Simulating Agility for Networked Military Systems

This paper describes some extensions to CAVALIER, a tool for studying the performance of networked military organisations. Specifically, we have improved event handling and added neural-network-based learning techniques. To demonstrate the use of this simulation tool, the paper describes in detail an experiment studying networking and agility. Agility refers to the ability of an organisation, person, or military force to perceive an upcoming threat, and to respond quickly enough to it. In the experiment reported here, a networked force is required to conduct a simple abstraction of a military mission: picking up 100 “items” scattered around a micro-world, while at the same time dealing with hostile forces. The experiment illustrates some key principles of networked operation: First, agents require either early awareness of upcoming threats, or the ability to respond to them very rapidly. These two attributes can be traded off against each other. Second, there is little benefit in networking if agents already have enough information, or if they do not have any information worth sharing. Third, high-quality information creates a situation where motion causes risk; but if this breeds risk averseness, overall mission success may suffer. In this paper, we describe two improvements to the agent-based simulation system embedded within our network farming tool [7]. These improvements are the use of event queues for more efficient event management, and neural-network learning to reduce the amount of fine-tuning of parameters that is required to obtain realistic behaviour.