Research map / Image generation / Fast diffusion sampling
Sparse ActionGen: Accelerating Diffusion Policy with Real-time Pruning
Kangye Ji, Jianbo Zhou, Yuan Meng, Ye Li, Hanyun Cui, 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 |
|---|---|---|---|
| replace the real-time observation-conditioned pruner with a fixed pruning schedule (ablation) | SAG · Average over 10 tasks | success rate (%): 87 -> 83 | Hurt |
| add one-for-all reusing instead of block-wise reusing (ablation) | SAG · Average over 10 tasks | success rate (%): 74 -> 87 | Improved |
| add global sparsity loss instead of block-wise uniform loss (ablation) | SAG · Average over 10 tasks | success rate (%): 53 -> 87 | Improved |
| add sinusoidal positional encoding to the pruner (ablation) | SAG · Robomimic Tool Hang | success rate (%): 28 -> 54 | Improved |
| raise target pruning rate from 91% to 97% (ablation) | SAG · Average over 10 tasks | success rate (%): 85 -> 51 | Hurt |
| lower target pruning rate from 91% to 80% (ablation) | SAG · Average over 10 tasks | success rate (%): 85 -> 85 | No clear effect |
| combine SAG with DPM-Solver++ 50 steps | DPM50 · Robomimic MH Square | success rate (%): 72 -> 78 | Improved |
| use fixed caching schedule instead of rollout-adaptive pruning (EfficientVLA) | DP-T · Franka Kitchen p4 | success rate (%): 99 -> 0 | Hurt |
| apply SAG to DP-T on real-world Assembly Line | DP-T · Real-world Assembly Line | success rate (%): 26 -> 58 | Improved |
| apply SAG to DP-T on real-world Stack Bowls | DP-T · Real-world Stack Bowls | success rate (%): 82 -> 78 | No clear effect |
Closest papers by meaning
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.90)
- GenPO: Generative Diffusion Models Meet On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- FASTER: Rethinking Real-Time Flow VLAs (NeurIPS 2026 · similarity 0.88)
- Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising (NeurIPS 2025 · similarity 0.88)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.88)
- Falcon: Fast Visuomotor Policies via Partial Denoising (ICML 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.
