AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Training-Free Guided Diffusion for Planning: A Unified Framework via Doob’s h-Transform with Safety Guarantees

Kenta Hoshino, Yashaswi Shashank Aluru, Xiyu Deng, Yorie Nakahira

ICML 2026 · Image generation › Fast diffusion sampling · approach: Guidance for conditional diffusion

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
increase PIC Monte Carlo rollout budget from L=16 to L=256 (ablation)PIC · 2D navigation Mediumtask success: 41/80 (51.2%) at L = 16 to 67/80 (83.8%) at L = 256Improved
increase PIC rollout budget from L=16 to L=256 (ablation)PIC · 2D navigation Mediummedian latency (s): 5.319 s to 10.128 sHurt
increase PIC rollout budget from L=128 to L=256 (ablation)PIC · 2D navigation Mediumobstacle-safe rate: 91.2% at L = 128 and 88.8% at L = 256No clear effect
add PIC rollout-based guidance over potential-gradient guidancediffusion planner · 2D navigation Hardtask success (%): 0.0 -> 67.5Improved
add PIC guidance over potential-gradient guidancediffusion planner · 2D navigation Simpletask success (%): 2.5 -> 98.8Improved
add reflected diffusion to gradient guidancePotential-Gradient Guidance · 2D navigation Mediumtask success (%): 3.8 -> 1.2No clear effect
use Full-Body PIC instead of direct-gradient guidanceFranka diffusion prior · Franka Hard-Bartask success (%): 0.0 -> 80.0Improved
use Full-Body PIC instead of diffusion inpainting with IKFranka diffusion prior · Franka Paneltask success (%): 10.0 -> 88.0Improved
use PIC instead of gradient guidance on unseen randomized layoutsdiffusion planner · Randomized layouts Overalltask success (%): 3.9 -> 92.8Improved
use PIC instead of gradient guidance on unseen randomized layoutsdiffusion planner · Randomized layouts Overallobstacle-safe rate (%): 58.3 -> 97.8Improved

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