Seonsoo Kim

Diffusion ReRoll: Revisable Denoising for Robotic Sequential Prediction

Seonsoo Kim, Seongil Hong, Jun-Gill Kang

We propose Diffusion ReRoll, a diffusion-based framework for robotic sequential prediction that enables revisable denoising over horizons. Existing diffusion-based sequence predictors typically perform a single monotonic denoising process. In contrast, Diffusion ReRoll selectivel…Read more