Research map / Reinforcement learning / Policy gradient and actor-critic
FlowCritic: Bridging Value Estimation with Flow Matching in Reinforcement Learning
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 flow-matching value critic to PPO | PPO · 12 Isaac Gym control tasks | final episodic return: Ant 5340.04 -> 9345.29 | Improved |
| add flow-matching value critic to PPO | PPO · AllegroHand | final episodic return: 604.59 -> 1044.12 | Improved |
| add flow-matching value critic to PPO | PPO · Task-averaged CVaR 0.1 | CVaR: 0.269 -> 0.968 | Improved |
| adapt existing value estimation methods (AVC, CVE, QD, IQN, C51, TQC) to PPO | PPO · 12 control tasks | final episodic return | Mixed |
| add CoV reweighting (ablation) | FlowCritic · Ant | episodic return | Improved |
| add velocity-field clipping (ablation) | FlowCritic · Ant | episodic return | Improved |
| add truncated sampling (ablation) | FlowCritic · Ant | episodic return | Improved |
| increase Euler steps to 10 (default 5) (ablation) | FlowCritic · Ant | episodic return | Improved |
| increase Euler steps from 10 to 20 (ablation) | FlowCritic · Ant | episodic return | No clear effect |
| CoV temperature alpha 0.1 vs 0.01 (ablation) | FlowCritic · Ant | episodic return | No clear effect |
| clipping threshold delta 0.2 vs 0.1 (ablation) | FlowCritic · Ant | episodic return | No clear effect |
| increase N from 10 to 50 with M=1 (ablation) | FlowCritic · Ant | episodic return | Hurt |
Closest papers by meaning
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.88)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Designing Reinforcement Learning for Diffusion Models: A Unified Path-Space View (NeurIPS 2026 · similarity 0.87)
- Random Policy Evaluation Uncovers Policies of Generative Flow Networks (ICML 2025 · similarity 0.87)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.87)
- Enhancing Value Function Estimation through First-Order State-Action Dynamics in Offline Reinforcement Learning (ICML 2024 · similarity 0.87)
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.
