On-Line Coevolution for Action Games
Pedro Demasi, Adriano Joaquim de Oliveira Cruz · IJIGS. International journal of intelligent games & simulation · 2003
Coevolutionary algorithms (CEAs) have been widely explored in recent years. Cooperative and competitive methods were proposed and evaluated, and many theoretical studies have been made about them and important results have been achieved, however few works have been published about a real-time approach to CEAs, with online agent evolution. The goal of this work is to explore this field of application of CEAs, proposing some methods and strategies for online evolution in an action (real-time) game. In this game, a human player interacts with computer-controlled agents, which begin with very naive or random behaviour and gradually get “smarter”, resulting in improved difficulty levels of gameplay. We present four different methods to do online evolution of the agents: using game specific information; merging offline-evolved data with online evolution; using online data only; and using them together. We will, finally, present some results and a brief discussion of the advantages and disadvantages of each one of the methods proposed, based upon these results.