Basic Usage

This section covers everything you need to get started with FluidGym — from choosing an environment to running your first training loop.

FluidGym provide GPU-accelerated fluid simulations in a Gymnasium-like interface. Each environment simulates a distinct flow control problem and exposes structured observations (velocity fields, pressure, sensor readings) and continuous action spaces for actuator control.

Quick start — create an environment, interact with it, and save a visualisation:

import fluidgym

env = fluidgym.make("CylinderJet2D-easy-v0")
obs, info = env.reset(seed=42)

for _ in range(50):
    action = env.sample_action()
    obs, reward, terminated, truncated, info = env.step(action)
    env.render()
    if terminated or truncated:
        break

env.save_gif("cylinder.gif")

Browse the sections below to learn about the available environments, how to plug FluidGym into common RL frameworks, the observation wrappers, and rendering options.