Um framework para pesquisa de inteligência artificial em jogos de estratégia por turnos

Abstract

In digital games, the Artificial Intelligence (AI) is one of the most important elements to ensure both challenge and fun, especially in genres such as strategy games. However, due to the high complexity of games of this genre, the AI performance is often very weak when compared to fairly experienced players. With this scenario, the need for more robust AI techniques and algorithms bring very interesting challenges to researches in the area. However, a major difficulty faced by researchers is the lack of proper tools for conducting experiments in real games, either because the maintenance of the code is too complicated or because real commercial games do not offer the access for modifying everything needed. In this context, we propose the creation of a framework appropriate for the research on AI algorithms in Turn-Based Strategy (TBS) games by creating a game with its architecture specially designed for AI research on TBS games, in order to have a proper tool for experiments for future works. The main goal of the work is to provide a comprehensive tool for the development and experimentation of AI in TBS games. The main requirements which must be with are to (i) allow the performance of contests between different AI agents without the need for a human player, (ii) enable the development of AI algorithms with no need of writing much code not related to the algorithm itself, (iii) have a well modularized architecture so the AI is totally in independent modules from the rest of the game and is not allowed to cheat and (iv) contain a TBS game which includes the main elements present on other games of the genre, so that the experiments on it are applicable in real games.Upon completion of the development phase, we conducted evaluation experiments with participants from different experience levels, both relating to AI area and games development in general. According to the results, we found that the tool built on this work achieved the proposed objectives and has a great potential for use in experiments of new algorithms and techniques applied to TBS games.