Details
- Google AI announces Gemini Robotics 2 as the new intelligence layer for the next generation of adaptable robots.
- The system extends Gemini’s multimodal capabilities into whole-body robot control, enabling walking, crouching, stretching, and complex object manipulation.
- Gemini Robotics 2 supports multi-robot collaboration, allowing different robots to coordinate in shared spaces and complete workflows no single robot could handle alone.
- The release includes three models: Gemini Robotics 2 (vision-language-action), Gemini Robotics ER 2 (embodied reasoning), and Gemini Robotics On-Device 2 (optimized for local, on-robot execution).
- Gemini Robotics 2 can adapt to entirely new robotic bodies in just a few hours of data, widening applicability across humanoids and bi-arm platforms.
- ER 2 acts as a high-level reasoning engine, orchestrating multi-step tasks, monitoring progress via continuous video, and calling tools such as Google Search.
- On-Device 2 brings Gemini’s capabilities to robots with limited compute, enabling general-purpose dexterity and task generalization directly on hardware.
- ER 2 is available via the Gemini API, Google AI Studio, and in private preview on Gemini Enterprise Agent Platform, with VLA and On-Device models offered to early-access partners.
- The announcement builds on earlier Gemini Robotics and On-Device releases, positioning Robotics 2 as a major upgrade in physical AI and real-world autonomy.
Impact
Gemini Robotics 2 strengthens Google DeepMind’s push to make general-purpose robots commercially viable by combining high-level reasoning with whole-body motor control and on-device deployment. This accelerates the shift from lab demos to real-world workflows and pressures rivals in humanoid robotics and foundation model robotics to match multi-robot collaboration and fast embodiment adaptation capabilities.
