Reproduce “Adaptive Online Replanning with Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siyuan Zhou, Yilun Du, Shun Zhang, Mengdi Xu, Yikang Shen, Wei Xiao, Dit-Yan Yeung, Chuang Gan
RDM proposes a method for adaptive online replanning in diffusion models, employing a likelihood-based criterion to determine when to replan and choosing between full regeneration or efficient local refinement strategies. The approach achieves higher success rates and computational efficiency across long-horizon planning, stochastic environments, and complex robotic control tasks.
Key insights
- The internal likelihood estimates of a diffusion model can serve as a principled and adaptive indicator for plan executability and deviations, enabling intelligent replanning triggers.
- Efficient local plan refinement ('Replan with Future Context') can achieve performance comparable to full regeneration while requiring significantly fewer computational steps.
- Adaptive replanning strategies, tailored to the severity of plan degradation, are more effective than fixed-interval replanning, which can sometimes hinder consistent progress.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.