The SFighterAI project trained an intelligent AI agent for clearing the bottom boss of Street Fighter II Special Champion Edition based on deep reinforcement learning.The intelligent agent makes decisions based entirely on the game screen (RGB pixel value). In the first round of the last level in the given archive of the project, it can achieve a 100% winning rate (in fact, there is an “overfitting” phenomenon, seerun testpart of the discussion).

file structure

├───data
├───main
│   ├───logs
│   ├───trained_models
│   └───scripts
├───utils
│   └───scripts

Game configuration files are stored in data/ folder; the main code folder for the project is main/. in,logs/ Contains the terminal text and data curve recording the training process (use Tensorboard to view);trained_models/ Contains model weight files at different stages, which can be used in test.py Run the test in , and watch the effect of the playing strategy learned by the intelligent agent at different training stages.

run guide

This project is based on the Python programming language, mainly using OpenAI Gym Retro,Stable-Baselines3 and other standard code libraries.The Python version used by the program is 3.8.10, it is recommended to use Anaconda Configure the Python environment. The following configuration procedures have been tested on Windows 11 systems. The following are console/terminal (Console/Terminal/Shell) commands.

Environment configuration

# 创建 conda 环境，将其命名为 StreetFighterAI，Python 版本 3.8.10
conda create -n StreetFighterAI python=3.8.10
conda activate StreetFighterAI

# 安装 Python 代码库
cd [项目上级文件夹]/street-fighter-ai/main
pip install -r requirements.txt

# 运行程序脚本定位 gym-retro 游戏文件夹位置
cd ..
python .\utils\print_game_lib_folder.py