* About

This was university work, a major part of my diploma thesis. It simulated a multiagent scenario and tested groups of agents with XCS (a variant of learning classifier systems). Their goal was to follow a moving goal object and keep it in surveillance range. The algorithm I developed outperformed the standard implementation of XCS. The results are soon available in a paper that is being prepared and presented at the upcoming AAMAS 2010 conference.
The emphasis in this work was on simulation, data analysis, parameter optimization, Windows GUI, genetic algorithms and heuristics.

* Downloads
- Source Code (JAVA) (use at your own risk. If you want to use any part of the software for other works please contact me first or reference me in the credits)
- Thesis (german, ~180 pages) (the usual copyrights apply)
- Our GECCO 2010 paper (the usual copyrights apply)

* Screenshots

The (huge) configuration window. The settings are on the left while the database output is on the right.

A closeup of the configuration options.

A short example run (500 steps) with heuristics. The white dots are the agents the green dot is the goal object, the red dots are obstacles and the remaining colors denote sight and surveilance range of the agents.