H3DP: Triply-Hierarchical Diffusion Policy for Visuomotor Learning

Authors: Y. Lu, Y. Tian, Z. Yuan*, X. Wang, P. Hua, Z. Xue, H. Xu
Status: Submitted to ICLR 2026.
Preprint: arXiv: 2505.07819

H3DP proposes a triply-hierarchical diffusion policy for visuomotor learning. The method decomposes control into coarse, mid, and fine granularities so that high-level intent, trajectory refinement, and low-level actuation can be optimized jointly. By coupling diffusion-based rollouts with hierarchical credit assignment, the approach targets better sample efficiency and stability on manipulation benchmarks.