Boston Dynamics, a global leader in mobile robotics, and Google DeepMind, one of the world's leading artificial intelligence research centers, announced a new partnership at CES 2026 in Las Vegas focused on integrating advanced AI models into humanoid robots. The collaboration aims to combine the athletic intelligence of Boston Dynamics robots with the foundational artificial intelligence capabilities of Google DeepMind.
The partnership will focus on using Gemini Robotics foundation models together with a new fleet of Atlas robots. The joint research is expected to begin later this year and will take place at both companies' facilities.
Gemini Robotics: AI Models for the Physical World
Google DeepMind has developed specialized AI models called Gemini Robotics, which are built on the large multimodal Gemini model. These models enable robots of any shape and size to perceive their surroundings, reason, use tools, and interact with people.
The system uses a dual approach that combines two specialized models. The first is Gemini Robotics 1.5, a vision-language-action (VLA) model that can "see" (vision), "understand" (language), and "act" (action) in the physical world. It processes visual inputs and user commands, learns across different embodiments, and improves its problem-solving capabilities.
The second is Gemini Robotics-ER 1.5, an embodied reasoning model that specializes in understanding physical spaces, planning, and making logical decisions about its surroundings. This model does not directly control robotic limbs, but provides high-level insights that help the VLA model decide what to do next.
Atlas: A Humanoid Robot
The Atlas robot represents the latest generation of Boston Dynamics humanoid robots. It has 56 degrees of freedom/mobility with rotational joints and human-scale hands with tactile sensing. The robot can lift up to 50 kg and is designed to perform repetitive movements.
For safe interaction with people, Atlas is equipped with 360-degree cameras that allow it to see when people approach. According to Albert Rodriguez, director of behavior for the Atlas robot at Boston Dynamics, the ability to interact naturally with people is crucial to the success of humanoid robots.
"We are building the world's most capable humanoid and knew we needed a partner to help us create new types of vision-language-action models for these complex robots. No one in the world is better equipped than DeepMind to create reliable, scalable models that can be safely and effectively deployed across a wide range of tasks and industries," Rodriguez said.
Focus on the Automotive Industry
The strategic partnership focuses on enabling humanoid robots to complete a wide range of industrial tasks and is expected to become a driving force behind the transformation of manufacturing, starting with the automotive industry.
Carolina Parada, senior director of robotics at Google DeepMind, explained the project's vision: "We developed our Gemini Robotics models to bring AI into the physical world. We are excited to begin working with the Boston Dynamics team to explore what is possible with their new Atlas robot as we develop new models to expand the impact of robotics and scale robots safely and efficiently."
Hyundai Motor Group, the majority shareholder of Boston Dynamics, plans to bring Atlas to its factory later this year and gradually deploy the robots for tasks such as parts sequencing by 2028. This year, the company is opening a facility in the United States called the Robot Metaplant Application Center (RMAC), which will teach robots how to map movements such as lifting and turning.
Training data from RMAC will be combined with real-world data collected through the software platform used at the Georgia factory so that the robots can continuously improve.
Boston Dynamics already has products such as the four-legged Spot robot, which is in the hands of customers in more than 40 countries. Since its launch in 2023, the Stretch warehouse robot has unloaded more than 20 million boxes worldwide.
Collaboration with Other Robot Manufacturers
Google DeepMind is also working with more than 60 trusted testers, including companies such as Apptronik, Agility Robotics, Agile Robots, Universal Robots, and others. These partnerships are intended to help shape the future of Gemini Robotics-ER models.
Apptronik, which is developing the humanoid Apollo robot, is one of Google DeepMind's key partners. This University of Texas startup recently expanded its Series A funding to a total of $935 million (approximately CZK 19.2 billion), with a valuation exceeding $5 billion (about CZK 103 billion).
Gemini Robotics Capabilities
Gemini Robotics models provide robots with several key capabilities. They can understand the physical world and adapt their behavior to new situations. They break goals down into manageable steps so they can create longer-term plans and overcome unexpected problems.
Robots using these models can autonomously use digital tools to solve complex tasks, including native tool calling such as Google Search to find information. The system allows robots to "think before acting," which improves the quality of their actions and makes their decision-making more transparent in natural language.
Another advantage is the ability to learn across different robot bodies. A single model can be used across all these robots, which in turn accelerates its learning across multiple embodiments.
Boston Dynamics did not announce its intention to build a commercial humanoid until 2024, after it became clear that recent advances in AI had accelerated the pace at which robots could be trained and deployed in real-world applications.
Sources: bostondynamics.com and deepmind.google
