# Google DeepMind releases Gemini Robotics 2 model family

> Google DeepMind launched Gemini Robotics 2, three robot AI models, on 30 July 2026.

*The three models give humanoid and two-armed robots whole-body control, multi-robot coordination and fast adaptation to new machines.*

By Behzad Hosseini · WireRead
Canonical: https://wireread.com/news/google-deepmind-releases-gemini-robotics-2-model-family

Google DeepMind announced Gemini Robotics 2 on 30 July 2026, a family of three AI models for controlling humanoid and two-armed robots. Google DeepMind said the release gives robots "whole body intelligence", covering motor control, high-level planning and on-device adaptation.

The family includes Gemini Robotics 2, a vision-language-action model that converts what a robot sees and hears into motor commands for full humanoid bodies and dexterous hands. A second model, Gemini Robotics ER 2, handles embodied reasoning: planning, following instructions, talking with people and coordinating several robots on multi-step tasks. A third, Gemini Robotics On-Device 2, runs locally on the robot and adapts to a new robot body in hours using fewer than 200 examples.

One checkpoint can control different robot bodies and plan over several minutes, according to [Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/). Supported platforms include Apptronik Apollo 2, Franka Duo, Dexmate, SO101 and Trossen.

Google DeepMind reported success rates of 45.7% to 76.3% on general whole-body manipulation tasks, 74.2% to 89.6% on gripper tasks, and 32% to 92% on multi-finger dexterity tasks. The company acknowledged that "the multi-finger dexterous manipulation remains challenging".

Gemini Robotics ER 2 is available now in Google AI Studio and through private preview. The vision-language-action and on-device models are limited to early-access partners.

Chinese manufacturers currently dominate global shipments of humanoid robots. Google is positioning its models as the underlying software that could run on robots built by multiple manufacturers.

Wider availability of the vision-language-action and on-device models beyond early-access partners has not been announced.

## Key takeaways

- Gemini Robotics 2 covers vision-language-action control, embodied reasoning and an on-device model
- The on-device model adapts to a new robot body in hours using fewer than 200 examples
- Multi-finger dexterity success rates ranged from 32% to 92% across tested tasks

## Sources

- [Gemini Robotics 2 brings whole body intelligence to robots](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/) — Google DeepMind, 2026-07-30