AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

HDFlow: Hierarchical Diffusion-Flow Planning for Long-horizon Tasks

Gireesh Nandiraju, Yuanliang Ju, Chaoyi Xu, Weiheng Liu, Yuxuan Wan, He Wang

ICML 2026 · Image generation › Fast diffusion sampling · approach: Flow matching distillation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use diffusion at high level and rectified flow at low level instead of flat diffusion (ablation)HDFlow · FurnitureBench lamp Lowsuccess rate (%): 24 -> 68Improved
use rectified flow at low level instead of diffusion (hierarchical diffusion) (ablation)HDFlow · FurnitureBench lamp Lowinference time (ms/step): 142 -> 88Improved
use diffusion at high level instead of flow (hierarchical flow) (ablation)HDFlow · FurnitureBench lamp Lowsuccess rate (%): 24 -> 68Improved
add manifold projection to EBM guidance (ablation)HDFlow · one_leg Lowsuccess rate (%): 84 -> 92Improved
add manifold-aware EBM guidance (ablation)HDFlow · one_leg Lowsuccess rate (%): 61 -> 92Improved
add contrastive world model training (ablation)HDFlow · one_leg Lowsuccess rate (%): 58 -> 92Improved
add contrastive world model training (ablation)HDFlow · lamp Lowsuccess rate (%): 27 -> 68Improved
use K=20 subgoals instead of K=10 (ablation)HDFlow · one_leg Lowsuccess rate (%): 61 -> 92Improved
use K=20 subgoals instead of K=30 (ablation)HDFlow · one_leg Lowsuccess rate (%): 85 -> 92Improved
use HDFlow instead of SHDSHD · cabinet Lowsuccess rate (%): 21 -> 55Improved
use HDFlow instead of RVT-2RVT-2 · RLBench Close Jarsuccess rate (%): 100.0 -> 96.0Hurt
use HDFlow instead of 3D Diffuser Actor3D Actor · RLBench Insert Pegsuccess rate (%): 65.6 -> 93.3Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy