Dyna-2 Pre-trained on 1M Hours of Egocentric Video, Boosting Robot Success to 53%

JasonMa2020 · x · 2026-08-11

Dyna Robotics introduced Dyna-2, a new world-action model (WAM) pre-trained on over 1 million hours of real egocentric human video.

Mechanism: It learns by jointly predicting the next video frames and actions. Video prediction (world modeling) is the key to transferring human knowledge to robots.
Scaling Effects: More human video predictably improves action prediction. Task accuracy improves across embodiments (bimanual arms and a humanoid) with zero robot data in pre-training.
Performance: Post-trained on just a few hours of robot data, mean task performance jumped from 20% to 53% as human pre-training scaled from 1k to 1M hours.
WAM vs VLA: Matched apple-to-apple, the WAM hit 1.55x the VLA's success rate and won 65% of head-to-heads. Deployed zero-shot at unseen customer sites, it passed 87% of tasks vs the VLA's 46%.

More from Embodied