Here's a pretty good high-level overview of genetic algorithms, and an animation demo that used them to evolve simple virtual actors that could locomote:

The algorithm generated 100 animated characters, each with a randomly assembled neural network controlling its muscles. Then the algorithm let them all try walking. Predictably enough, the first generation was almost completely inept. But a few figures were slightly better than the rest—they took one hesitant step before crumbling to the ground. By the standards of the fitness function, they became the winners of round one. The software made 20 copies of their neural networks, introduced subtle mutations in each of them, added 80 new participants with randomly wired networks, and started the next generation walking.

And it worked. They got little virtual critters that walked, others that crawled, did somersaults, and so on.

This is similar to what Philip and I are working on, though we're trying to evolve networks that play Go.

And as a sidenote, I hope that such algorithms should be demonstrated and taught to students taking high school and freshman biology, as a practical demonstration of the power of bottom-up evolutionary design.