Research map / Reinforcement learning / Visual and representation RL
Plug-and-Play Benchmarking of Reinforcement Learning Algorithms for Large-Scale Flow Control
Jannis Becktepe, Aleksandra Franz, Nils Thuerey, Sebastian Peitz
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 a 2D-trained policy transferred to 3D instead of training in 3D | PPO/SAC on CylinderJet3D · CylinderJet3D easy and hard | drag reduction | Improved |
| use a 2D-trained policy transferred to 3D on medium difficulty | PPO/SAC on CylinderJet3D · CylinderJet3D medium | drag reduction | No clear effect |
| train on small TCF domain and transfer to large domain instead of training on large | MARL policy · TCF large channel | drag reduction | Improved |
| use SAC instead of PPO | SB3 agents · FluidGym environments | normalized relative improvement | Improved |
| use MA-SAC instead of MA-PPO | multi-agent agents · FluidGym environments | normalized score | No clear effect |
| use differentiable predictive control (DPC) instead of PPO/SAC | policy learning · CylinderJet2D-easy-v0 | training speed: one and two orders of magnitude over PPO and SAC | Improved |
| use DPC instead of TD-MPC and SAC | policy learning · RBC2D-hard-v0 | training reward | No clear effect |
| use DPC instead of standard RL | policy learning · all FluidGym classes | training wall-clock time: 1.5-2× higher | Hurt |
| use TD-MPC instead of PPO/SAC | policy learning · FluidGym environments | training runtime | Improved |
| coarsen RBC grid to resolution 96 from 192 (ablation) | simulation · RBC uncontrolled | mean Nusselt number: 4.786 -> 4.896 | No clear effect |
Closest papers by meaning
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- ReinFlow: Fine-tuning Flow Matching Policy with Online Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.83)
- Normalizing Flows are Capable Models for Continuous Control (NeurIPS 2025 · similarity 0.83)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.83)
- DiffPhyCon: A Generative Approach to Control Complex Physical Systems (NeurIPS 2024 · similarity 0.83)
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.
