Google Cloud released Gemini Robotics Embodied Reasoning 2 in public preview. The model adds advanced spatial-temporal reasoning to enterprise robotics. It enables robots to verify tasks, self-correct and generalize to new situations. It works across robot types and supports multi-robot collaboration. Developers can orchestrate low-level control interfaces as tools. It integrates with Google Cloud for simulation, training and deployment. The model achieves 91.3% accuracy in detecting critical events. It runs at four times the speed of many larger models. It is now available via Model Garden in the Gemini platform.

• 91.3% accuracy on critical event detection
• 0.96-second mean absolute frame distance
• 4x execution speed