Jörg Denzinger's
Research
 
      

AI methods for the development and set-up of computer games

For many people, especially non-experts, the obvious way to use Artificial Intelligence techniques around computer games is in the so-called AI component(s) of a game during run time to offer realistic challenges to the game player(s). Unfortunately, this usually comes with efficiency requirements that many techniques can not fulfill or that require very specialized techniques and compromises (a good example is path planning). Consequently, many AI components are using scripts and other non-AI techniques. Some classes of AI techniques, for example machine learning techniques, additionally come with a lot of potential risks that most game developers simply do not want to be exposed to when considering them as part of the run-time behavior of a game.

Due to these concerns, we have seen AI techniques applied a lot to the set-up of games, which is known as automated content generation for games. While there are still efficiency requirements around such a use of the techniques, they are less severe than the near real-time requirements mentioned above and there is also the possibility to do some risk analysis around the outcome of machine learning attempts, although the use of such techniques in this context still comes with a lot of problems. We have contributed some works to this area and a description of our ideas can be found here.

But recently the idea of using AI techniques, especially learning techniques, in the development phase of games also has found some traction, since there are no dangers with regard to unforeseen game behaviors, but quite some potential for reducing development costs and improving product quality. We have looked into using learning techniques for testing computer games (among other things; our results are published at CEC 2004, CIG 2005, CIG 2009 and AIIDE 2011, see our bibliography page) and we also looked into the automated creation of the behavior of non-player characters.

For more information follow the links above or take a look at our papers cited on our bibliography page. A list of the persons that are or were involved in this research can be found here.
to our ideas on automated content generation.

Last Change: 26/5/2013