Research map / Reinforcement learning / Visual and representation RL
FlowMAP: Flow Matching for Generalizable Agent Planning
Jiarun Fu, Lizhong Ding, Ye Yuan, Qiuning Wei, Zhaohuan Linghu, Yurong Cheng, Changsheng Li, Tianlong Gu, Liang Chang, Guoren 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 planning-time flow (VTFM) on top of the RL backbone (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 6.0 ± 0.4 -> 9.0 ± 0.2 | Improved |
| add value-transport flow matching to the consistency regularizer (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 6.2 ± 0.4 -> 9.0 ± 0.2 | Improved |
| add the consistency term to the RL baseline (ablation) | RL-only baseline · Minecraft (Diamond) | episodic return: 6.0 ± 0.4 -> 6.2 ± 0.4 | Improved |
| use value-shaped target instead of uniform target (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 6.5 ± 0.5 -> 9.0 ± 0.2 | Improved |
| use distributional transport instead of pointwise value ascent (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 6.7 ± 0.4 -> 9.0 ± 0.2 | Improved |
| add consistency loss to VTFM (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 7.1 ± 0.6 -> 9.0 ± 0.2 | Improved |
| co-train the flow with the policy instead of freezing it (ablation) | FlowMAP · Minecraft (Diamond) | episodic return: 7.6 ± 0.3 -> 9.0 ± 0.2 | Improved |
| add planning-time flow (VTFM) on top of the RL backbone (ablation) | FlowMAP · BSuite | mean episodic return: 54.2 -> 69.0 | Improved |
| add value-transport flow matching to the consistency regularizer (ablation) | FlowMAP · BSuite | mean episodic return: 55.5 -> 69.0 | Improved |
| use value-shaped target instead of uniform target (ablation) | FlowMAP · BSuite | mean episodic return: 53.8 -> 69.0 | Improved |
| use distributional transport instead of pointwise value ascent (ablation) | FlowMAP · BSuite | mean episodic return: 61.3 -> 69.0 | Improved |
| add consistency loss to VTFM (ablation) | FlowMAP · BSuite | mean episodic return: 57.8 -> 69.0 | Improved |
Closest papers by meaning
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.88)
- Learn to Follow: Decentralized Lifelong Multi-Agent Pathfinding via Planning and Learning (AAAI 2024 · similarity 0.88)
- Traffic Flow Optimisation for Lifelong Multi-Agent Path Finding (AAAI 2024 · similarity 0.87)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.87)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.86)
- TodoEvolve: Learning to Architect Agent Planning Systems (ICML 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.
