Research map / Reinforcement learning / Visual and representation RL
FocalPolicy: Frequency-Optimized Chunking and Locally Anchored Flow Matching for Coherent Visuomotor Policy
Qian He, Zhenshuo Yang, Wenqi Liang, Chunhui Hao, Nicu Sebe, Jiandong Tian
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 Foresight Composite Objective and Locally Anchored Sampling to consistency flow policy | FocalPolicy · Adroit and MetaWorld 53 tasks | success rate (%): 79.9 -> 83.6 (vs DP3) | Improved |
| add Locally Anchored Sampling (ablation) | FocalPolicy · 24 tasks ablation | average success rate (%): 64.3 -> 77.9 | Improved |
| add FCO on top of LAS (ablation) | FocalPolicy · 24 tasks ablation | average success rate (%): 54.1 -> 64.3 | Improved |
| use frequency-domain loss over multi-chunk instead of time-domain loss over long horizon (ablation) | FCO · 24 tasks ablation | average success rate (%): 61.1 -> 77.9 | Improved |
| add proximal time-domain loss alongside frequency-domain loss (ablation) | FCO · 24 tasks ablation | average success rate (%): 72.8 -> 77.9 | Improved |
| sample anchor time r stochastically near terminal instead of fixing r = 1 (ablation) | LAS · 24 tasks ablation | average success rate (%): 73.3 -> 77.9 | Improved |
| bias anchor time r toward terminal (mu 4.0, sigma 1.6) instead of near 0.6 (ablation) | LAS · 7 tasks | average success rate (%): 13.9 -> 69.7 | Improved |
| integrate FCO into DP3 | DP3 · 7 simulation tasks | average success rate (%): 57.8 -> 60.6 | Improved |
| integrate LAS into FlowPolicy | FlowPolicy · 7 simulation tasks | average success rate (%): 55.4 -> 63.3 | Improved |
| aggregate N=3 chunks instead of N=5 (H=4) (ablation) | FocalPolicy · 24 tasks | average success rate (%): 64.6 -> 77.9 | Improved |
| increase action chunk size H from 4 to 8 or 12 (ablation) | FocalPolicy · 24 tasks | average success rate (%): 77.9 -> 70.3 (H=8, N=3) | Hurt |
| use FocalPolicy instead of FlowPolicy for inter-chunk coherence | FocalPolicy · MetaWorld Sweep Into | Action Total Variation: 1.66 -> 1.45 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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 5 of 5 shared results
- FreqPolicy* — FreqPolicy: Frequency Autoregressive Visuomotor Policy with Continuous Tokens (NeurIPS 2025) · better on 2 of 2 shared results
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 0.92)
- FreqPolicy: Frequency Autoregressive Visuomotor Policy with Continuous Tokens (NeurIPS 2025 · similarity 0.90)
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · similarity 0.88)
- Falcon: Fast Visuomotor Policies via Partial Denoising (ICML 2025 · similarity 0.88)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · similarity 0.88)
- Frequency‑Aware Flow Matching for Continuous and Consistent Robotic Action Generation (NeurIPS 2026 · 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.
