Considerations in the Design of Realistic Agents for Serious Games

Abstract

Artificially intelligent software agents are a staple of the video game and simulation industry, and are used for many purposes ranging from entertainment to training to analysis and decision support. Using data from live training exercises, this paper considers the process of constructing an effective data pipeline to determine the critical features that should be represented in intelligent agents for use in game-based training environments. Intelligent agents are often developed to represent the cognitive processes and behaviors of individuals or aggregate units or teams. The goal of the research described herein is to determine if the construction, training and validation of autonomous agents can be enhanced by modeling their perception, decision, and action processes after real entity behaviors captured during live training exercises.