AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Model-Based Diffusion Sampling for Predictive Control in Offline Decision Making

Haldun Balim, Na Li, Yilun Du

ICML 2026 · Image generation › Fast diffusion sampling · approach: Guidance for conditional diffusion · 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
alternate dynamics-model and planner denoising steps instead of planner alone (ablation)MPDiffuser · D4RL medium-replay (3 tasks)average normalized return: 69.3 -> 74.4Improved
give planner twice the denoising steps (200 vs 100) (ablation)Planner · D4RL medium-replay (3 tasks)average normalized return: 69.3 -> 68.8No clear effect
alternate planner and dynamics updates instead of combining their scores (ablation)MPDiffuser · D4RL Hopperaverage normalized return: 98.2 -> 102.4Improved
add a ranker selecting among 64 samples (ablation)MPDiffuser · D4RL Locomotion averagenormalized return: 84.9 -> 85.1Improved
add ranker (ablation)MPDiffuser · D4RL Kitchen partialnormalized return: 67.9 -> 73.8Improved
add cheap random trajectories for dynamics model training (ablation)MPDiffuser · FetchPickAndPlacesuccess rate: 0.75 -> 0.86Improved
add random trajectories for dynamics model training (ablation)D-MPC · FetchPickAndPlacesuccess rate: 0.60 -> 0.60No clear effect
fine-tune only the dynamics model on 100 play episodes under defectMPDiffuser · Walker2d-medium with ankle defectnormalized return: 58.6 -> 66.4Improved
add dynamics model to planner under defect (ablation)Walker2d-medium with ankle defect · Walker2d-medium defect, post fine-tuningnormalized return: 56.0 -> 66.4Improved
increase number of ranked samples to 64 without dynamics feasibility (ablation)Planner · Pendulum with velocity constraintsuccess rate: 89 -> 66 (4 to 64 samples)Hurt
condition dynamics model on task information (ablation)MPDiffuser · D4RL medium-replay (3 tasks)average normalized return: 71.1 -> 74.4Improved
use acausal instead of causal denoisers (ablation)MPDiffuser · D4RL medium-replay (3 tasks)average normalized return: 69.0 -> 74.4Improved

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