Unified Video-Action Joint Denoising for Dexterous Action and Data Generation

Dingrui Wang, YuAn Wang, Jinkun Liu +4

Recent world action models leverage video foundation models by aligning broad visual-dynamics priors with executable robot actions. We revisit this alignment from a distributional…