Written by Ethan Smith

Table of Contents

The repo:

https://github.com/ethansmith2000/MazeSolver

Intro

After seeing some posts by Francois Fleuret about solving mazes with neural networks, I got inspired to try my hand at it. I don’t know much on the methods they used, so I came up with a few different things to try, and I figured I would test them against each other in performance.

For all methods, the mazes themselves are represented as a 2D array where

• Walkable paths = 0
• Walls = 1
• Start = 2
• End = 3

Then depending on the method, the maze will either be normalized to -1 and 1 or be encoded as discrete tokens

Here’s how it looks, matplotlib style.