Evolving Game NPCs Based on Concurrent Evolutionary Neural Networks

Evolutionary Artificial Neural Networks (EANNs) has been highly effective in Artificial Intelligence (AI) and in training Non-Player-Characters (NPCs) in video games. An important question in training NPCs in games is how we can choose the appropriate way to make NPCs smart. We focus on (1) choosing a principled method of high dimensional data space, (2) designing adaptive fitness functions which can make the proper evolution. In this work, we describe the Concurrent Evolutionary Neural Networks (CENNs) based on EANNs for competitive team game playing behaviors by teams of virtual football game players. We choose Darwin Platform as our test bed to show its efficiency. The Red team and the Blue team are competing in the soccer field, the field players in Red team are evolved during the virtual game playing. The experimental results show that the Blue team programmed by Rule-Based System leads the evolution successful.