Google released Gemini Robotics ER 2 to help robots complete complex tasks

Google launched Gemini Robotics ER 2 to serve as a high-level brain for robots. The system enables machines to perform multi-step planning and real-time spatial reasoning. It processes video feeds to verify task completion and can adjust actions when errors occur. Developers can now access the model through the Gemini API and Google AI Studio.

The update introduces the ability for different robots to collaborate in shared spaces. By using a shared semantic understanding, a fleet of diverse machines can divide labor for tasks that are too difficult for a single unit. The system integrates with the Gemini Live API to reduce latency during communication and robot movement.

Safety features include improved human proximity detection. The model can monitor its surroundings and pause operations if a person enters the workspace, resuming only when the area is clear. These benchmarks were tested on Boston Dynamics Spot robots and other industrial platforms.

While the model handles high-level decision-making, it relies on lower-level vision-language-action models to execute physical motor commands. The performance capabilities such as progress tracking and moment-finding are limited to the accuracy levels documented in the technical benchmarks, and the system is still being evaluated for more complex real-world workflows.