Adversarial intent modeling using embedded simulation and temporal Bayesian knowledge bases

Abstract

To foster shared battlespace awareness among air strategy planners, BAE Systems has developed Commander's Model Integration and Simulation Toolkit (CMIST), an Integrated Development Environment for authoring, integration, validation, and debugging of models relating multiple domains, including political, military, social, economic and information. CMIST provides a unified graphical user interface for such systems of systems modeling, spanning several disparate modeling paradigms. Here, we briefly review the CMIST architecture and then compare modeling results using two approaches to intent modeling. The first uses reactive agents with simplified behavior models that apply rule-based triggers to initiate actions based solely on observations of the external world at the current time in the simulation. The second method models proactive agents running an embedded CMIST simulation representing their projection of how events may unfold in the future in order to take early preventative action. Finally, we discuss a recent extension to CMIST that incorporates Temporal Bayesian Knowledge Bases for more sophisticated models of adversarial intent that are capable of inferring goals and future actions given evidence of current actions at particular times.