More believable and intelligent control of entities and aggregates is needed in military simulations for prediction, evaluation, planning, and training in the area of battlefield operations. Intelligent agents are a technology that can be used to achieve this level of believable behavior in a simulation autonomously by modeling various aspects of the complex goal-oriented decision making of the units involved. While some simpler behaviors at the entity level can currently be simulated quite well (e.g. route-planning for a single tank platoon), the behavior of larger aggregates, such as battalions and brigades, are considerably more complex, requiring information-gathering, situation assessment, resource allocation, coordination with other units, etc. In the University XXI project, we have been developing an intelligent agent architecture to support the training of tactical operations center (TOC) staff officers at the brigade level. We have focused on simulation of the behavior of battalions, to which the brigade commander issues orders and with which many of the personnel in the TOC interact (e.g. to provide information or support). Our simulation of battalions inter-operates with OneSAF Testbed (OTB) to control the friendly units at the battalion level, and to receive responses from the enemy in real-time. In this paper, we describe the underlying agent architecture we have developed for simulating battalion behavior. There are three central components: a generic task representation language for capturing procedural knowledge about when and how to carry out various functions within the battalion TOC, an agent algorithm (interpreter) for carrying out the execution of these tasks in an interleaved way that allows for sufficiently reactive behavior, and an implementation of this architecture in Java and Jess (Java Expert System Shell). We have used our task representation language to do a large amount of knowledge acquisition about staff functions in a battalion TOC by encoding doctrine (e.g. techniques, tactics, and procedures) acquired through reading documentation/manuals and through interviews with military expert. We have implemented a prototype of our system on a Movement-To-Contact scenario. The success of our approach validates the potential utility of intelligent agent technology for these types of military applications, and our system will be further developed as the basis of simulation-based training system for digital forces staff operations. The main contribution of this research has been to provide insight on how agents can be used to model specific functions within a sophisticated aggregate such as a battalion, and to effectively make decisions that synthesize both scenario-specific orders as well as general knowledge about how to adaptively carry out those functions as a situation diverges from expectations.
展开▼