This was one of my first big projects. It was done in Unity game engine with C#. I did not know about machine learning or AI design at the time but I wanted to finally get into it.
My goal was to create an AI that could solve any maze that was thrown at it. I decided to design my algorithm to accomplish this task. Here is the final product:
Before starting this project I had no clue how any AI or machine learning worked. I decided to rely on the mathematics to help me accomplish my goal. First I had to decide what programming language and IDE I would use for this project. I chose to use Unity because I could visually represent the AI accomplishing its task and since Unity uses C# I was stuck with C# as the programming language. I quickly set up a set of mazes just to have a variety for the AI to solve. The green block is the goal that the AI has to reach and the yellow spheres are AI's sensors that have to reach the green block. I programmed in the movement and physical collision so that the yellow balls couldn't go over the red walls of the maze. For the algorithm, I decided to go the multi-generational route. The way this algorithm works is a generation is given a certain amount of time to live. Each generation has a certain number of steps during which each yellow ball will pick and move in a random direction. After each step is complete each ball records the distance from the goal and the starting point. Depending on the data it's assigned a weight (factor of importance) which determines if the data is good or bad. When another generation comes along its random direction is influenced based on the progress of the previous generation's steps. If the progress for a certain step is the same over and over (the ball got probably stuck in a corner or dead-end) then I just assigned a fresh random direction. This cycle keeps repeating until the green block is reached. This algorithm works pretty well however it's not perfect. Since there is a random factor involved an AI could get extremely lucky and solve a maze in a few generations. But, it could also get extremely unlucky and take hundreds of generations to solve a maze (also depends on maze complexity). When back tracking is involved it also relies heavily on luck of random direction which also slows down the algorithm.
Overall this was an important project for me. It's not perfect and needs improvement (in fact I was thinking of redoing it) however it helped me experience working on a large project which takes a lot of time and patience to complete. The knowledge that I gained from this project helped me prepare for my college CS classes where we have to do several large projects throughout a semester. This project also taught me the importance of mathematics in computer science and gave me a basic understanding of how AI works.