Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
deepmind ·
Google DeepMind has launched Gemini Robotics ER 2, an advanced embodied reasoning model engineered to function as a high-level cognitive system for physical robots. The platform significantly improves real-time spatial analysis by enabling machines to continuously monitor video feeds, quantify task progression across five performance tiers, and pinpoint exact moments when specific actions conclude. Leveraging the Gemini Live API, the system delivers low-latency orchestration that removes traditional processing interruptions while dynamically adjusting to operational errors or environmental shifts. Through accessible developer interfaces, engineers can now automate complex multi-step sequences and coordinate cooperative workflows among heterogeneous robotic units operating in shared environments. The architecture also introduces upgraded spatial perception tools, including the ability to detect mid-task failures and accurately read both analog and digital measurement devices from live streams. Enhanced safety mechanisms allow the model to autonomously halt movements when personnel approach nearby zones and resume operations only once those areas are cleared. By merging rapid execution capabilities with reliable self-correction and cross-machine communication, this release positions the technology as a more viable solution for deploying intelligent agents in dynamic physical settings.
deepmind