Skip to content

Latest commit

 

History

History
33 lines (29 loc) · 1.16 KB

README.md

File metadata and controls

33 lines (29 loc) · 1.16 KB
Raw

gym-sor

Gym Streets of Rage is an environment bundle for OpenAI Gym built on top of the Retro Learning Environment. Users can train one or two agents on any single level of Streets of Rage 1-3. Users

Installation

First install the Retro Learning Environment:

git clone https://github.com/jmichaux/Retro-Learning-Environment.git
cd Retro-Learning-Environment
pip install .

Next, install gym_streets_of_rage:

git clone https://github.com/jmichaux/gym-streets-of-rage.git
cd gym-streets-of-rage
pip install .

Choosing an Environment

To choose an environment for training a single agent, use the following pattern: game-character-level-difficulty

env = gym.make('streets_of_rage_ii-max-1-4-v0')

Rewards

By default, the environment returns a dictionary of rewards corresponding to: health, lives, points, kills

TODO

  • set rewards
  • re-map the action space
  • Fix two player

Acknowledgements

  • @nadavbh12 for his work on RLE.
  • @gsaurus for his work hacking and reverse engineering the Streets of Rage series.