Robot Harbour / Physical AI platform
Gemini Robotics ER 2
Source-backed intelligence on the models, software and compute systems giving robots the ability to perceive, reason and act.
- Developer
- Google DeepMind
- Status
- Preview
- Access
- API access
- First released
- Not verified
- Systems
- Robot systems using VLA or lower-level control
- Last verified
- 2026-08-06
Gemini Robotics ER 2 is Google DeepMind’s embodied-reasoning model for understanding physical spaces, planning multi-step robot tasks and monitoring progress.
What it does
ER 2 acts as a higher-level robot brain. It can interpret multimodal inputs, reason about objects and spaces, plan a sequence of actions and hand motor execution to a lower-level vision-language-action model. Google DeepMind also presents multi-robot coordination, tool use and success tracking as core capabilities.
Current access
The official model page lists ER 2 as a public preview in Google AI Studio and the Gemini API, with a separate private preview in the Gemini Enterprise Agent Platform. API access to reasoning is distinct from access to Gemini Robotics 2’s motor-control model.
Evidence and limits
Google DeepMind publishes evaluation results for spatial reasoning, success detection and safety instruction following. These are first-party benchmarks. Real-world reliability still depends on the connected VLA, robot hardware, sensors, safety controls and deployment environment.