Google Cloud opened Gemini Robotics Embodied Reasoning 2 in public preview, supporting robot task planning and multi-robot coordination.

Gemini Robotics Embodied Reasoning 2 (ER 2) is available in public preview on the Gemini platform. Google describes it as the high-level reasoning component of Gemini Robotics 2, designed for embodiments ranging from stationary robot arms to humanoids.

The model combines spatial and temporal video understanding with long-horizon planning and safety guardrails. Google says it can converse with people, coordinate low-level action models and hardware application programming interfaces (APIs), execute multi-minute tasks involving hundreds of decisions, and coordinate multiple robots.

Developers can expose low-level controls, including vision-language-action (VLA) models and navigation APIs, as tools while streaming video, audio, or text to the model. Its task-verification functions classify progress using binary outcomes, discrete categories, or continuous scores, supporting correction of failed steps without restarting an entire workflow.

Google reports 91.3% accuracy and a 0.96-second mean absolute frame distance for identifying critical events in video, alongside execution at four times the speed of “many larger model categories”; the excerpt does not name those comparison models. ER 2 can also interpret more than 10 instrument categories, including circular dials, sight glasses, digital displays, linear scales, rulers, and liquid thermometers, while the cloud offering adds enterprise security, multi-agent orchestration, and lifecycle management.