Gemini Robotics On-Device 2 is an efficient Vision-Language-Action model (VLA) developed by Google DeepMind that runs entirely locally on robotics hardware without requiring an internet connection. Based on Gemini Robotics 1.5 technology and On-Device Gemma models, it can be adapted to new robot embodiments with different form factors, sensors, and degrees of freedom in just a few hours using typically fewer than 200 examples, thanks to "Motion Transfer" technology. The model was unveiled on July 30, 2026, alongside Gemini Robotics 2 and Gemini Robotics ER 2, and is currently accessible only to select "Trusted Testers" and Early-Access partners such as Apptronik, Boston Dynamics, and Agile Robots.

Sources (3)