Bos22apl
Bos22apl is a project that tries to integrate agent technology and reinforcement learning into the open source real-time strategy game Bos Wars.
It is originally developed for my master’s thesis project in which I compared different approaches in learning winning strategies. However, it may be used as a more generic research platform for intelligence and agent technology in computer games. It will be used as starting point during the Games and Agents course at the Intelligent Systems group of the Utrecht University, organised by dr. Frank Dignum.
Wednesday 7 may 2008, I gave a talk about the project at ISIS. The presentation can be downloaded.
Interested in using this AI research platform? The full instruction manual can be downloaded. It describes the design, how to obtain the code and how to use it. The reinforcement learning extension for the C++ version of 2apl can be used separately with any 2apl agent and environment.
In June 2008 I graduated on this project. My master thesis work has been rewarded with a mark of 8,5 out of 10. The full thesis can be downloaded as a PDF. I am pretty proud of the work done an hope that future students will continue work on learning agents in games.
A little video showing an agent that has learned a winning strategy against a static opponent:
[...] Bos22apl « Bos22apl presentation [...]
Hello!I am following your posts for many weeks now. I have to admit that it is very interesting . It is already added in my favourite list and i will make sure that i will follow it when possible . Thanks for the interesting inputs . Furthermore, i really like your template and the way you have organised your site . Could you the name of your template ? Thanks