Our primary activities in this research involve interviewing military aviators and encoding their knowledge into a large rule-based system. From one point of view, this is an exercise in constructing an expert system. However, the particular domain we are tackling (high-fidelity air combat simulation) distinguishes itself from traditional expert-systems problems in that it requires the flexible generation of very complex behavior in a rapidly changing real-time environment. The domain also requires the software system to communicate and coordinate with humans and other simulators, and to interpret sensor information and generate behavior in a manner similar to humans. Thus, we find it preferable to categorize this research in the areas of situated, autonomous, intelligent agents and cognitive models of high-level behavior.
This research was originally funded by a grant from the Defense Advanced Research Projects Agency (DARPA). That culminated in an expert system that provided behaviors for all of the fixed-wing aircraft missions in the Synthetic Theater Of War Advanced Concept Technology Demonstration in October 1997. This demonstration included flying over 700 missions during 48 straight hours of an operational military training exercise, with minimal human intervention.
Soar Technology, Inc. is building from this early work to field a variety of highly autonomous intelligent agents for various applications and research. The air combat expert system continues to add capabilities and is currently used in a number of training systems, experiments, and technology demonstrations.