Introduction

> Achieving a natural-looking gait for our humanoid robots has been a long and demanding journey, taking us around 15 years to perfect. From our early experimental stages with the PETMAN prototype in 2008 to finally seeing the fluid, human-like walking in Atlas last year, it’s been both a technical challenge and an evolving learning process, where new control algorithms unexpectedly brought us closer to our goal.

> The iterative development at Boston Dynamics, encompassing innovative projects like Atlas and Spot, reflects our dedication to blending elegance and functionality in robotics. Seeing our creations not just walk, but also perform complex actions like dancing and backflips, is a testament to the relentless effort and creativity of our team over the decades, originating from our formative work back at the MIT Leg Lab.

Early days of Boston Dynamics

> I fell in love with robotics when I saw a robot do a somersault during a visit to MIT; the connection to my gymnastics background made it clear to me that movement was the heart of the fascination. The quest for elegance and efficiency in robotics mirrors the gymnastics philosophy of “letting your body do what it wants to do”—an intrinsic understanding of physics that empowers both the body and the robot to move naturally and gracefully.

> One lasting lesson I learned from Marc Raibert is to pursue what truly engages you—when you love what you do, everything gets better, and that passion creates a fulfilling environment for everyone involved. It’s fascinating how this pursuit brings joy to roboticists, akin to the happiness I observe in skydive enthusiasts, fueled by the magic of bringing robots to life after overcoming challenges.

> Finally, tackling complex problems like legged locomotion requires courage and a willingness to simplify them to their core essence; starting with something as basic as a pogo stick helped us solve fundamental issues. Marc taught me to seek out and chase big challenges driven by curiosity, rather than just the prospect of profit, which has been a guiding principle in my career ever since.

Simplifying robots

> I remember the early days at Boston Dynamics, where we had to pioneer understanding the central principles behind building robots and creating software around them. It required real expertise to develop feedback-control algorithms that could run in real-time and make the hardware work.

> Breaking robots to learn, repair, and repeat became a core principle at the company. It allowed us to be fearless in our work, understanding that sometimes you have to break things to make progress and innovation possible.

Art and science of robotics

> The development of dynamic movement in robots involves a blend of both scientific principles and artistic intuition. While principles like the spring-mass system are crucial for transitioning from one-legged to multi-legged robots, humanoid robots leverage an innate human intuition about how they should move. This intuition taps into a deeper, almost artistic sense, akin to how athletes and coaches perceive and refine movements.

> Achieving truly elegant movement in robots requires embracing dynamic stability and natural gaits. Early robots struggled with awkward, inefficient motions because they focused on preventing falls rather than harmonizing with the physics of movement. By allowing robots like Atlas to "fall" and catch themselves naturally, we create machines that not only appear more lifelike and beautiful but are also more efficient and stable, capable of walking stably over various conditions.

Atlas humanoid robot

> Achieving natural walking in robots is a complex challenge that took over a decade of development. "I probably didn't really see the nice, natural walking that I expected out of our humanoids until maybe last year," which reflects the painstaking process of refining control algorithms for fluid movement.

> The complexities of humanoid robots, especially with their mass and inertia, can complicate locomotion. "Dealing with all of that interaction does make the humanoid a much more complicated platform," highlighting the intricacies of balancing movement with gravity and weight distribution.

> Recent advancements, like model-predictive control, are revolutionizing how robots handle movements and interactions with objects. Now, “we're starting to develop the tools that let us do that in a matter of days,” showcasing a shift towards rapid prototyping and improved agility in robotics.

> Pushing the envelope, like attempting backflips, requires the courage to experiment and learn from failures. "By breaking the robot repeatedly, you find the weak points, and then you end up redesigning it so it doesn't break so easily next time," emphasizes a philosophy that encourages innovation through trial and error.

DARPA Robotics Challenge

> The DARPA Robotics Challenge was a significant event, humbling us all with the difficulty of getting robots to perform seemingly simple human tasks. Our first-generation Atlas robots were part of this challenge, highlighting just how hard it is to create a general-purpose robot that can handle a variety of activities autonomously.

> Building general-purpose robots is fundamentally challenging due to the need for them to navigate and manipulate in unpredictable environments. We learned that while specific-task robots can excel in controlled settings, achieving versatility in robots involves intricate feedback loops and real-time adjustments, which demand immense computational power.

> A critical part of our development process has been the creation of sophisticated physics-based simulation tools. These tools, originating from MIT, are essential for testing and iterating on robot designs efficiently. Even though simulating real-world interactions like foot-ground contact or dextrous manipulation is complex, these simulations are vital for refining our robots' capabilities before actual deployment.

BigDog robot

> One thing that stands out to me is the challenge of testing robots in varying environments, like walking on sand or rocks. Saltwater, for example, caused corrosion issues for robots like BigDog when not properly cleaned after exposure. This highlights the importance of considering environmental factors in robot design and maintenance.

> A key aspect of Boston Dynamics' work has been pushing the boundaries of legged robot capabilities, starting with BigDog. The early versions faced challenges with stabilizing and managing weight, using gas-powered engines that were loud and not optimized. Despite these obstacles, they were able to progress to walking on rough terrain and demonstrating the feasibility of legged locomotion.

> Exploring the integration of emotion communication in robots through body language is an intriguing notion. By enabling robots to express feelings like excitement or fear through movement, it could enhance human-robot interactions and usability in various applications. Boston Dynamics is evolving to enhance robot usability beyond just locomotion, allowing for clearer communication with operators in complex environments.

Spot robot

> Spot's evolution began with the shift to Google to continue progress in robotics, aiming for consumer-level products, but now focuses on industrial robots due to cost challenges for home use.

> Transitioning from R&D to commercialization brought key challenges in ensuring quality, reliability, and cost effectiveness, but the team excelled by leveraging their engineers' problem-solving mindset.

> Boston Dynamics' focus on manufacturing enhancements such as part casting and cost reduction processes reflects a continuous learning journey towards improving efficiency and scalability while iterating on robot design.

> The innovative inclusion of a manipulator arm on Spot highlights the company's direction towards mobile manipulation, enabling autonomous tasks like door opening and breaker operation, showcasing the potential of advanced robots for industrial applications.

Stretch robot

> "Stretch started with a clear vision—we recognized there’s a trillion boxes shipped around the world each year, mostly moved by hand." This realization led us to create a mobile robot that can efficiently handle box-moving tasks in warehouses, which is something that the industry desperately needs.

> "We’re already seeing the enthusiasm from customers, with commitments for hundreds of robots before we even finished shipping." The early demand for Stretch has been incredible, proving that we’re tapping into a significant market opportunity that will transform operations in warehouses everywhere.

Handle robot

> Creating Handle was a fascinating journey in understanding and mastering the balance and coordination of robotics. We built this sci-fi-inspired robot with wheels on legs, an arm, and a tail, moving in a synchronized dance. It was a beautiful machine but, in practical terms, too slow for the logistics applications we had in mind. This experience taught us the balancing aspect deeply, but we had to pivot to more efficient designs.

> Transitioning from Handle to creating Stretch marked a significant evolution. Stretch, with its heavy base and omnidirectional movement, was optimized for warehouse efficiency, capable of operating for 16 hours and handling up to 50-pound boxes. This robot addresses real-world challenges like unloading trucks, transforming a backbreaking manual job into a role where workers become robot operators. Building Stretch illustrates our shift to focus on practical, sustainable business solutions that enhance productivity and worker safety.

Robots in our homes

> - The key to creating successful social robots like Spot lies in prioritizing utility over cuteness, focusing on making robots perform useful tasks effectively first.

>

> - Building a consumer-level social robot goes beyond just cuteness; it involves complex tasks like distinguishing between items, cleaning up, and ensuring safety, especially in interactions with children.

> - While performance and utility are crucial foundations for robot development, the potential for robots to serve as companions with emotional connections, like the Sony Aibo, should not be underestimated, opening up possibilities for meaningful interactions beyond utility.

Tesla Optimus robot

> The excitement surrounding humanoid robots, particularly with Elon's commitment to innovation, has shed a bright light on our work at Boston Dynamics, invigorating "the competitive juices" and validating over a decade of effort in robotics.

> Our recent Atlas video demonstrated that we’re not just focused on dexterous tasks; we aim "to show that we can pick up and move heavy things," marrying mobility with the ability to handle weight and balance effectively, which is essential for manufacturing environments.

> We're moving toward a future where external developers can easily enhance our robots' capabilities with new behaviors and machine learning applications, creating "an ecosystem of providers," while we simultaneously leverage reinforcement learning to improve locomotion, manipulation, and perception across our systems.

ChatGPT

> The advancements in transformer neural networks, like GPT-4, are both exciting and concerning. Disinformation is a significant risk these tools could exacerbate, but their application in robotics, like judging task performance, provides a verifiable check that mitigates these concerns.

> Integrating language models with robots such as Spot is thrilling. Enhancing communication through verbal commands increases the bandwidth and variety of interaction, making technology like the demo with Levatas, where they talked to Spot to give commands, truly exciting and showcasing practical advancements.

Boston Dynamics AI Institute

> Marc Raibert, founder of Boston Dynamics, established the Boston Dynamics AI Institute to focus on "pure research without the confinement or demands of commercialization", allowing for deeper exploration into long-term, over-the-horizon problems in AI and robotics.

> While Boston Dynamics continues to focus on practical research with a commercial outlook within a five-year perspective, the AI Institute aims to push the boundaries of innovation even further, tackling multi-decade challenges and advancing the field in unprecedented ways.

Fear of robots

> The fear surrounding legged robots often stems from a century of cultural conditioning; as I explained, “we’ve been taught to be afraid for over 100 years,” and that fiction often dictates our perceptions more than reality does.

> Technological transformations have historically led to job anxieties, but I believe “this is the right technology at the right time,” especially as we face a shrinking workforce and an increasing need for efficiency—much like when we transitioned from hand tools to machines in farming.

> My commitment to ensuring the responsible development of robotics is strong; by co-authoring a letter against weaponizing robots, I emphasized that “we thought it was important to draw a bright line,” as fostering trust in technology is vital for it to thrive and genuinely enhance our lives.

Running a company

> Courage and adaptability have been essential in my journey from engineer to CEO. Leading a team of brilliant individuals required the courage to trust myself and tackle any challenge that arose. The job has evolved over the years, and being open to shifting roles and responsibilities was crucial.

> Our interview process focuses on finding passionate individuals who truly care about their work. Asking probing questions to understand their expertise and seeing their genuine excitement helps us identify the right fit for our team. It's about tapping into their passion and talent to build a strong team dedicated to tackling the toughest challenges in the world of engineering.

Consciousness

> The discussion around consciousness in robots and large language models is fascinating. I doubt that simulating consciousness by stringing together statistically associated words truly captures the essence of knowledge. Truth and emotions are a different kind of knowledge that is absolute and unique.

> Large language models like GPT seem to excel in presenting nuanced perspectives, especially in controversial topics. The ability of these models to show confusion and a childlike element of humility raises important ethical questions about how we interact with AI and the potential manipulation of human emotions.

Advice for young people

> Follow your heart and your curiosity—it’s the key to having a fulfilling career. When you truly love what you’re doing, it just becomes more enjoyable and makes you better at it. Don’t stress too much about meticulously planning your future; embrace the "happy mistakes" that come along the way and let your interests guide you.

> Life is full of unexpected opportunities, and being open to change can lead to exciting directions. Like my encounter with Marc in the AI lab, that moment of realization—"Oh, boy, I gotta go do this"—can steer you towards amazing paths you never anticipated.

Future of robots

> Robots will become an integral part of our lives, complementing human labor rather than replacing it. It’s essential that we continue to engage in meaningful work because “self-satisfaction and feeling productive is such an ingrained part of being human” — I believe it keeps us healthy and fulfilled.

> I also see a future where robots can address human loneliness by forming emotional connections; companionship from machines could enrich our lives. There's a compelling study that shows “companionship and friendship are the things that make for a better and happier life,” and I think robots can play a role in offering that connection, even if it’s through simulated intelligence.