Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
DeepMind Blog · 2026-07-30
Google DeepMind launches Gemini Robotics ER 2, an embodied reasoning model achieving 91.3% accuracy on moment-finding tasks at 4x the speed of larger competing models, enabling real-time video-based task progress tracking and multi-robot collaboration for complex multi-step physical workflows.
Appears in
Extraction
Topics: embodied-reasoningroboticsmulti-robot-collaborationphysical-airobot-safety
Claims
- Gemini Robotics ER 2 achieves 91.3% accuracy on moment-finding tasks with a 0.96-second mean absolute distance, at 4x the execution speed of competing larger models and at a fraction of their compute cost.
- The model achieves 57.4% accuracy on continuous progress classification across five progress levels, outperforming previous-generation and competing frontier models.
- Multi-robot collaboration enables heterogeneous robots to communicate via shared semantic understanding to hand off and complete tasks no single robot could accomplish alone.
- The model integrates with Gemini Live API for sub-second latency execution, enabling fluid real-time orchestration of physical robotics without stop-and-think pauses.
- The new ASIMOV-Agentic safety benchmark evaluates whether foundation models can enforce safety constraints, assess physical feasibility, and seek human clarification when uncertain.
Key quotes
Gemini Robotics ER 2 achieves 91.3% accuracy and a 0.96s mean absolute distance [on moment-finding]. It competes closely with much larger model categories, but delivers this precision at a fraction of the compute cost and 4x the execution speed.
Gemini Robotics ER 2 successfully halts a humanoid robot when a person is nearby and autonomously resumes work only once the area is clear.
Think of Gemini Robotics ER 2 as a high-level brain for robots. It allows robots to chat with humans, understand the physical world, and plan multi-step tasks.