Optimal Training Systems STTR

Abstract : Successful training in complex environments is normally accomplished through the interaction of a trainee and a skilled expert, but experts are an expensive commodity. Using an optimal model of task performance subject to human constraints may be a more efficient way to develop models of skilled human performance for use in training, especially since optimal models are simpler to validate, test, and debug than corresponding expert models. In addition, constrained optimal models can be constructed in domains where no experts are available or even exist. Using a simulated task environment (STE) permits the necessary close model-trainee interaction by enabling the construction of optimal performance models that perform the same task as the trainee using the same interface while closely observing and guiding trainee performance. We have developed a methodology for using a normatively correct task model as the core engine of an automated tutor for a national missile defense (NMD) task STE. This methodology has allowed us to explore: the relative impact of expert versus optimal feedback, the locus of learning within the NMD task, the differential impact of providing feedback on strategy selection, and methodologies for constructing tutors directly from expert performance data.