic imager, visual cameras, and vibration measurements, the robots can be used in industries such as beer factories, chip fabs,electric utilities, and nuclear power plants. Playter also talks about the company's second robot, Stretch, which is a mobile robot designed to move boxes around warehouses. With over a hundred spots and several customers already committed to buying between 10 and 20 robots, Playter believes that Stretch has the potential to be profitable, particularly given the need to mechanize box moving tasks in the warehouse.

16) Robert discusses the possibility of social robots like Spot being used in the home as pets. While the company does see afuture in this area, they believe that the pathway to getting there is likely through industrial applications and mass manufacturing. Playter emphasizes theimportance of functionality and performance first and believes that social robots need to perform useful tasks before adding on additional features likepet-like interactions. While he acknowledges the potential for robots to become companions as seen with the Sony IBO, Boston Dynamics is currently focused on building a foundation of utility before expanding into areas like AI-generated conversation.

17) Robert discusses the potential for robots to communicate with humans through verbal conversation and gestures, using learning models and generative AI. He believes that robots will eventually provide a warm feeling of connection and friendship, building memories and creating a foundation for shared experiences. Playter acknowledges that personalized robotic communication is a technical problem, not a limitation, and that embodied robots like Spot have the potential to fulfill this role.

18) Robert discusses the exciting energy in the company, fueled by recent competition and goals of pushing Atlas to do more cool stuff while driving costs down. The next phase of Atlas will focus on more dexterous hands that can manipulate and grab more things, starting with heavy items that affect balance. Playter sees the potential for two-armed robots, like Atlas, in manufacturing and construction environments, where they can transfer objects from one hand to the other, reorient them in ways that one hand cannot, and accomplish tasks that require a higher level of dexterity. Playter also notes that scaling and building efficiencies are key to driving down costs, something the company can learn from the automotive industry.

19) Robert discusses the Boston Dynamics AI Institute, its mission, and its research goals. The AI Institute, led by Boston Dynamics founder Mark Rayburn, is a separate organization that focuses purely on unadulterated research. While Boston Dynamics will continue to commercialize and do research in the next five years, the AI Institute will tackle over-the-horizon problems. Playter also touches on people's fear of humanoid robots, stating that we have been taught to fear them through fiction for over a hundred years. However, he believes that it's much less threatening to have a super intelligent AI embodied in a robot as it is visible and verifiable in physical reality.

20) Robert discusses the company's decision to co-sign a letter stating that they won't put weapons on their robots. The motivation behind this decision is that as robots become more widely available, people have found ways to attach guns and operate them remotely, which is a justifiably scary prospect. Boston Dynamics wants robots to ultimately be helpful and make lives more fulfilled and productive, and they believe that it's better for the industry as a whole to draw a bright line and ban weapons on robots. Playter also talks about what it takes to lead a company and hire a great team, highlighting the importance of courage and adaptability.

21) Robert discusses the hiring and interview process at Boston Dynamics and compares it to that of Google. While Boston Dynamics used to require candidates to give presentations to showcase their passion and problem-solving skills, they have since scaled back on this process as they look to increase their rate of hiring. Playter emphasizes the importance of finding candidates who have a genuine passion for their work and can speak to the details of their projects. Playter also briefly touches on the topic of consciousness in robots, stating that it is fairly remote for the machines Boston Dynamics currently works with and that the anthropomorphization of robots is largely due to physical gestures that are similar to those of humans and animals.

22) Playter discusses his doubts about whether large language models like GPT can simulate consciousness by stringing words together, believing that there is a different kind of knowledge associated with truth and emotions that contributes to the growth and evolution of sentient beings. However, he acknowledges that language models will be able to simulate consciousness and that GPT-4 when fine-tuned correctly, would be able to do exactly that, raising ethical questions on accurate labeling of information and dealing with disinformation. Playter also emphasizes that embodiment and the connection between human and robot are crucial and holds promise.

23) Robert gives advice to young people on how to have a career or life they can be proud of. He advises them to follow their curiosity, love what they are doing, and be willing to make decisions and pivot if they see something exciting. He envisions a future where robots will be ubiquitous and doing some of the hard labor, but he believes that people need to work to feel productive and satisfied. Therefore, he hopes that robots and computers will not be able to do all the creative work, as that is what gives humans the feeling of satisfaction and accomplishment. Instead, he envisions robots and humans working together in a complementary fashion, where humans keep their faculties and physical health while doing some labor.

24) Robert discusses his optimism for a future where some robots can be pets and create an emotional connection with humans. Playter believes that companionship and friendship are what make for a better and happier life, and he sees no reason why some of these friends can't be robots.He acknowledges that machines may never have true intelligence or sentience but adds that they could simulate it and serve as companions that collect and remember individual histories. Playter subtly hints that some of his best friends are already robots, and he thanks Boston Dynamics for leading the way in robotics engineering, specifically in accessible and elegant movement.