This project was done to test the capabilities of machine learning. At any given position in the maze the AI will decide which way to progress through the maze. Using tensorflow JS I created a neural network with three hidden layers summing over 100 nodes to predict the probability that a choosen path will lead the AI to solving the maze. Each decision the AI makes and the result of that decision is stored and used to train the neural network towards reaching the goal. That being said, with so many nodes being optimized it is likely that the network will take a long time to find a consistent solution model.
It is crucial to note that this network is not optimizing the neural network towards finding the shortest path, but rather finding a path to the goal without revisiting tiles. This increases the difficulty since backtracking is not a valid strategy for this goal. I choose to optimize for this goal since there isn't a clear solution to this problem.
The check boxes below give you control over certain aspects of the maze. That being said, the neural networks will optimize faster the more moves can be made per second so allowing faster transitions is recommended. Lastly, the neural networks are stored between sessions so refreshing or closing the tab is not a problem.