Research map / Image generation / Fast diffusion sampling
One-Step Flow Policy: Self-Distillation for Fast Visuomotor Policies
Shaolong Li, Lichao Sun, Yongchao Chen
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 |
|---|---|---|---|
| one-step OFP versus 100-step Diffusion Policy | OFP · Adroit+DexArt (2D image, 7 tasks) | success rate average: 64.2 -> 68.3 | Improved |
| one-step OFP versus 100-step FM Policy | OFP · Adroit+DexArt (2D image, 7 tasks) | success rate average: 67.2 -> 68.3 | Improved |
| OFP self-distillation instead of MeanFlow-based MP1 (one step) | OFP · Adroit+DexArt (2D image, 7 tasks) | success rate average: 60.5 -> 68.3 | Improved |
| one-step OFP versus 100-step DP3 | OFP · 56 tasks (3D pointcloud) | success rate average: 66.4 -> 71.6 | Improved |
| one-step OFP versus 100-step 3D FM Policy | OFP · 56 tasks (3D pointcloud) | success rate average: 59.8 -> 71.6 | Improved |
| one-step OFP versus Adroit Bucket DP at NFE=100 | OFP · DexArt Bucket (2D image) | success rate: 34.7 -> 33.3 | No clear effect |
| increase OFP sampling steps from 1 to 4 (ablation) | OFP · DexArt 4 tasks (2D, few-step) | success rate average: 64.5 -> 66.2 | Improved |
| increase CP sampling steps from 1 to 4 (ablation) | CP · DexArt 4 tasks (2D, few-step) | success rate average: 60.7 -> 64.2 | Improved |
| replace 100-step DP3 sampling with one-step OFP | OFP · Inference latency per action chunk (56 tasks) | time (ms): 3225.67 -> 17.58 | Improved |
| increase demonstrations for OFP (ablation) | OFP · DexArt Faucet data scaling | success rate: 32.7% to 51.0% | Improved |
| increase demonstrations beyond 100 for MP1 (ablation) | MP1 · DexArt Faucet data scaling | success rate | Hurt |
| integrate OFP into pi0.5 at one step versus original 10-step pi0.5 | pi0.5 · RoboTwin 2.0 (4 tasks) | success rate average: 94.7% | Improved |
Closest papers by meaning
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.93)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · similarity 0.92)
- Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry (NeurIPS 2026 · similarity 0.90)
- Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control (NeurIPS 2026 · similarity 0.89)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · similarity 0.89)
- Frequency‑Aware Flow Matching for Continuous and Consistent Robotic Action Generation (NeurIPS 2026 · 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.
