In robotics foundation models, transformers ingest large amounts of data, while diffusion refines outputs into smooth, precise, fine-grained motions.

“our consideration of the sitarata is generating image so that's why from first principle you know the whole field not just us have been using this kind of diffusion transformer architecture to train robotics foundation models you need the transformer aspect because transformer is the thing that can just ingest the most amount of data to learn then diffusion helps you to refine the outcome to get to the smooth precise fine-grained motion you need”