Agents / Skills
Dyna 2
A world-action model scaling robot performance through 1 million hours of human video.
Playable
Dyna 2 media is blocked
Allow external media to connect to the provider and play this content.
Description
Dyna-2 is a generative world-action model (WAM) built on a DiT-style video diffusion backbone that jointly predicts future video and robot actions. It establishes a first-of-its-kind human-to-robot transfer scaling law, demonstrating that pre-training on 1 million hours of egocentric human manipulation video predictably improves performance across diverse robot platforms, including bimanual arms and five-fingered dexterous hands.
Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.