Gymnasium Library in Reinforcement Learning

Posted on Edited on In Views:

Sources:

  1. --> Gymnasium github repo
  2. --> Documentation

Gymnasium is the successsor of the famous OpenAI Gym library. It's highly recommended to move to gymnasium since gym is not being maintained.

Gymnasium is maintained by the Farama Foundation, an open-Sources nonprofit foundation.

Install

On Linux:

1
pip install gymnasium

Tested on Ubuntu22.04 X86_64 with python=3.9.

This does not include dependencies for all families of environments. According to your requirement, you may need to install dependencies for one family like pip install "gymnasium[atari]" or use pip install "gymnasium[all]".

Customize

Make your own custom environment

https://gymnasium.farama.org/environments/box2d/car_racing/

https://gymnasium.farama.org/_modules/gymnasium/utils/play/

Flappybird

1
2
3
4
conda create -n flappy_bird python==3.11 # Must be this version
pip install flappy-bird-env
pip install gymnasium\[other\]
python -m flappy_bird_env

Scripts

Box2D

For box2d environments, we must also install:

1
pip install gymnasium[box2d]

CarRacing

To play:

1
2
3
git clone [email protected]:Farama-Foundation/Gymnasium.git
cd Gymnasium
python gymnasium/envs/box2d/car_racing.py

Atari

For Atari environments, we must also install:

1
pip install "gymnasium[atari, accept-rom-license]"

Errors

  1. OpenGL error: when running

    1
    python gymnasium/envs/box2d/car_racing.py

    The output is

    1
    2
    3
    4
    5
    6
    7
    libGL error: MESA-LOADER: failed to open swrast: /usr/lib/dri/swrast_dri.so: cannot open shared object file: No such file or directory (search paths /usr/lib/x86_64-linux-gnu/dri:\$${ORIGIN}/dri:/usr/lib/dri, suffix _dri)
    libGL error: failed to load driver: swrast
    X Error of failed request:  BadValue (integer parameter out of range for operation)
      Major opcode of failed request:  149 (GLX)
      Minor opcode of failed request:  3 (X_GLXCreateContext)
      Value in failed request:  0x0
      Serial number of failed request:  153

    This problem occurs randomly. I can successfully run the experiment for one day and meet this error the next day with not any conguguration changed.

    The solution is simple:

    1
    conda install -c conda-forge libffi

    Works on my Ubuntu22.04 x86_64 Python=3.10 environment.

  2. Can't create Atari environment:

    1
    gymnasium.error.NamespaceNotFound: Namespace ALE not found. Have you installed the proper package for ALE?

    Solution: There seems sth wrong with gymnasium. We have to use gym instead of gymnasium, and

    1
    pip install "gym[atari, accept-rom-license]"

    Run:

    1
    2
    3
    4
    import gym # or "import gymnasium as gym"

    if __name__ == '__main__':
        env = gym.make("ALE/Breakout-v5")

    The output is

    1
    2
    A.L.E: Arcade Learning Environment (version 0.8.1+53f58b7)
    [Powered by Stella]