Google DeepMind

Google's AI research division, source of the RT series and Gemini Robotics.

DeepMind was founded in London in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman, acquired by Google in January 2014, and merged with Google Brain in April 2023 to form Google DeepMind, still headquartered in London. Around the time of the merger (with Google Brain's own robotics team, Robotics at Google, folded in), its landmark robotics work included RT-1, RT-2 (one of the first VLAs), the cross-embodiment dataset Open X-Embodiment, RoboCat, and ALOHA Unleashed. Starting in 2025, the focus shifted to the Gemini Robotics series: the first version in March, 1.5 in September, and Gemini Robotics 2, which supports whole-body humanoid control, on July 30, 2026. On the hardware side, it partners with Apptronik and Boston Dynamics.

ExampleRT-2 directly fine-tuned a vision-language model trained on web image-text data to output robot action tokens, helping popularize the term 'VLA.'

Also called

DeepMind, Robotics at Google, Google Brain robotics team

Related

RT-2 · Gemini Robotics · Gemini Robotics 2 · Open X-Embodiment · Genie 3 · Vision-Language-Action Model

Sources

Google DeepMind - Wikipedia
Gemini Robotics 2 brings whole-body intelligence to robots

As of

2026-07