Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Product launches· 日本語で読む

Dyna Robotics unveils robot foundation model trained on 1 million hours of human video

Dyna Robotics unveiled DYNA-2, a robot foundation model trained on more than 1 million hours of human egocentric video.

What happened

According to an Aug. 10, 2026, report by Interesting Engineering, Dyna Robotics unveiled the DYNA-2 World-Action Model, a robot foundation model. The company said it was trained on more than 1 million hours of human egocentric video, equivalent to about 170 years of waking experience, and improved task success rates in high-precision manufacturing from 20% to 80%-90%.

Why it matters

For people responsible for teaching robots to perform tasks on manufacturing floors, the announcement points to the possibility of using a model pretrained solely on human egocentric video rather than robot operation records. Dyna Robotics said tests in high-precision manufacturing raised success rates from 20% to 80%-90% and enabled knowledge transfer among fixed robotic arms, humanoid prototypes and dexterous robotic hands. This offers a basis for assessing how the burden of collecting training data for each type of robot could be reduced.

Sources

Translated from the Japanese original: /news/dyna-robotics-100-man-jikan-no-ningen-eizou-de-gakushuu-shit-1d03b5. The Japanese version is authoritative if they differ.