Neat idea: jointly diffuse pixels and DINO features with separate noise levels. Then optimise the trajectory through 2D noise level space. Could do this with DINO + traditional VAE latents as well to get a souped-up version of ReDi (representationdiffusion.github.io Thodoris Kouzelis et al.)!

03.10.26

To bring generalist intelligent robots to the real world, we have to overcome the data scarcity problem. At Rhoda, we are solving it by reformulating robot policies as video generation. Today, we introduce the Direct Video-Action Model (DVA)

I love this result It shows that a robot foundation model ought to be a model trained to understand the world that happens to be able to perform robot tasks. Not just action behavior cloning from teleop/umi data to fit a demo. For this, the DVA approach is the clear direction.

Two months ago, I vaguely posted a number: 0.9 FID, one-step, pixel space. Now it is 0.75, and can be even lower. Many wonder how. I thought it might end as a small FID prank: simple and deliberate. It started with one question: can FID be optimized directly, and what does it