Research map / Reinforcement learning / Offline RL
Continuity-Regularized Flow Matching for Offline Reinforcement Learning
Xiaocong Chen, Siyu Wang, Lina Yao
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 Jacobian (PDE) regularizer together with Beta timestep sampling (ablation) | PQL · D4RL Hopper medium-expert | normalized score: 98.7 -> 111.8 | Improved |
| add Beta timestep sampling to the PDE-regularized model (vs uniform sampling) (ablation) | PQL · D4RL HalfCheetah medium-expert | normalized score: 76.2 -> 90.2 | Improved |
| add PDE regularizer to Beta-sampled model (ablation) | PQL · D4RL HalfCheetah medium-expert | normalized score: 84.3 -> 90.2 | Improved |
| add PDE regularizer with uniform time sampling (ablation) | FQL-style flow policy · D4RL Gym-MuJoCo | normalized score | Hurt |
| add Beta timestep sampling without regularizer (ablation) | PQL-PDE · D4RL Gym-MuJoCo | normalized score | Improved |
| penalize full Jacobian Frobenius norm instead of divergence only (ablation) | PQL · D4RL Gym-MuJoCo, harder Medium-Expert and Medium-Replay | normalized score | Improved |
| penalize full Jacobian instead of divergence on Medium datasets (ablation) | PQL · D4RL Gym-MuJoCo Medium | normalized score | No clear effect |
| vary Beta shape alpha in [2, 5] and lambda in [0.01, 0.02] (ablation) | PQL · D4RL Gym-MuJoCo | normalized score | No clear effect |
| PQL vs FQL on D4RL Gym-MuJoCo medium-expert Hopper | FQL · D4RL Hopper medium-expert | normalized score: 21 -> 112 | Improved |
| PQL on OGBench long-horizon navigation and manipulation vs FQL | FQL · OGBench puzzle-4x4-play | success rate (%): 11 -> 16 | Improved |
| PQL on HalfCheetah medium vs CAC | PQL · D4RL HalfCheetah medium | normalized score: 54 vs 69 (CAC) | No clear effect |
Closest papers by meaning
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.92)
- Flow Q-Learning (ICML 2025 · similarity 0.91)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.89)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.89)
- Online Reward-Weighted Fine-Tuning of Flow Matching with Wasserstein Regularization (ICLR 2025 · similarity 0.89)
- Flow-Based Policy for Online Reinforcement Learning (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.
