Research map / Image generation / Fast diffusion sampling
Block-wise Adaptive Caching for Accelerating Diffusion Policy
Kangye Ji, Yuan Meng, Hanyun Cui, Ye Li, Jianbo Zhou, Shengjia Hua, Lei Chen, Zhi Wang
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 Adaptive Caching Scheduler (unified across blocks) instead of uniform intervals (ablation) | BAC / DP-T · simulation benchmarks | success rate | Improved |
| apply scheduler naively per block (Block-wise ACS) vs unified ACS (ablation) | BAC / DP-T · simulation benchmarks | success rate | Hurt |
| add Bubbling Union Algorithm to block-wise ACS (ablation) | BAC / DP-T · simulation benchmarks | success rate: highest score of 0.79 | Improved |
| use BAC (S=10) instead of full precision | DP-T · Robomimic PH + Push-T | avg success rate: 0.76 -> 0.79 | No clear effect |
| use BAC (S=10) instead of full precision | DP-T · Robomimic MH | avg success rate: 0.76 -> 0.77 | No clear effect |
| use BAC (S=7) instead of S=10 (fewer cache updates) | BAC / DP-T · Robomimic MH | avg success rate: 0.77 -> 0.60 | Hurt |
| use BAC (S=10) instead of Uniform(fastest) | DP-T · Block-Pushing and Kitchen | avg success rate: 0.66 -> 0.98 | Improved |
| use BAC instead of TeaCache-style adaptive caching | DP-T · Block-Pushing and Kitchen | avg success rate: 0.52 -> 0.98 | Improved |
| replace DDPM K=100 with BAC (S=7, n=3) on real robot | DP-T · real-world bag task | success rate: 3% -> 71% | Improved |
| apply BAC to RDT-1B VLA at S=2 | RDT-1B · ManiSkill | success rate: 3.55x speedup | No clear effect |
Closest papers by meaning
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.89)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.89)
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.89)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.88)
- Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching (NeurIPS 2024 · similarity 0.88)
- Accelerating Diffusion Transformers with Token-wise Feature Caching (ICLR 2025 · similarity 0.88)
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.
