DYNA-2

DYNA-2 is Dyna Robotics' flagship World-Action Model (WAM), pre-trained on over 1 million hours of human egocentric video — roughly 170 years of continuous waking experience. Unlike VLA (Vision-Language-Action) models, DYNA-2 is built on a video-diffusion backbone that jointly denoises future video and future actions, enabling physical intuition and spatial reasoning to transfer directly from human video to robot hardware without any robot data in pre-training. Announced August 10, 2026, it demonstrates the first human-to-robot scaling law: robot performance improves smoothly and predictably with every added hour of human data, without hitting plateaus. DYNA-2 enables zero-shot performance at production-level speeds across new deployment sites, and can be adapted to new robot platforms with just a few hours of local fine-tuning. In one benchmark, 13 minutes of data was sufficient for DYNA-2 to command five-fingered robot hands to twist open a bottle cap.

1 robot model on the DEPLOY registry runs on DYNA-2: Sophy Swiftfold.

DYNA-2 is credited to Dyna Robotics on the DEPLOY registry. Each developer attribution is verified via primary sources.

No. DYNA-2 is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.

DYNA-2 is a world model on the DEPLOY registry.