Gemini Robotics 2 gives humanoids whole-body control down to the fingertips

9:00 AM PT · July 30, 2026

Google DeepMind announced Gemini Robotics 2 on Thursday, a family of three models meant to move robots beyond single arm pick and place demos toward coordinated, whole body behavior. The centerpiece vision language action model controls a humanoid from feet to fingertips, handling locomotion and fine manipulation together rather than as separate systems, while a companion reasoning model, Gemini Robotics ER 2, acts as the robot’s planning brain, breaking a task into hundreds of decisions across several minutes and coordinating multiple robots on the same job. A third, on device model adapts to a new robot body in a matter of hours using fewer than 200 training examples, DeepMind said, a sharp drop from the data typically needed to port a model between hardware platforms. In testing, whole body manipulation tasks succeeded between 45.7 percent and 89.6 percent of the time depending on difficulty, with precise gripper insertions reaching nearly 90 percent. The release also introduces ASIMOV Agentic, a safety benchmark scoring how reliably a robot refuses unsafe tool calls and detects humans working nearby. Gemini Robotics ER 2 is available now on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform, while the vision language action and on device models are open to early access partners including Apptronik and Franka.