Note: including code for both training and testing!

AI Solutions for Fighting Game Based on Traditional Reinforcement Learning Algorithms

Two traditional reinforcement algorithms Q-learning and Monte Carlo

Introduction of Fighting Game AI Competition：http://www.ice.ci.ritsumei.ac.jp/~ftgaic/index.htm

[Paper] [DemoVideo] [Project Page]

1. Environment Setup
2. Code structure

This code has been tested on Window 10, Python 3.7

• Clone the repository

git clone https://github.com/TongRu/FightingAI.git

• Install OpenJDK (java environment)

  1. Download the installation package for window from https://jdk.java.net/14/

  2. Unzip it to a fixed path, such as D:\jdk-14.0.1

  3. Add java program to system path: Right-click to open My Computer → Properties → Advanced System Settings → open the environment variables → edit and create a new path "D:\jdk-14.0.1\bin" in the user variable -Path

  4. whether java is successfully installed: open the cmd command window and enter “java -version” to display java information.

  if you apply the code in in Linux, please directly execute the following instructions to install OpenJDK

  $ sudo apt-get install openjdk-8-jre
  

• Setup python environment

conda create -n FightingAI python
source activate FightingAI
pip install gym
pip install py4j
pip install port_for
pip install opencv-python

.\fightingice_env.py: starting code of the Fighting Game Platform;

.\data\ai\MctsAi.jar: Opponent bot based on java env;

.\train-MC.py: code for training traditional Q-learning algorithm;

.\train-QLearning.py: code for training traditional MC algorithm;

.\train-ImprovedQL.py: code for training improved Q-learning algorithm;

.\test-MC.py: code for testing traditional Q-learning algorithm;

.\test-QLearning.py: code for testing traditional MC algorithm;

.\test-ImprovedQL.py: code for testing improved Q-learning algorithm.