Research map / Image generation / Fast diffusion sampling
Fisher-Preserving Guidance: Training-Free Manifold Constraints for Safe Diffusion Control
Hao Ren, Zetong Bi, Yiming Zeng, Le Zheng, Zhi Li, Zhaoliang Wan, Lu Qi, Hui Cheng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Fisher-preserving guidance to NoMaD | NoMaD · CARLA | success rate (%): 51.11 -> 60.00 | Improved |
| add uncertainty-guided action blending to NoMaD | NoMaD · CARLA | success rate (%): 51.11 -> 57.78 | Improved |
| add both FPG and action blending (ablation) | NoMaD · CARLA | success rate (%): 48.89 -> 75.55 | Improved |
| add FPG on top of action blending (ablation) | NoMaD with AB · CARLA | success rate (%): 57.78 -> 75.55 | Improved |
| add action blending on top of FPG (ablation) | NoMaD with FPG · CARLA | average collisions: 0.911 -> 0.445 | Improved |
| use uncertainty-guided blending instead of random blending (ablation) | FPG + AB · CARLA | success rate (%): 57.78 -> 75.55 | Improved |
| add FPG to the Maze2D sampler without task guidance | DDPM baseline · Maze2D | collisions: 0.243 -> 0.170 | Improved |
| add FPG on top of TSDF guidance | DDPM baseline + TG · Maze2D | collisions: 0.071 -> 0.016 | Improved |
| add FPG to official Diffusion Policy | Diffusion Policy · PushT | score: 0.91 -> 0.94 | Improved |
| add FPG on real robot | NoMaD · Real robot, 3 scenes | success rate (successes/30): 19/30 -> 28/30 | Improved |
| use OPS low-rank projection instead of VJP-Fisher (ablation) | Fisher guidance · RTX 3060, 10 steps | total latency (ms): 95.32 -> 61.93 | Improved |
| reduce denoising steps from 10 to 4 with OPS guidance (ablation) | FPG-OPS · RTX 3060 | total latency (ms): 61.93 -> 45.12 | Improved |
Closest papers by meaning
- Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models (CVPR 2025 · similarity 0.88)
- Versatile Navigation Under Partial Observability via Value-guided Diffusion Policy (CVPR 2024 · similarity 0.88)
- DynaGuide: Steering Diffusion Polices with Active Dynamic Guidance (NeurIPS 2025 · similarity 0.88)
- Understanding and Improving Training-free Loss-based Diffusion Guidance (NeurIPS 2024 · similarity 0.87)
- DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising (NeurIPS 2026 · similarity 0.86)
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
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.
