CHD: Coupled Hierarchical Diffusion for Long-Horizon Tasks

Ce Hao, Anxing Xiao, Zhiwei Xue, Harold Soh
Proceedings of The 9th Conference on Robot Learning, PMLR 305:1420-1451, 2025.

Abstract

Diffusion-based planners have shown strong performance in short-horizon tasks but often fail in complex, long-horizon settings. We trace the failure to loose coupling between high-level (HL) sub-goal selection and low-level (LL) trajectory generation, which leads to incoherent plans and degraded performance. We propose Coupled Hierarchical Diffusion (CHD), a framework that models HL sub-goals and LL trajectories jointly within a unified diffusion process. A shared classifier passes LL feedback upstream so that sub-goals self-correct while sampling proceeds. This tight HL–LL coupling improves trajectory coherence and enables scalable long-horizon diffusion planning. Experiments across maze navigation, tabletop manipulation, and household environments show that CHD consistently outperforms both flat and hierarchical diffusion baselines.

Cite this Paper

BibTeX
@InProceedings{pmlr-v305-hao25a, title = {CHD: Coupled Hierarchical Diffusion for Long-Horizon Tasks}, author = {Hao, Ce and Xiao, Anxing and Xue, Zhiwei and Soh, Harold}, booktitle = {Proceedings of The 9th Conference on Robot Learning}, pages = {1420--1451}, year = {2025}, editor = {Lim, Joseph and Song, Shuran and Park, Hae-Won}, volume = {305}, series = {Proceedings of Machine Learning Research}, month = {27--30 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v305/main/assets/hao25a/hao25a.pdf}, url = {https://proceedings.mlr.press/v305/hao25a.html}, abstract = {Diffusion-based planners have shown strong performance in short-horizon tasks but often fail in complex, long-horizon settings. We trace the failure to loose coupling between high-level (HL) sub-goal selection and low-level (LL) trajectory generation, which leads to incoherent plans and degraded performance. We propose Coupled Hierarchical Diffusion (CHD), a framework that models HL sub-goals and LL trajectories jointly within a unified diffusion process. A shared classifier passes LL feedback upstream so that sub-goals self-correct while sampling proceeds. This tight HL–LL coupling improves trajectory coherence and enables scalable long-horizon diffusion planning. Experiments across maze navigation, tabletop manipulation, and household environments show that CHD consistently outperforms both flat and hierarchical diffusion baselines.} }
Endnote
%0 Conference Paper %T CHD: Coupled Hierarchical Diffusion for Long-Horizon Tasks %A Ce Hao %A Anxing Xiao %A Zhiwei Xue %A Harold Soh %B Proceedings of The 9th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2025 %E Joseph Lim %E Shuran Song %E Hae-Won Park %F pmlr-v305-hao25a %I PMLR %P 1420--1451 %U https://proceedings.mlr.press/v305/hao25a.html %V 305 %X Diffusion-based planners have shown strong performance in short-horizon tasks but often fail in complex, long-horizon settings. We trace the failure to loose coupling between high-level (HL) sub-goal selection and low-level (LL) trajectory generation, which leads to incoherent plans and degraded performance. We propose Coupled Hierarchical Diffusion (CHD), a framework that models HL sub-goals and LL trajectories jointly within a unified diffusion process. A shared classifier passes LL feedback upstream so that sub-goals self-correct while sampling proceeds. This tight HL–LL coupling improves trajectory coherence and enables scalable long-horizon diffusion planning. Experiments across maze navigation, tabletop manipulation, and household environments show that CHD consistently outperforms both flat and hierarchical diffusion baselines.
APA
Hao, C., Xiao, A., Xue, Z. & Soh, H.. (2025). CHD: Coupled Hierarchical Diffusion for Long-Horizon Tasks. Proceedings of The 9th Conference on Robot Learning, in Proceedings of Machine Learning Research 305:1420-1451 Available from https://proceedings.mlr.press/v305/hao25a.html.

Related Material