Realistic agent populations for large-scale virtual training environments

Abstract

This paper explores the problem of effectively simulating realistic populations of background agents in large-scale virtual environments for training and mission rehearsal. It explains why such populations are needed, and surveys the behaviors one would want to see them exhibit. It argues that a successful simulation of a background population ought to be credible, scalable and maintainable, while defining what is meant by those terms. It identifies the key technical challenges as being the efficiency of behavior authoring, and the efficiency of simulation. Finally, it describes some ways such a background population simulation can be verified and validated.