|Thesis abstract: |
Modern computer games achieved an impressive level of realism featuring advanced 3D graphics and sophisticated physics engine able to simulate complex game environments. However, the increasing complexity of the game environment poses many challenges for developing the Artificial Intelligence which controls the non-player characters. In many games, the non-player characters show very simple and predictable behaviors which are not able to evolve over time or to adapt to the user¿s gameplay.
Another problem, caused from the complexity of the game environment, is the generation of a large quantity of game content, e.g., levels, maps, racing tracks, etc. In most complex computers games, the content generation can require the work of several designers and can last even for months. Thus, there is the necessity of techniques that can automate or support some aspects of the game development process.
In this context, Computational Intelligence can be a valid approach to deal with some of the issues in computer games. Computational Intelligence is a field of the Artificial Intelligence which includes techniques and mathematical models which are inspired by nature, such as Evolutionary
Algorithms and Neural Networks. In the recent years, the number of scientific works involving Computational Intelligence and Game is quickly growing and suggests that this is a promising research direction.
In this thesis, we focus on the application of Computational
Intelligence, and in particular Evolutionary Algorithms, to computer games.
In the first part, we investigate different learning paradigms which can support the development of non-player characters. In particular, we investigate on-line learning, i.e., a learning process applied during the game to generate sophisticated non-player characters that can evolve
over time. In the second part, we focus on game content generation. We investigate several representations and different approaches for evaluating the game content. In our experimental analysis we take into account different computer games: racing games, shooters and platform games.
The results show that all the proposed methods are promising and Computational Intelligence represents an effective approach to support the game development process.