Google DeepMind announces Gemini Robotics 2 for whole-body intelligence

Google DeepMind has introduced Gemini Robotics 2, a major advancement designed to give robots whole-body intelligence, advanced dexterity, and multi-robot collaboration capabilities. While previous systems largely focused on upper-body table-top manipulation, the new intelligence layer expands physical AI across entire humanoid and multi-arm robotic hardware platforms.

A Three-Tiered Architecture for Physical Intelligence

The Gemini Robotics 2 release is built around three specialized models designed to handle everything from low-level motor execution to high-level strategic planning and local device deployment:

• Gemini Robotics 2: An advanced vision-language-action model that translates visual inputs and natural language instructions into precise motor control, enabling full-body movements ranging from feet to fingertips.
• Gemini Robotics ER 2: A vision-language model serving as an embodied reasoning brain, enabling robots to process live video feeds, plan multi-step workflows lasting several minutes, track task progress, and coordinate multiple robots.
• Gemini Robotics On-Device 2: An efficient vision-language-action model optimized to run locally without internet connectivity, capable of adapting to entirely new robot embodiments within hours using fewer than 200 examples.