Anton Arapin

Extracting the local shape of the action-conditioned data manifold once lets any number of candidate actions be scored by matrix multiplication alone, making sixteen candidates over an order of magnitude faster than the source generator and more accurate at short horizons.

Extracting the local shape of the action-conditioned data manifold once lets any number of candidate actions be scored by matrix multiplication alone, making sixteen candidates over an order of magnitude faster than the source generator and more accurate at short horizons.

D3WM splits next-frame prediction into a deterministic part and an uncertain part, using a learned uncertainty map to steer a diffusion model toward occlusions and dynamic textures, and it beats comparison world models on both robotic manipulation and camera motion.

D3WM splits next-frame prediction into a deterministic part and an uncertain part, using a learned uncertainty map to steer a diffusion model toward occlusions and dynamic textures, and it beats comparison world models on both robotic manipulation and camera motion.