Google DeepMind announces the robot AI model "Gemini Robotics 2," capable of controlling the whole body from walking to intricate two-handed tasks

Google DeepMind has unveiled the Gemini Robotics 2 family of robot AI models.

The system uses vision and language instructions to control full-body motion and perform fine-grained object manipulation.

It also introduced Gemini Robotics ER 2, a higher-level model for managing long-horizon task plans and supervision.

A lightweight on-device version, Gemini Robotics On-Device 2, was announced as well.

By integrating walking, posture control, bimanual manipulation, and multi-robot coordination, it marks progress toward general-purpose physical AI.