Dynin-Omni Omnimodal Masked-Diffusion Backbone for Robotics
September 10, 2026
Dynin-Omni uses an omnimodal masked-diffusion backbone to represent language, visual observations, goals, and actions as discrete tokens. This enables unified trajectory modeling for action prediction, goal-state prediction, and instruction reconstruction.
HOW THIS AFFECTS YOU
●
builderYou can leverage a single model for multiple robot policy tasks including action and goal prediction.
●
researcherThis provides a unified framework for vision-language-action modeling in robotics.