Stable-Baselines3

In order to use Stable-Baselines3 in the MARL setting, we can leverage the VecEnv interface. FluidGym environments are compatible with the SB3 VecEnv interface, allowing seamless integration with various reinforcement learning libraries that support SB3. Here is a simple example from examples/interfaces/sb3_vec_env.py:

import numpy as np

import fluidgym
from fluidgym.wrappers import FlattenObservation
from fluidgym.integration.sb3 import VecFluidEnv

fluid_env = fluidgym.make(
    "Airfoil3D-easy-v0",
    use_marl=True,
)

# We flatten the observation space to receive a 1D array of observations
fluid_env = FlattenObservation(fluid_env)

# For the SB3 VecEnv interface, wrap the FluidGym environment. This will give us a
# vectorized environment with a pseudo-enviroment for each agent
env = VecFluidEnv(fluid_env)

obs = env.reset(seed=42)

for i in range(50):
    actions = np.array([env.action_space.sample() for _ in range(env.num_envs)])
    obs, reward, done, info = env.step(actions)
    print(f"Step: {i}; Rewards:", reward.tolist())

    env.render()

    if np.any(done):
        break

env.save_gif("cylinder.gif")