Google released Gemini Robotics ER 2, a model that plans and supervises robot tasks while leaving the actual movement to separate control software. It can follow continuous video to judge whether a step is finished, correct itself part-way through, and coordinate two different robots on a shared job. It is available to developers through the Gemini API and AI Studio, and only in private preview on Google's enterprise agent platform.
What changed
The previous model, Gemini Robotics ER 1.6, worked from still images rather than continuous video and did not support robots working together.
What it unlocks
Building a robot controller that watches a live video feed, judges how far along a task is, corrects itself mid-task, and hands work between two different robots.
- 57.4% accuracy on progress classification
- 91.3% accuracy on moment-finding
- 0.96s mean absolute distance
- 4x the execution speed
What you need to act on it
- access to the Gemini API or Google AI Studio
- private preview access for the Gemini Enterprise Agent Platform
- a robot with a control interface the model can call as a tool
- blog.google2026-07-30