Research map / Reinforcement learning / Offline RL
Guided Flow Policy: Learning from High-Value Actions in Offline Reinforcement Learning
Franki NGUIMATSIA TIOFACK, Théotime Le Hellard, Fabian Schramm, Nicolas Perrin-Gilbert, Justin Carpentier
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 value-aware guidance (GFP) over FQL | FQL · OGBench cube-double-noisy | success rate: 38.2 -> 63.1 | Improved |
| add value-aware guidance (GFP) over FQL | FQL · OGBench cube-triple-noisy | success rate: 3.5 -> 24.5 | Improved |
| add value-aware guidance (GFP) over FQL | FQL · OGBench average (105 tasks) | average score: 46.7 -> 53.2 | Improved |
| add value-aware guidance (GFP) over ReBRAC | ReBRAC · Minari average (21 tasks) | average score: 65.9 -> 74.1 | Improved |
| use GFP over ReBRAC on D4RL | ReBRAC · D4RL average (18 tasks) | average score: 64.8 -> 63.0 | No clear effect |
| tune discount factor to larger value for ReBRAC | ReBRAC · OGBench antmaze-large-navigate | success rate: 81 -> 95.9 | Improved |
| use bigger minibatch for FQL | FQL · OGBench antmaze-giant-navigate | success rate: 9 -> 16.3 | Improved |
| add modified Bellman target using VaBC policy (ablation) | GFP · OGBench cube-double-play | success rate: ~28 -> 47.2 | Improved |
| add modified Bellman target using VaBC policy (ablation) | GFP · OGBench puzzle-4x4-play | success rate: 26.1 -> ~22 | Hurt |
| replace soft-max guidance with AWR-style weight (ablation) | GFP · OGBench (65 tasks) | average score: 38.6 -> 35.9 | Hurt |
| replace soft-max guidance with AWR-style weight (ablation) | GFP · OGBench cube-triple-noisy | success rate: 24.5 -> 11.2 | Hurt |
| replace soft-max guidance with AWR-style weight (ablation) | GFP · OGBench puzzle-3x3-play | success rate: 23.1 -> 28.3 | Improved |
Closest papers by meaning
- Learning Multimodal One-step Flow Policy via Value-weighted Optimal Transport (NeurIPS 2026 · similarity 0.91)
- Flow Q-Learning (ICML 2025 · similarity 0.90)
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.90)
- DeFlow: Decoupling Behavior-Prior Modeling and Value Maximization for Offline Policy Extraction (NeurIPS 2026 · similarity 0.89)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.89)
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.
