Skip to main content

HydroGym

A Reinforcement Learning Platform for Fluid Dynamics

Getting Started

About the Project

HydroGym is a transparent, extensible & comprehensive platform for applying reinforcement learning to fluid dynamics flow control. With environments ranging from canonical benchmarks to turbulent flows, HydroGym provides a standardized Gymnasium-compatible interface for training RL agents on challenging CFD problems.

Diverse Environments

88 pre-configured environments across 6 CFD solvers. Flow environments are supported by multiple backends: Finite Element (Firedrake), Lattice Boltzmann (MAIA LBM), Finite Volume (MAIA FV), Spectral Element (NEK5000), and fully Differentiable solvers (JAX-Fluids)

Scalable Implementation

Highly optimized GPU & CPU backends for efficient RL deployment ranging from local workstations to exascale HPC systems

Research Ready

Includes checkpoints, observation strategies, and reward formulations managed by a complementary HuggingFace repository. Gymnasium-compatible API works with Stable-Baselines3, RLlib, and other RL libraries

Join researchers training flow-control agents on canonical benchmarks in minutes.

  • NACA 0012 Airfoil
    NACA 0012 Airfoil

    Zero shot evaluation at Rec = 200,000. Drag reduction trained on channel flow of Reτ = 206, and evaluated on a large scale airfoil.

  • Fluidic Pinball
    Fluidic Pinball

    Multi-body wake interactions at Re = 30 - 150. Coordinated control of three cylinders demonstrates chaos suppression.

  • Turbulent Boundary Layer
    Turbulent Boundary Layer

    Reinforcement learning agent performs net power savings with travelling wave control for Reτ = 200, 1550, and 2200

  • Cylinder
    Cylinder

    2D & 3D flows at Re = 100 - 3,900. Drag reduction via rotation and jet actuation. Achieves more than 20% drag reduction.

  • Turbulent Channel Flow (Differentiable)
    Turbulent Channel Flow (Differentiable)

    Wall-shear stress reduction at Reτ = 180. Gradient-enhanced training with JAX for efficient optimization.

  • NACA 0012 Airfoil
    NACA 0012 Airfoil

    Gust mitigation at Re = 100 - 50,000. Transverse gust encounters with load alleviation strategies.

  • Kolmogorov Flow (Differentiable)
    Kolmogorov Flow (Differentiable)

    Extreme event mitigation in 2D turbulence. Control energy bursts and enhance mixing efficiency.

  • Open Cavity Flow
    Open Cavity Flow

    Shear layer stabilization at Re = 4,140 - 7,500. Acoustic feedback disruption through leading-edge control.