Gemini Robotics is Google DeepMind's family of AI models, built on Gemini, that let robots understand what they see and control their movements.
In plain words
Chatbots work with text and images, but robots must also move real objects. Gemini Robotics adds that step. One model is a vision-language-action model that turns camera views and instructions directly into robot movements. A second, the embodied reasoning model, plans tasks and reasons about space, such as where to point or grasp. The first version was presented in March 2025; the latest are Gemini Robotics 2.5 and ER 2.
A quick example
Imagine telling a two-armed robot to put a banana in a bowl. The model looks at the scene, works out where each object is, and moves the arms to do it, even with objects it has not seen before.
Where you'll meet it
- Robot arms and humanoid robots.
- A preview API for robot reasoning in the Gemini API.
Related
Sources
- PaperGemini Robotics: Bringing AI into the Physical World, Gemini Robotics Team (arXiv) · read 6 Oct 2026
- DocsGemini Robotics, Google DeepMind · read 10 Nov 2026
- DocsGemini Robotics ER overview, Gemini API docs · read 2 Nov 2026
Checked by WhatIsAI against the sources above · 10 Nov 2026. Spotted an error? Report it.