My project is to create a neural network and investigate its capabilities to solve
a 2D maze with very limited information. The maze is built in Python and can be rendered
out using PyGame and the Neural network is built with Tensorflow and uses Q learning a
reinforcement learning algorithm to teach itself how to solve the maze. The maze is
randomly generated and the amount of walls in the maze varies based on it’s difficulty level.
I created an experiment to test the Neural networks capabilities by getting it to train
continuously on randomly generated mazes and increasing the difficulty over time until the
maze is too difficult for it to complete. Once the experiment has finished we generate metrics
and graphs and are able to come to a number of interested conclusions.
Contact e-mail: emmettbyrne@hotmail.com