Introduction

If after many hours of trying to teach your little sister to avoid walls, she still can't cope with the concept of obstacle avoidance, then there's not much I can do for you. Besides, you shouldn't be applying the previous tutorial to siblings! Robots at a pinch, but not people ;)

This tutorial builds upon our work tackling basic obstacle avoidance, and anyone that is not familiar with is is recommended to tackle it first. The model discussed therein does not have any major flaws, and the quality of the obstacle avoidance here will be pretty much the same. This articles goes into more details of advanced issues that can get it to perform more reliably, with a simpler solution. This is more of an optimisation phase, and the resulting performance improvement does become obvious.

We will start by disecting the representation used in the previous approach, and give very practical insights into how this can be improved upon. Then we discuss how learning by evolution in real-time environments can be tricky business, and we tackle the major issues with a slant towards motion control, obviously. The third section will look into the famous Genetic Algorithms in much more detail, and describe pratical issues rather than the theoretical background. Then, we will look into things that are possible with this basic framework, and show that this is a great setup for more experimentation

Remember you can visit the Message Store to discuss this tutorial. Comments are always welcome!. There are already replies in the thread, why not join in?

Next: Representation Improvements