NVIDIA unveils GR00T 2, a robot model that learns skills from human video

Presented November 9th at CoRL, the world-action model reports 40% zero-shot success, rising to 95% after five demonstrations and a 30-minute fine-tune.

By · Published

AI-generated illustration of a person watching video of hands stacking blocks, beside a whiteboard linking human video, learning and robot skills.

Presented November 9th at CoRL, the world-action model reports 40% zero-shot success, rising to 95% after five demonstrations and a 30-minute fine-tune.