arXiv cs.AI· Kyowoon Lee, Yunhao Luo, Anh Tong, Jaesik Choi·· 3 天前AI 评分24
RCD：用免训练引导改进组合式扩散规划，缓解模式平均问题
Refining Compositional Diffusion for Reliable Long-Horizon Planning
AI 导读
针对组合式扩散规划在局部计划多模态时出现的模式平均问题，研究者提出免训练引导方法 RCD，用预训练扩散模型的自重建误差作为组合计划对数密度的代理，并结合重叠一致性项约束分段边界。在 OGBench 的移动、物体操作和像素观测等长时程任务上，RCD 持续优于现有方法，该工作已被 NeurIPS 2026 接收。
正文 · 原文
Abstract:Compositional diffusion planning generates long-horizon trajectories by stitching together overlapping short-horizon segments through score composition. However, when local plan distributions are multimodal, existing compositional methods suffer from mode-averaging, where averaging incompatible local modes leads to plans that are neither locally feasible nor globally coherent. We propose Refining Compositional Diffusion (RCD), a training-free guidance method that steers compositional sampling toward high-density, globally coherent plans. RCD leverages the self-reconstruction error of a pretrained diffusion model as a proxy for the log-density of composed plans, combined with an overlap consistency term that enforces consistency at segment boundaries. We show that the combined guidance concentrates sampling on high-density plans that mitigate mode-averaging. Experiments on challenging long-horizon tasks from OGBench, including locomotion, object manipulation, and pixel-based observations, demonstrate that RCD consistently outperforms existing methods.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.03075 [cs.RO] |
| (or arXiv:2605.03075v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2605.03075 arXiv-issued DOI via DataCite |
Submission history
From: Kyowoon Lee [view email]
[v1]
Mon, 4 May 2026 18:44:19 UTC (10,405 KB)
[v2]
Sat, 10 Oct 2026 12:47:25 UTC (10,409 KB)
来源：arXiv cs.AI · arxiv.org