Reproduce “CHD: Coupled Hierarchical Diffusion for Long-Horizon Tasks” 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
Ce Hao, Anxing Xiao, Zhiwei Xue, Harold Soh
Coupled Hierarchical Diffusion (CHD) from the National University of Singapore introduces a framework that jointly models high-level subgoals and low-level trajectories within a unified diffusion process, enabling bidirectional feedback and asynchronous parallel generation for long-horizon robot tasks. The approach improves performance in maze navigation by 10-15% and robot task planning, while reducing inference time by 30-40% compared to prior methods.
Key insights
- Bidirectional interaction between high-level subgoal generation and low-level trajectory execution is essential for robust and coherent long-horizon planning.
- A unified diffusion process, guided by a coupled hierarchical classifier, can effectively model the joint generation and refinement of both abstract subgoals and concrete actions.
- Asynchronous parallel generation significantly boosts the inference speed of hierarchical diffusion planners, making them more practical for real-time robotic applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.