Nvidia’s GR00T 2 foundation model trains robotic arms by transferring human movements from video rather than hours of remote control. As a world-action model it predicts the next movement and a short future video; training used about 400,000 hours of head-mounted camera footage on a cluster of more