Dyna Robotics has unveiled a new robot foundation model trained on more than 1 million hours of human video. The company said its DYNA-2 World-Action Model was trained entirely on human egocentric video rather than robot action data. The dataset represents 170 years of continuous waking experience.

In tests, DYNA-2 raised task success rates in high-precision manufacturing from 20 per cent to 90 per cent through pre-training scale alone. It also demonstrated transfer of across stationary robot arms, humanoids, and dexterous robotic hands, so knowledge of human behavior can be transferred across platforms.