Google DeepMind just dropped Gemini Robotics 2, and it’s the kind of announcement that makes you rethink what robots will actually look like in five years. The new platform gives AI-powered robots the ability to handle complex physical tasks, continuously read their surroundings, and coordinate with other machines in real time. One of the sub-models, Gemini Robotics ER 2, is now publicly accessible to developers through Google AI Studio.
DeepMind scientists have a term for this ambition. They call it “physical AGI.”
What’s actually new under the hood
Gemini Robotics 2 is built on top of the broader Gemini 2.0 foundation and introduces what the company calls vision-language-action capabilities for whole-body intelligence. In English: the robot can see its environment, understand natural language instructions, and translate both into coordinated physical movements across its entire body.
The upgrade path here has been relatively swift. Google first launched Gemini Robotics on March 12, 2025. An on-device variant followed in June 2025. Now the 2.0 release adds long-horizon planning, meaning robots can chain together sequences of actions over extended periods without needing constant human supervision.
