dynamicslab/hydrogym

An RL-Gym for Challenge Problems in Data-Driven Modeling and Control of Fluid Dynamics.

What it solves

HydroGym provides a standardized way to apply reinforcement learning (RL) to fluid dynamics and flow control. It bridges the gap between complex Computational Fluid Dynamics (CFD) solvers and RL libraries by providing a consistent interface, allowing researchers to train agents to control fluid flows without needing to build custom integrations for every different solver.

How it works

The platform implements a Gymnasium-compatible API, making it compatible with popular RL libraries like Stable-Baselines3 and RLlib. It integrates six different solver backends—including Finite Element (Firedrake), Lattice Boltzmann (MAIA LBM), Finite Volume (MAIA FV), Spectral Element (NEK5000), and differentiable solvers (JAX-Fluids)—to simulate fluid behavior across 2D and 3D environments. It supports high-performance computing (HPC) via MPI-parallelized solvers and GPU acceleration for efficient training.

Who it’s for

It is designed for researchers and engineers working in fluid dynamics, flow control, and reinforcement learning who need a scalable, research-ready platform to test RL agents on both canonical benchmarks and high-Reynolds number turbulent flows.

Highlights

  • Extensive Environment Library: Over 61 pre-configured environments covering cylinder wakes, airfoils, and turbulent channel flows.
  • Multi-Backend Support: Integrates six distinct CFD solver types, including fully differentiable solvers for gradient-based optimization.
  • High Scalability: Optimized for everything from local workstations to exascale HPC systems with MPI and GPU support.
  • Advanced Analysis Tools: Includes built-in support for modal decompositions (DMD, POD) and global stability analysis.

Related

  • Project
    Farama-Foundation/Gymnasium

    A standard Python API and environment suite for developing and comparing reinforcement learning algorithms.

  • Project
    Farama-Foundation/Gymnasium-Robotics

    A collection of Reinforcement Learning robotic environments powered by the MuJoCo physics engine and the Gymnasium API for training AI agents in manipulation and navigation tasks.

  • Project
    Farama-Foundation/Minigrid

    A collection of lightweight, discrete grid-world environments for Reinforcement Learning research, supporting goal-oriented tasks and grounded language learning.

  • Project
    X-GenGroup/Flow-Factory

    A unified framework for applying reinforcement learning to fine-tune diffusion and flow-matching models across image, video, and audio-video tasks.

  • Project
    alibaba/ROLL

    An efficient scaling library for reinforcement learning with LLMs that optimizes training and inference across large GPU clusters using a distributed Ray-based architecture.