At CES 2026 in Las Vegas, Boston Dynamics and Google DeepMind announced a new artificial intelligence partnership for humanoid robots. The goal is to combine the athletic capabilities of Boston Dynamics robots with advanced artificial intelligence models from Google DeepMind. Work on the joint research will begin this year and will focus on integrating Gemini Robotics models into new Atlas robots. This will enable the robots to handle a wide range of industrial tasks, particularly in the automotive industry.
Boston Dynamics, known for its robots with impressive physical abilities, plans to use this partnership to turn humanoid robots into more than just preprogrammed machines. Google DeepMind is contributing its Gemini Robotics models, which enable robots to perceive their surroundings, reason, use tools, and interact with people. Carolina Parada, head of robotics at Google DeepMind, explained that the industry is only beginning to discover the potential of artificial intelligence. Robots learn through teleoperation, meaning remote control, and simulations rather than manual programming.
New Atlas robot
CES 2026 featured the first-ever public demonstration of Boston Dynamics' Atlas robot. Zack Jackowski, general manager of Atlas, presented the electric version of the robot, which was first unveiled in early 2024. The robot stood, walked, waved, turned, bent over, and waved again before leaving the stage. They later revealed a blue production version of the robot behind a curtain.
Marc Raibert, founder of Boston Dynamics and executive director of the Robotics and AI Institute, was also present at the event. His participation underscored the importance of the announcement, which also involved executives from Hyundai Motor Group. Hyundai acquired Boston Dynamics in December 2020 for $880 million. In 2025, the company deployed more than 500 robots, including the quadrupedal Spot and the Stretch trailer-loading system, and generated revenue of around $130 million, or approximately 2.86 billion Czech korunas.
Partnership details and robot training
The partnership with Google DeepMind is crucial to enabling humanoid robots to perform different tasks without preset routines. Boston Dynamics is increasingly relying on teleoperation, imitation learning, and simulations. Carolina Parada noted that robots are limited when tasks require sensors they do not have, but they can learn anything that can be consistently demonstrated through teleoperation. Collecting real-world data is expensive, so DeepMind is exploring methods that will allow robots to learn without having to see every object in advance.
Boston Dynamics recently completed a proof of concept for Hyundai involving parts sequencing. At CES, Hyundai outlined potential applications not only in the automotive industry but also in construction, energy, and facility management. To support training, the Robot Metaplant Application Center (RMAC) will open in 2026, where data will be collected from Hyundai factories. Wong Jae Lee, vice president of Hyundai's manufacturing division, said that no company has more real-world data than Hyundai Motor Group.
Boston Dynamics is also collaborating on artificial intelligence with the Toyota Research Institute and the Hyundai-funded Robotics and AI Institute. Robert Playter, CEO of Boston Dynamics, emphasized that the remaining automation work is challenging because of the variability of tasks, and this is precisely where artificial intelligence will help. Aaron Saunders, who stepped down as Boston Dynamics' chief technology officer in November 2025 after nearly two decades, joined Google DeepMind as vice president of robotics hardware engineering.
Atlas robot specifications
The new production version of Atlas, unveiled at CES, stands 1.89 meters tall, has a reach of 2.29 meters, fully rotational joints, and can lift up to 50 kilograms. It has 56 degrees of freedom, an IP67 rating, a new four-fingered hand, and a standard battery life of four hours with autonomous battery swapping. The robot can be controlled autonomously, by teleoperation through a VR headset, or using a tablet.
Boston Dynamics focused on simplifying the design. They reduced the number of motors to lower complexity, and all limbs can be replaced on-site within five minutes for easy maintenance. Each component is designed to minimize failures and allow easy upgrades. The components are compatible with automotive supply chains, which, thanks to Hyundai's support, will ensure reliability and cost-effectiveness.
All units for 2026 are already reserved for RMAC and Google DeepMind, with additional customers scheduled for 2027. Hyundai plans to produce up to 30,000 humanoid robots annually by 2028. Atlas is expected to handle highly precise, large-scale sequencing within two years and complex assembly tasks by 2030. Zack Jackowski described this generation as the most manufacturing-friendly they have ever designed. Robert Playter called it the best robot they have ever built.
The reality of humanoid robots
Robert Playter noted that humanoid robots have proliferated rapidly over the past three to five years, but success depends on commercial maturity. Boston Dynamics has a team for integration, service, and repairs. Most other companies only conduct demonstrations or sell robots as platforms while hoping someone else will develop the applications. The company draws on its experience deploying quadrupedal robots.
Playter criticized the strategy of some companies that want to put humanoids into homes first, given their high costs, limited capabilities, and lack of safety standards. Nevertheless, he sees a future for them in homes, just not yet. Alberto Rodriguez, director of robot behavior for Atlas at Boston Dynamics, said they are building the world's most capable humanoid and need a partner like DeepMind to provide reliable, scalable models for different tasks and industries.
Additional source: therobotreport.com
