Research map / Reinforcement learning / Policy gradient and actor-critic
Q-Flow: Stable and Expressive Reinforcement Learning with Flow-based Policy
JaeHyeok Doo, Byeongguk Jeon, Seonghyeon Ye, Kimin Lee, Minjoon Seo
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 |
|---|---|---|---|
| use Q-Flow intermediate-value gradient matching instead of FQL one-step distillation | FQL · OGBench standard setting average | average score: 43.8 -> 54.4 | Improved |
| use Q-Flow instead of FQL on antmaze-giant | FQL · antmaze-giant (standard) | success rate (%): 9 -> 40 | Improved |
| use Q-Flow instead of FQL on puzzle-3x3 | FQL · puzzle-3x3 (standard) | success rate (%): 30 -> 49 | Improved |
| use Q-Flow instead of QAM-E in advanced setting | QAM-E · OGBench advanced setting average | average score: 44.8 -> 47.5 | Improved |
| use Q-Flow instead of QAM-E on puzzle-4x4-sparse | QAM-E · puzzle-4x4-sparse (advanced) | success rate (%): 36 -> 0 | Hurt |
| use Q-Flow instead of QAM-E on cube-double | QAM-E · cube-double (advanced) | success rate (%): 65 -> 38 | Hurt |
| add intermediate value gradient matching in place of BPTT through the ODE solver (ablation) | Q-Flow with learned intermediate value · 5 OGBench default tasks (standard) | success rate | Improved |
| use flow-consistent intermediate value instead of outer Q function guidance (ablation) | Q-Flow gradient matching · 5 OGBench default tasks (standard) | success rate | Improved |
| replace FBRAC BPTT with Q-Flow in training | FBRAC · Puzzle-4x4 training cost, 50 flow steps | training time per step: approximately 2x faster | Improved |
| use Q-Flow instead of FQL during online fine-tuning | FQL · 5 OGBench default tasks (offline-to-online) | average success: +23% points | Improved |
Closest papers by meaning
- Flow Q-Learning (ICML 2025 · similarity 0.92)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.90)
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.89)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.89)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · 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.
