Nicolas's Ownblog
← All posts
Robotics

Google brings its robot reasoning model to Google Cloud

Two robots, one with two arms at a table and one humanoid, beside the words Gemini Robotics ER 2.

The robots Duo and Apollo in Google's announcement image.

Source: Google, press image, cropped

Google has made Gemini Robotics ER 2 available on Google Cloud in public preview. Google first introduced the model in July through the Gemini API and describes it as "a high-level brain for robots". It handles conversation, spatial understanding and multi-step planning, then hands motor control to lower-level models.

According to Google, the model follows task progress in continuous video, reads instruments such as dials, sight glasses and thermometers, and plans tasks lasting several minutes by calling lower-level controls as tools. It is meant to work across robot types, from stationary arms to humanoids.

The Cloud release connects it to Google's enterprise tools, with simulation through MuJoCo-Warp and NVIDIA Isaac Sim and edge deployment through Google Distributed Cloud.

The performance figures are Google's own: 91.3% accuracy at finding key moments in video, such as when to stop pouring, and 57.4% at classifying task progress.

Sources