Robot Harbour / Physical AI platform
Gemini Robotics 2
Source-backed intelligence on the models, software and compute systems giving robots the ability to perceive, reason and act.
- Developer
- Google DeepMind
- Status
- Limited access
- Access
- Selected partners
- First released
- Not verified
- Systems
- Tabletop robots, arms, bi-arm systems and humanoids
- Last verified
- 2026-08-06
Gemini Robotics 2 is Google DeepMind’s current vision-language-action model for converting visual input and natural-language instructions into robot actions.
What it does
The model is designed to control robots of different forms, from tabletop systems and arms to full humanoids. Google DeepMind highlights whole-body control, dexterous manipulation and behaviour that adapts when tasks or surroundings change.
Current access
The official model page labels Gemini Robotics 2 as a private preview and offers a waitlist for early access. It is therefore not a generally available robot software product, and Robot Harbour does not treat demonstrations or partner access as commercial availability.
What remains unknown
The live product page does not state a release date, pricing or a public deployment schedule. Those fields remain unverified rather than inferred. Performance figures shown by Google DeepMind are company evaluations and should be read alongside future independent testing.