I've been working on genetic algorithms and neural networks lately. I want to put together a creature demo similar to black and white so I've been reviewing my AI text for decision trees and reading everything I can on learning algorithms. My experience with AI is very basic: I've coded mini max and alpha beta tic tac toe games, used A* for 2d and 3d path finding, created simple belief networks, and pretty much faked all the rest.

Reading through Mat Buckland's AI Techniques book and playing with his example code has given me a great introduction to neural nets and machine genetics. I've been using what I learned in that book to create a smart lawn mower using a neural net. The application has two stages - the learning and the practical. The learning stage is where most of the time is spent creating a decent network by evolving generations of chromosomes with elitism and/or optional user interaction. The practical stage involves loading a network, created in the learning stage, to cut the grass. The world has boundaries and no tall grass at the start. The user moves a cloud around in order to water the grass and make it grow taller. Once the rain stops the mower is activated and cuts the grass as quickly as it is able before deactivating. It's so much fun to make these little brains. [smile]

Here is a screen of the learning phase graphing mode. It graphs each generation's best and average fitness scores as well as identifying the best of the best.