A Mental Simulation-Based Decision-Making Architecture Applied to Ground Combat

Abstract : At last year's BRIMS conference, we described a model of mental simulation based on statistical event prediction (Kunde and Darken, 2005). In this paper, we describe a new decision-making architecture based on our mental simulation model. We have developed and tested the model using a scenario built in COMBAT XXI, where the model is used to make fire/hold fire decisions. While the choice of what is to be predicted and the basis for the prediction are chosen by a human modeler, the details of the predictive models are constructed by machine learning based on actual simulation data. Three different predictive models are used to support the decision, one for target richness, one for the effects of obscuring terrain, and one for losses. The outputs of the predictions are integrated by a decision component, which is currently implemented by a decision tree. Preliminary experimental results indicate that the predictive ability of the model and the resulting firing behavior are similar to human performance.