This is a project I've done to study neat-python package. Also used: pygame.
It is a racing game: there is a track and a car, controlled by player (or AI). The task is to drive the car on the track. If car is on the grass than game is over. Player (or AI) can select velocity and angular velocity of car from fixed ranges. There are several radars on the car, which measure distance from the center of the car to the border of the road. NEAT studies by taking the results of the measures from the car and passing gas and turn intesivity to it. The greater car velocity or the less car angular velocity the greater fitness of the genome of the NEAT.
First, run following command to install pygame dependencies:
sudo apt build-dep python-pygame -y
Alternatively:
sudo apt-get install python-dev libsdl-image1.2-dev libsdl-mixer1.2-dev libsdl-ttf2.0-dev libsdl1.2-dev libsmpeg-dev python-numpy subversion libportmidi-dev ffmpeg libswscale-dev libavformat-dev libavcodec-dev
Second, clone this repo.
Third, go to the folder of repo and install virtual enviroment:
pipenv install
To play the game run
pipenv run python play.py
To train AI run
pipenv run python trainai.py
To run trained AI run
pipenv run python playai.py [path]
path to AI is optional. By default it runs last trained ai. Also it is possible to pass several paths.