Research map / Reinforcement learning / Policy gradient and actor-critic
Policy-DRIFT: Dynamic Reward-Informed Flow Trajectory Steering
Atharva Mahajan, Abhijeet Yashwant Vishwasrao, Yuning Wang, Ricardo Vinuesa
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Policy-DRIFT, a method for steering flow trajectories with reward information to reduce skin-friction drag in wall-bounded turbulence via deep reinforcement learning-based flow control.
Problem
Reducing skin-friction drag in wall-bounded turbulent flows for real-time flow control; the abstract suggests DRL performance is limited by something other than the algorithm (text truncated).
Approach
Abstract is truncated; the method name suggests dynamic reward-informed steering of flow trajectories for DRL-based flow control, but the specifics are not stated in the available text.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace scalar reward with TRG-guided CFM full-field target tracking | TD3 policy · Turbulent channel Re_tau=180 | drag reduction (%): 42.86 -> 48.95 (vs TD3-WEN) | Improved |
| use Policy-DRIFT instead of TD3-WSE | TD3 policy · Turbulent channel Re_tau=180 | actuation energy: 3.692 -> 0.0990 (x1e-1) | Improved |
| use Policy-DRIFT instead of opposition control | opposition control · Turbulent channel Re_tau=180 | drag reduction (%): 35.23 -> 48.95 | Improved |
| apply TRG correction after the velocity step (post-placement) instead of before (ablation) | TRG · Turbulent channel Re_tau=180 | reward hacking / realisability | Hurt |
| add Terminal Reward Guidance to unguided CFM sampling (ablation) | CFM · held-out uncontrolled snapshots | drag reduction: +0.026 to +0.030 DR | Improved |
| raise TRG guidance scale from 5 to 20 (ablation) | Policy-DRIFT · Turbulent channel Re_tau=180 | drag reduction (%): 48.95 -> 36.58 | Hurt |
| raise TRG guidance scale from 5 to 10 (ablation) | Policy-DRIFT · Turbulent channel Re_tau=180 | drag reduction (%): 48.95 -> 47.22 | No clear effect |
| use SAC instead of TD3 under WSE reward (ablation) | DRL policy · Turbulent channel Re_tau=180 | drag reduction (%): 42.13 -> 32.31 | Hurt |
| add actuation-energy penalty to the DRL reward (WEN vs WSE) (ablation) | TD3 · Turbulent channel Re_tau=180 | drag reduction (%) and E_act: DR 42.13 -> 42.86; E_act 3.692 -> 0.0609 | Mixed |
Limitations
- Full text unavailable; only the truncated abstract was available
Closest papers by meaning
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.87)
- Learning Multimodal One-step Flow Policy via Value-weighted Optimal Transport (NeurIPS 2026 · similarity 0.86)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.86)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.86)
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- ReinFlow: Fine-tuning Flow Matching Policy with Online Reinforcement Learning (NeurIPS 2025 · 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.
