This is a gym compatible version of various games for reinforcement learning.
For PyGame Learning Environment, the default observation is a non-visual state representation of the game.
For MinAtar, the default observation is a visual input of the game.
-
PyGame learning environment:
- Catcher-PLE-v0
- FlappyBird-PLE-v0
- Pixelcopter-PLE-v0
- PuckWorld-PLE-v0
- Pong-PLE-v0
-
MinAtar:
- Asterix-MinAtar-v0
- Asterix-MinAtar-v1
- Breakout-MinAtar-v0
- Breakout-MinAtar-v1
- Freeway-MinAtar-v0
- Freeway-MinAtar-v1
- Seaquest-MinAtar-v0
- Seaquest-MinAtar-v1
- SpaceInvaders-MinAtar-v0
- SpaceInvaders-MinAtar-v1
-
Exploration games:
- NChain-v1
- LockBernoulli-v0
- LockGaussian-v0
- SparseMountainCar-v0
- DiabolicalCombLock-v0
Please read the instruction here.
-
On OSX:
brew install sdl sdl_ttf sdl_image sdl_mixer portmidi pip install pygame==1.9.6 -
On Ubuntu:
sudo apt-get -y install python-pygame pip install pygame==1.9.6 -
Others: Please read the instruction here.
pip install git+https://github.com/ntasfi/PyGame-Learning-Environment.git
pip install minatar==1.0.10
pip install git+https://github.com/qlan3/gym-games.git
Run python test.py.
Please use this bibtex to cite this repo:
@misc{gym-games,
author = {Lan, Qingfeng},
title = {Gym Compatible Games for Reinforcement Learning},
year = {2019},
publisher = {GitHub},
journal = {GitHub Repository},
howpublished = {\url{https://github.com/qlan3/gym-games}}
}