The announcement
On July 30, 2026, Google DeepMind introduced Gemini Robotics 2, which it calls the "intelligence layer" for the next generation of robots. The previous version showed Gemini could turn multimodal understanding into physical action. The new one adds three advances: whole-body control, greater dexterity and multi-robot collaboration.
What changes in practice
- A humanoid can walk, crouch, stretch and manipulate objects to tidy a cluttered room, and other robots can join in to finish faster.
- The model can run locally on the robot and, per DeepMind, adapt to a new robot body in just a few hours.
- Gemini Robotics ER 2 is the embodied reasoning model, a vision-language model that talks with people, understands its surroundings and plans multi-step tasks lasting minutes.
Availability
- ER 2: available in Google AI Studio and in private preview on Gemini Enterprise Agent Platform.
- VLA model and on-device version: early-access partners only.
Context
In its July recap, Google also listed Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber for production agents. Gemini Robotics 2 is the most visible part of that wave because it takes the model off the screen.
Why it matters
Robots that adapt to different bodies in hours cut one of robotics' biggest costs: training each machine from scratch. These are still company demos, so independent real-world results will show how well it scales.