Key Moments
Boris Sofman: Waymo, Cozmo, Self-Driving Cars, and the Future of Robotics | Lex Fridman Podcast #241
Lex FridmanKey Moments
Boris Sofman discusses Anki's social robots, human-robot interaction, the challenges of robotics, and Waymo's autonomous trucking.
Key Insights
Simplicity and character-driven design were key to Cozmo's success, allowing for emotional connection despite minimal physical form.
Hardware and software constraints, combined with the power of software, enabled Anki to create affordable, intelligent consumer robots.
The character aspect is crucial for human-robot interaction, fostering empathy and forgiveness for mistakes, extending beyond entertainment.
Autonomous trucking aims to address truck driver shortages, enhance safety, reduce costs, and revolutionize logistics networks.
Waymo's L4 autonomous driving focuses on specific, controlled environments (like freeways and transfer hubs) to prove safety and enable gradual expansion.
The future of robotics will see increasing collaboration, with AI and machine learning driving advancements in perception, planning, and evaluation, ultimately leading to widespread societal adoption.
THE ALLURE OF CHARACTER IN ROBOTICS
Boris Sofman, co-founder of Anki and head of trucking at Waymo, begins by expressing his admiration for fictional robots like Wall-E and R2-D2, highlighting their ability to convey emotion without language. This fascination with emotional interaction significantly influenced the design of Cozmo, Anki's social robot. Sofman notes that Anki intentionally moved away from humanoid forms, finding that simpler designs, similar to cartoon characters, allow for more profound emotional communication. By limiting physical complexity, engineers and animators are forced to focus on the nuances of personality and interaction, leading to a richer, more engaging user experience.
ANKI'S VISION: BRINGING PIXAR CHARACTERS TO LIFE
Anki was founded by three PhD students from Carnegie Mellon's Robotics Institute, driven by a passion for applying robotics and AI to consumer applications. They recognized a gap in the entertainment space between rich video game and movie experiences and the limited interactivity of traditional toys. Leveraging plummeting component costs due to the mobile phone industry, Anki aimed to shift complexity from hardware to software, using inexpensive components to power sophisticated AI. Cozmo, their second product, epitomized this vision: a physical character with an understanding of its surroundings, capable of emotional connections typically found only in movies, much like bringing a Pixar character into the real world. Their first product, 'Drive,' was a physical racing game that brought virtual gaming concepts into the physical world, revealing the amplified impact of physical interaction compared to virtual experiences.
THE GENIUS OF SIMPLICITY: COZMO'S DESIGN AND IMPACT
Cozmo's design success stemmed from a radical simplification of hardware. With only four degrees of freedom (two treads, a head that moves up and down, and a lift arm), sound, and a low-resolution screen, Cozmo managed to convey a wide range of emotions and intentions. This minimalist approach was a deliberate choice, contrasting with traditional roboticists who often strive for complex, human-like forms. Anki's collaboration with experts from Pixar and low-cost manufacturing specialists enabled them to achieve a sub-$200 retail price. This affordability and the robot's endearing personality made Cozmo accessible and compelling. Sofman emphasizes that perfection isn't fun in robotics; intentional 'noise' and imperfections in Cozmo's behavior made it more endearing and fostered empathy from users, creating a feedback loop where even mistakes strengthened the human-robot bond.
THE CHARACTER-DRIVEN APPROACH TO HUMAN-ROBOT INTERACTION
Anki's character-driven approach prioritized emotional connection over gaming mechanics, recognizing that physical characters have a far greater impact than virtual ones. They developed a behavioral engine that interpreted real-world context and generated parameterized emotional responses. This allowed Cozmo to react spontaneously and intelligently, showing happiness, frustration, or boredom. Simple actions, like eye contact, significantly increased user engagement, demonstrating the power of nuanced emotional cues. The games Cozmo played were designed not as ends in themselves, but as contexts to trigger and showcase the robot's personality, creating scenarios for genuine emotional responses. This character-first philosophy ultimately fosters human-robot empathy, turning robot mistakes from perceived failures into endearing qualities.
THE CHALLENGES AND THE DEMISE OF ANKI
Despite Cozmo's critical and commercial success—selling millions of units and approaching $100 million in annual revenue—Anki faced significant business hurdles. The company's revenue was highly seasonal, with 85% concentrated in the fourth quarter, making cash flow management extremely difficult. Expensive marketing, manufacturing costs, and long payment cycles from retailers led to brutal financial swings. The hardware sector's declining valuations further complicated fundraising efforts. Anki's attempt to transition beyond entertainment into other categories while maintaining its core business proved unsustainable. A failed financing round ultimately led to the company's shutdown in 2019, leaving a profound emotional impact on Sofman and his team, who were proud of their accomplishments but victims of unfavorable market dynamics and the inherent difficulties of consumer robotics.
LOOKING AHEAD: THE FUTURE OF SOCIAL ROBOTS
Sofman believes that companion robots, akin to pets, will arrive much sooner than human-like AI, as the latter poses immense technical and societal challenges (e.g., the 'uncanny valley'). He highlights that for effective human-robot interaction, embracing the concept of a 'character' is more flexible and appealing than striving for human resemblance or voice. Cozmo and the subsequent Vector robot demonstrated that a physical character could proactively initiate interactions and disarm privacy concerns more effectively than passive voice assistants. While acknowledging the complexity of creating fulfilling relationships, Sofman stresses the need for more intelligent robots. He expresses concern that the unique, character-driven approach pioneered by Anki hasn't been widely adopted, but predicts that future technological advancements will eventually reignite this space, forming the basis of trillion-dollar opportunities.
WAYMO'S MISSION: AUTONOMOUS DRIVING ACROSS DOMAINS
After Anki, Sofman joined Waymo as the Head of Trucking, bringing his robotics expertise to autonomous vehicles. Waymo's core mission is to develop the 'Waymo Driver,' a universal autonomous driving system applicable across various environments, vehicle platforms, and use cases. Originating from Google's self-driving car project, Waymo now focuses on both consumer transportation (Waymo One, active in Phoenix) and logistics (Waymo Via, for autonomous trucking). The underlying technology—perception, decision-making, and safety—is largely shared across these applications. This multi-product strategy allows for significant leverage, as core R&D advancements benefit both passenger cars and heavy-duty trucks, despite their unique challenges, such as the dynamic limits and blind spots of articulated trailers.
THE VISION FOR WAYMO VIA: REINVENTING LOGISTICS
Waymo Via aims to develop Level 4 autonomy for Class 8 trucks, specifically 53-foot trailers, which dominate goods transportation. The long-term vision involves expanding routes, vehicle types, and operating conditions, moving beyond initial freeway-focused operations. The growing shortage of truck drivers (projected to reach hundreds of thousands), increasing e-commerce demand, and significant safety concerns (especially fatigue) highlight the critical need for autonomous trucking. Beyond cost savings, autonomous trucks can fundamentally reshape logistics by removing human-imposed constraints like daily driving limits. This enables longer, more efficient hauls and could lead to a complete reinvention of the supply chain, from warehouse locations to goods distribution, creating a more agile and responsive network.
TACKLING THE CHALLENGES OF AUTONOMOUS TRUCKING
Autonomous trucking presents a mix of easier and harder problems compared to passenger cars. Freeway driving, a primary focus, offers more structure and consistency than surface streets, simplifying the AI challenge. Waymo is strategically establishing transfer hubs at metropolitan entry points, such as an upcoming facility in Dallas, to facilitate automated hub-to-hub operations, deferring the complex 'last-mile' surface street problem. This phased approach, prioritizing market entry, allows Waymo to gather crucial real-world data and iterate. Level 4 autonomy means fully driverless operation within specified domains and conditions. This initial deployment strategy emphasizes proving safety and value in well-defined areas, with future expansion into more complex scenarios driven by gathered experience and refined technology.
SENSOR SUITES AND DATA FUSION FOR TRUCKS
Autonomous trucks require a more extensive sensor suite than cars, roughly twice as many. Waymo's fifth-generation hardware stack incorporates dozens of cameras, multiple radars, and proprietary lidars. A key design adaptation for trucks involves placing two main sensor pods on the cabin's outer sides (where mirrors would be) instead of a single central roof pod, to avoid occlusion from the higher trailer. This configuration improves visibility, offers redundancy, and addresses unique truck blind spots. While lidar excels in near-range, consistent 3D perception, its range limitations make radar and cameras crucial for long-range detection (beyond a kilometer). Waymo emphasizes 'early fusion,' integrating data from all sensors at the earliest possible stage to leverage their complementary strengths and weaknesses, especially for managing challenging conditions like weather and sparse freeway events, and to enhance machine learning model performance.
THE CRITICAL ROLE OF MACHINE LEARNING AND SIMULATION
Machine learning (ML) is an increasingly central component of autonomous driving, revolutionizing perception, behavior prediction, and planning. Waymo leverages its vast existing data from passenger cars to pre-train truck models (e.g., 85% of vehicle behavior prediction data comes from cars). This cross-domain data reuse significantly reduces the amount of new data required for trucking. Simulation plays a massive role, with Waymo simulating about 1,000 miles for every physical mile driven. Simulation is used for development, regression testing, replaying real-world edge cases, and, crucially, for large-scale safety validation. The ultimate metric is human-relative safety: proving the autonomous system is significantly safer than a human driver, especially for rare, high-severity events that are almost impossible to test sufficiently in the real world. This requires sophisticated, multi-faceted evaluation strategies, including structured testing and data augmentation, to efficiently explore the infinite state space of driving scenarios.
WAYMO VS. TESLA: DIFFERENT PATHS TO AUTONOMY
Sofman contrasts Waymo's approach with Tesla's. Waymo's advantage lies in its singular focus on Level 4 autonomy from the outset, with purpose-built hardware, sensor suites (lidar, radar, cameras), compute, and a comprehensive safety framework. This specialized investment, refined over generations, enables Waymo to develop a fundamentally driverless system. Waymo's current operation of a fully driverless ride-sharing service in Phoenix provides invaluable real-world learnings that cannot be shortcut. In contrast, Tesla's strength lies in its massive data collection from a large fleet of human-driven vehicles. While a camera-only, end-to-end ML approach is theoretically capable of solving autonomous driving as humans do, Sofman views this as a higher-risk path, as it adds significant complexity to an already incredibly difficult problem, especially regarding long-tail safety cases and the robustness of computer vision in all conditions. The timeline to full driverless capability remains the key differentiator between these two prominent approaches.
THE SOCIETAL IMPACT OF AUTONOMOUS VEHICLES
Autonomous vehicles, particularly trucks, hold the potential to profoundly reshape society. Reduced commuting pain could alter urban planning and living patterns, while optimized logistics will reduce congestion and parking demand. The environmental benefits, from reduced emissions to decreased accidents (95% of which are currently human-error related), are substantial. While acknowledging concerns about job displacement, Sofman points to the existing shortage of truck drivers and the unpleasant nature of long-haul routes as areas where automation can fill a critical need and improve human quality of life. Automation will likely shift demand to more localized, quality human jobs and create entirely new support roles (e.g., remote assistance). Ultimately, the increased efficiency from autonomous transport will spur economic growth and generate new industries, leading to a net positive societal impact, with a cautious approach to managing transitional pains.
ADVICE FOR ASPIRING ROBOTICISTS AND STARTUP FOUNDERS
For those aspiring to careers in AI and robotics, Sofman stresses the importance of passion, emphasizing that sustained effort is nearly impossible without it. He advises finding the intersection of passion with a growing market need, like machine learning, which offers versatility across many industries. A PhD in robotics, while demanding, provides a broad foundation in AI, ML, computer vision, hardware, and systems, fostering collaboration across disciplines. For startup founders, he draws on Anki's lessons: be prepared for intense emotional swings and the 'gauntlet' of fundraising, manufacturing, and market education. Co-founders are vital for support. Crucially, understanding the market and timing is paramount; even brilliant teams can fail if they misjudge product-market fit or face unavoidable industry headwinds. While creating novel categories is possible, anchoring on familiar needs (like a 'smart thermostat' vs. an entirely new 'home robot') can ease market adoption and communication. The challenge lies in justifying price points with compelling functionality for the mass market.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●People Referenced
Common Questions
Anki was a robotics company co-founded by Boris Sofman with a vision to merge rich video game experiences with physical world interaction. They aimed to create intelligent robots that could leverage software to deliver value and experiences previously impossible with traditional robotics. Their first major commercial product, Cozmo, embodied this by bringing a Pixar-like character to life.
Topics
Mentioned in this video
Senior Director of Engineering and Head of Trucking at Waymo, co-founder and former CEO of Anki. He discusses his work on social robots and autonomous vehicles.
Mentioned as a person who designs IQ tests for machines, focusing on handcrafted validation sets for general intelligence.
CEO of Tesla, mentioned for his focus on the humanoid form for the Tesla Bot from a factory-efficiency perspective, a point of disagreement with Sofman who sees it as research.
Mentioned by the host as someone with whom he discussed the possibility of building a humanoid robot cheaply.
Mentioned as a previous guest on the Lex Fridman Podcast who discussed the human side of trucking.
A popular robot from Star Wars, cited by Boris Sofman as a favorite for its emotional expressiveness without language.
The actor who played the Terminator, mentioned by Boris Sofman.
Character director from Pixar who worked on 'Wall-E' and partnered with Anki to develop Cozmo's character and animation, bringing a deep understanding of emotional magic to robotics.
The computer scientist who proposed the Turing Test, referenced by the host when discussing how to 'test' an autonomous truck's intelligence.
A science fiction writer whose quote about robots and love concludes the podcast.
The company that originally started the self-driving car project that eventually spun out to become Waymo.
A consumer electronics company, whose former head of mechanical engineering joined Anki, bringing expertise in low-cost, high-volume consumer product manufacturing.
An animation studio, whose character design principles and film 'Wall-E' heavily influenced Anki's creation of Cozmo. Some of Anki's team members previously worked at Pixar.
Mentioned as a company that was on the verge of failure multiple times before succeeding, drawing a parallel to Anki's challenges.
The company that acquired Cozmo's assets after Anki's shutdown, continuing the product line.
A company building incredible robots (autonomous vehicles), whose distinct technical and philosophical approach (vision-only, large human-operated fleet) is compared to Waymo's.
An autonomous vehicle company, formerly the Google self-driving car project, where Boris Sofman currently works as Senior Director of Engineering and Head of Trucking.
Robotics company acquired by Amazon, known for its orange robots that automated warehouse picking and collecting previously, described as a 'favorite application of robotics'.
A company that had 'beautiful products' and found great success by reinventing existing categories like the thermostat, making marketing messaging clear.
Its success with products like the iPhone is contrasted with other companies, highlighting its meticulous design and launch approach. Also mentioned as a company that has fewer 'misfires' than Amazon.
A partner with Waymo, providing their trucks (Freightliners) for integration with the Waymo driver system.
Robotics company known for its legged robots; admired for their technical achievements in walking and jumping, though finding practical applications is noted as a long journey.
Mentioned for its Kiva Systems robots in warehouses and its Astro home robot. Identified as having deep pockets and a willingness to iterate aggressively and take risks.
A robotics company cited as an example of successful pivot and thriving by focusing on a profitable product (Roomba vacuum cleaner) to fund other ventures.
A science fiction robot from a movie, cited as a favorite by Boris Sofman for its ability to convey emotion and character without language. Anki studied it deeply as inspiration for Cozmo.
A movie about a human-AI relationship, brought up by the host to discuss the future of AI companions. Boris notes such advanced human-like AI is still 'complete R&D'.
A science fiction movie series mentioned by Boris Sofman, highlighting the dynamic range of AI and robotics, a 'dark world' but quite fascinating from an engineering perspective.
A video game mentioned as an example of characters without voice, which helps broaden their appeal across demographics.
An animated show whose creators partnered with Dream Labs to develop a 'Butter Robot'. Boris Sofman praises the show's brilliance and the 'existential angst' portrayed by the robot.
A fictional robot from Rick and Morty, whose character and existential questioning are admired. Dream Labs is developing a physical version, posing significant manufacturing and AI challenges.
A toy robot created by Anki, known for its emotional intelligence and engaging human-robot interaction, described by the host as doing 'exactly what Wall-E does'.
Its emergence in 2007 spurred a plummet in component costs and became an 'incredible interface device' and 'brain' for physical world experiences, enabling robotics advancements like Anki's products.
A humanoid robot being developed by Tesla. Sofman views humanoid robotics as primarily research, questioning its immediate application justification, especially in factory environments.
Anki's first product, a physical racing experience with virtual game elements, which served as a proving ground for human-robot interaction and led to insights applied to Cozmo.
A robotic vacuum cleaner manufactured by iRobot, noted for its strong product-market fit due to solving a clear need (annoyance of vacuuming) with a familiar concept.
An augmented reality headset, mentioned as an example of technology starting in enterprise and moving towards consumers.
A robotics company co-founded by Boris Sofman that created Cozmo, a social robot with emotional intelligence, which he describes as one of the most incredible social robots ever built.
Google's smart speaker, similar to Alexa, which people generally do not tolerate initiating conversations, used to illustrate the importance of character in proactive interaction.
A future version of Anki's social robot, described as always-on and allowing for non-intrusive initiation of interaction in a way voice assistants cannot.
Amazon's voice assistant, used as an example of a device that people generally do not tolerate initiating conversations proactively, contrasting it with character-driven robots like Cozmo.
A virtual office assistant mentioned by the host as an early, less successful attempt at character-driven AI interaction.
Specific European country mentioned for its high privacy bar, where Anki was worried about launching its robots.
Specific European country mentioned for its high privacy bar, where Anki was worried about launching its robots.
Specific European country where Anki launched its products.
A specific location within the Phoenix metropolitan area where Waymo started its driverless operations, expanding from there.
A major hub for Waymo Via's autonomous trucking, with a new facility being built for testing routes to Houston, Phoenix, and Austin.
A city on a key trucking route from Dallas that Waymo is targeting for autonomous operations.
The largest port in the United States, a huge opportunity for autonomous trucking due to the volume of goods moving east from there. Delays at the port due to labor shortages highlight current logistics inefficiencies.
Anki faced concerns about privacy regulations there regarding robots with cameras and microphones, but found that Cozmo's character disarmed these fears.
A city mentioned as part of Waymo's trucking routes from Dallas, and where Waymo's car side had previously operated.
Waymo's first fully autonomous service area for human transportation (Waymo One), serving as a crucial learning ground for driverless technology.
Another key city for Waymo's consumer transportation efforts, chosen for its dense urban challenges like pedestrians, which provide important learning opportunities.
Identified as one of the best places to start autonomous trucking operations due to volume, regulatory environment, and weather conditions.
More from Lex Fridman
View all 222 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