Ying-Shui Tan

StrucPhysVideo: Learning Physical Dynamics from Structured Captions and Robot Actions

StrucPhysVideo, a family of video world models that bridges physics-focused data curation with language- and action-conditioned prediction of scene evolution, is presented, and a sparse Mixture-of-Experts text-image-to-video model trained with a curriculum that progressively emphasizes physical dynamics while retaining...