Previous MSc Theses

"General Video Game Playing with Goal Orientation." B. Ross. J. Levine. Department of Computer and Information Sciences, University of Strathclyde. 2014. Download PDF (BibTeX) MSc Advanced Computer Science

Abstract:
Artificial Intelligence of the past has been primarily designed to deal with specific tasks, but within the field of general video game playing intelligent agents must be created that can cope with multiple different unknown scenarios. The AAAI started a competition in 2005 that allowed board games to be the focus of this research, and a new competition has been started in 2014 to help further this which now focuses on video games, and will be concluded at the IEEE CIG conference in Germany. This competition features a library of Java based games where anyone can try and create general video game playing agents and then submit them for evaluation. To date, focus has been given to tree search based approaches such as Monte-Carlo Tree Search, but this method has its drawbacks which does not make it applicable to all types of gameplay. In this report a possible solution is suggested to cope with one of these limitations by looking at a more goal oriented approach in combination with Monte-Carlo Tree Search. By using this newly designed method a comparison is able to be drawn with a variety of other techniques that do not have this goal based focus. The results are promising and show potential not only in the scores achieved during gameplay, but in the overall behaviour of AI controlled agents when playing video games with greater goal orientation.