Google DeepMind introduces Gemini Robotics ER 2

The Google DeepMind team introduces Gemini Robotics ER 2 — a robot brain that can interpret video in real time and command multiple machines simultaneously. The model achieves 57.4% accuracy in assessing task progress on a five-point scale, pinpointing the event of interest in a video with 91.3% accuracy. It can also halt a robot if it senses someone nearby and can read the scale of ten kinds of meters.

Such models help overcome the main barrier to industrial automation. They allow robots to break free of pre-defined scripts and begin autonomously assessing the quality of their work, adapting to change, and collaborating with other machines in multi-robot settings. They are a crucial step towards automating tasks that would have required a human before. Gemini Robotics ER 2 is accessible via the Gemini API and Google AI Studio.