Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model

Dynin-Robotics has implemented an omnimodal unified diffusion vision-language-action model that brings visual goal and dynamics prediction into action generation and selection through a shared trajectory model, according to the company. The approach is intended to give language-conditioned robot policies both a target outcome and a representation of action-dependent scene changes.

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 for