Google DeepMind unveils Gemini Robotics 2 for robots from tabletop arms to humanoids

Google DeepMind introduced Gemini Robotics 2, which it describes as its most advanced vision-language-action model yet. The model is built to help robots interpret images, process language, and control actions in physical environments.

The key takeaway: Google DeepMind is positioning Gemini Robotics 2 as an intelligence layer for adaptive robots, not just a model for one specific machine type.

According to DeepMind, Gemini Robotics 2 can control systems ranging from tabletop arms to full-body humanoid robots. That breadth matters because robotics software often needs to adapt to different bodies, tasks, and physical constraints.

The article says the model can manage full-body movement, perform fine motor tasks, and coordinate multiple robots. For readers tracking AI in practice, the focus is on whether general-purpose robotics models can move beyond demos into more flexible real-world use.

Google DeepMind also introduced Gemini Robotics ER 2, a model designed for “embodied reasoning.” In the article’s terms, embodied reasoning means understanding the physical world and deciding which actions to take based on that information.

ER 2 acts as a higher-level control system for robots and replaces Gemini Robotics ER 1.6, which was released in April. The new model is available in Google AI Studio, according to the article.

Developers can apply for early access to Gemini Robotics 2 through a waitlist mentioned in the article. The practical question is how well the model handles varied robot bodies, coordinated robot behavior, and tasks that require both fine motor control and physical-world reasoning.

For teams evaluating robotics AI, the immediate takeaway is to separate two layers: Gemini Robotics 2 for vision-language-action control, and Gemini Robotics ER 2 for higher-level reasoning about what a robot should do next.