Evolving reactive NPCs for the real-time simulation game

AI in computer games has been highlighted in recent, but manual works for designing the AI cost a great deal. An evolutionary algorithm has developed strategies without using features that are based on the developer. Since the real-time reactive selection of behaviors for NPCs is required for better playing, a reactive behavior system consisting neural networks is presented. Using only the raw information on games, the evolutionary algorithm optimizes the reactive behavior system based on a co-evolutionary method. For demonstration of the proposed method, we have developed a real-time simulation game called 'Build & Build'. As the results, we have obtained emergent and interesting behaviors that are adaptive to the environment, and confirmed the applicability of evolutionary approach to designing NPCs' behaviors without relying on human expertise.