Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
Google says ER 2 can watch a robot’s live video feed, judge task progress, and coordinate handoffs between different machines.
The model is positioned as a high-level controller that plans steps, calls tools, and passes execution to lower-level robot systems. DeepMind says it improves over ER 1.6 in tool orchestration, progress classification, moment-finding, spatial reasoning, and safety benchmarks. The note cites 57.4% accuracy on progress classification and 91.3% accuracy on moment-finding, with sub-second latency for physical tasks. It is available through the Gemini API, Google AI Studio, and private preview on Gemini Enterprise Agent Platform. Google DeepMind's note
The model is positioned as a high-level controller that plans steps, calls tools, and passes execution to lower-level robot systems. DeepMind says it improves over ER 1.6 in tool orchestration, progress classification, moment-finding, spatial reasoning, and safety benchmarks. The note cites 57.4% accuracy on progress classification and 91.3% accuracy on moment-finding, with sub-second latency for physical tasks. It is available through the Gemini API, Google AI Studio, and private preview on Gemini Enterprise Agent Platform. Google DeepMind's note
score 8