Gemini Robotics ER 2 Enhances Embodied Reasoning for Robots

Gemini Robotics ER 2 is introduced as Google's most capable embodied reasoning model, designed to enable robots to think and act in real-time, bridging high-level reasoning with motor execution. The model allows robots to chat with humans, understand the physical world, and plan multi-step tasks by handing off execution to vision-language-action models and natively calling tools.

The new release introduces multi-robot collaboration, allowing diverse robot types to communicate and work together in shared spaces. This enables the completion of complex workflows that would be beyond the capabilities of a single robot.

Gemini Robotics ER 2 offers significant upgrades over its predecessor by enabling robots to track their progress via continuous video feeds, adapt to errors, and know when to proceed to the next task. This allows for more fluid orchestration without jarring pauses, even during complex, multi-step processes.

Gemini Robotics ER 2 significantly improves temporal understanding for robust task completion, featuring progress classification with 57.4% accuracy and moment-finding at 91.3% accuracy. This allows robots to precisely identify task completion points and manage sequential actions effectively.

The model advances spatial reasoning with improved success/failure detection on raw video, broader instrument reading capabilities, and enhanced visual question answering. It also demonstrates significant gains in safety, halting operations when humans are near and adhering to physical constraints during reasoning tasks.

Gemini Robotics ER 2 is publicly available to developers via the Gemini API and Google AI Studio, with a private preview on the Gemini Enterprise Agent Platform. Code examples and configuration guides are provided to facilitate adoption.