FrontierJuly 30, 2026via MarkTechPost
Google DeepMind Ships Three Physical AI Models For Whole Body Control, Dexterity And Multi Robot Collaboration
Why it matters
Major capability release in physical AI: multi-model system for robot control and reasoning represents a significant step in embodied AI deployment. The on-device adaptation angle (hours to new bodies) is particularly actionable for practitioners building robot systems.
Key signals
- Three models released: vision-language-action (whole body control), Gemini Robotics ER 2 (embodied reasoning + task orchestration), on-device VLA (fast adaptation)
- Tested on Apptronik Apollo 2 and Franka Duo robot platforms
- On-device model adapts to new robot bodies in hours (significant speed improvement claim)
- Only ER 2 is publicly available; other two are integrated/proprietary
- Released July 30, 2026
- Represents multi-robot collaboration capability in single model system
The hook
Google DeepMind just shipped three robotics models in one drop—whole-body humanoid control, embodied reasoning, and on-device adaptation. Only one is public.
Google DeepMind has released Gemini Robotics 2, the intelligence layer for its next generation of robots. The release ships three models: a vision-language-action model for whole body humanoid control, Gemini Robotics ER 2 for embodied reasoning and task orchestration, and an on-device VLA that adap…