Google Deepmind's Gemini Robotics models will power Boston Dynamics' Atlas for industrial tasks

Robotics pioneer Boston Dynamics and AI lab Google Deepmind have announced a strategic partnership to integrate Deepmind's Gemini Robotics models into Boston Dynamics' Atlas humanoid robots.

The two companies will also conduct joint research at both company sites over the coming months, using a new fleet of Atlas robots. Google Deepmind hired former Boston Dynamics CTO Aaron Saunders back in November.

The collaboration aims to prepare humanoid robots for a wide range of industrial tasks. According to the announcement, the partnership will help drive manufacturing transformation, starting with the automotive industry. Hyundai Motor Group is Boston Dynamics' majority shareholder.

"We are building the world's most capable humanoid, and we knew we needed a partner that could help us establish new kinds of visual-language-action models for these complex robots," said Alberto Rodriguez, Director of Robot Behavior for Atlas at Boston Dynamics. The company said recent AI advances have sped up how quickly robots can be trained and deployed in real-world settings.

Deepmind's foundational robotics models power the system

Google Deepmind has been refining its foundational robotics models, including Gemini Robotics, over the past few years. These models build on the multimodal Gemini architecture and help robots of different shapes and sizes perceive their environment, reason through problems, use tools, and interact with humans.

The company shipped several AI models for robotics in 2025, most recently Gemini Robotics 1.5 and Gemini Robotics-ER 1.5. Robotics 1.5 is a vision-language-action model for direct robot control, while Robotics-ER 1.5 handles higher-level planning: it can tap into digital tools like Google Search, communicate in natural language, and evaluate how actions are progressing.

Google also introduced Deepmind Gemini Robotics On-Device, a model that runs directly on robot hardware without a cloud connection. All of these models are generalists, meaning they work across different robot platforms, including Boston Dynamics' systems.