Research map / Reinforcement learning / Visual and representation RL
Primary-Fine Decoupling for Action Generation in Robotic Imitation
Xiaohan Lei, Min Wang, Wengang Zhou, Xingyu Lu, Houqiang Li
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 mode-conditioned MeanFlow policy on top of primary mode policy (ablation) | PF-DAG without MeanFlow · Adroit/DexArt/MetaWorld weighted | weighted success: 0.01 -> 0.72 | Improved |
| add primary mode policy (ablation) | PF-DAG without primary mode · Adroit/DexArt/MetaWorld weighted | weighted success: 0.56 -> 0.72 | Improved |
| use K=64 modes instead of 8 (ablation) | PF-DAG · Adroit/DexArt/MetaWorld weighted | weighted success: 0.61 -> 0.72 | Improved |
| use K=64 modes instead of 1024 (ablation) | PF-DAG · Adroit/DexArt/MetaWorld weighted | weighted success: 0.58 -> 0.72 | Improved |
| use K=64 instead of 16, 32 or 128 (ablation) | PF-DAG · Adroit/DexArt/MetaWorld weighted | weighted success: 0.70 -> 0.72 | No clear effect |
| use VQ-VAE instead of k-means tokenization (ablation) | PF-DAG · Adroit/DexArt/MetaWorld weighted | weighted success: 0.70 -> 0.72 | No clear effect |
| replace CFM with MeanFlow (1-NFE) (ablation) | PF-DAG · Adroit/DexArt/MetaWorld weighted | success rate: 0.69 -> 0.72 | Improved |
| use 10-NFE Euler or Runge-Kutta instead of 1-NFE Euler for CFM (ablation) | CFM variant · Adroit/DexArt/MetaWorld weighted | success rate: 0.69 -> 0.69, 0.69 -> 0.68 | No clear effect |
| use PF-DAG instead of DP3 on real-world Wipe Table | PF-DAG · Wipe Table (real world) | total jerk: 1.25 -> 0.45 | Improved |
| increase number of function evaluations at inference (ablation) | PF-DAG · Simulation benchmarks | success rate | No clear effect |
| use PF-DAG instead of DP3 on MetaWorld Hard (5) | PF-DAG · MetaWorld Hard (5) | success rate: 0.35 -> 0.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlowPolicy — FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · similarity 0.88)
- Improving Generative Behavior Cloning via Self-Guidance and Adaptive Chunking (NeurIPS 2025 · similarity 0.87)
- Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry (NeurIPS 2026 · similarity 0.87)
- Chain-of-Action: Trajectory Autoregressive Modeling for Robotic Manipulation (NeurIPS 2025 · similarity 0.87)
- Phoenix: A Motion-based Self-Reflection Framework for Fine-grained Robotic Action Correction (CVPR 2025 · similarity 0.87)
- ET-SEED: EFFICIENT TRAJECTORY-LEVEL SE(3) EQUIVARIANT DIFFUSION POLICY (ICLR 2025 · similarity 0.87)
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.
