Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model
Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a representation of action-dependent scene changes. We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.