Google DeepMind has launched Gemini Robotics ER 2, an advanced "embodied reasoning" model designed to significantly enhance robot capabilities in the physical world. This new model allows robots to understand video feeds, orchestrate complex multi-step tasks, and collaborate with other robots.

Gemini Robotics ER 2 acts as a high-level "brain" for robots, enabling them to communicate with humans, interpret their surroundings, and plan tasks. It can also integrate with external tools like Google Search. A key improvement over its predecessor, Gemini Robotics ER 1.6, is its ability to track progress through continuous video monitoring, allowing for adaptation and self-correction during tasks. This temporal intelligence is crucial for tasks like tightening a light bulb or tying a trash bag, with Gemini Robotics ER 2 achieving 57.4% accuracy on progress classification and 91.3% accuracy on precision moment-finding.

The model also introduces multi-robot collaboration, enabling diverse robots to work together. Available via the Gemini API and Google AI Studio, Gemini Robotics ER 2 aims to accelerate the development of helpful physical AI. Demonstrations include a Boston Dynamics Spot robot fetching a snack on command, showcasing fluid task orchestration and sub-second latency.