Genetic2DCarGame

An AI is taught to beat a 2D car game with Genetic Algorithm for a Neural Network.

Background

In this project a genetic algorithm generates 100 cars and 100 neural networks (NN). In each generation, the fittest species breed, trying to create better offspring. A little mutation is involved too.

• The NN inputs are the state of the cars, that is the distance of the walls around the car in certain angles. It can be imagined as a lidar.
• The output of the NN is the directions to move towards (left-forward, forward, right-forward).

Fun facts

• As the NNs learn to move based on the distances of the walls around the car, it won't be track dependent.
• Learning to navigate on a track, maybe enough, to navigate on newly generated tracks.

Built With

• Python
• pyglet - The cross-platform windowing and multimedia library for Python
• keras - The Python Deep Learning library

Authors

• Kardos Tamás - Initial work - Swordy