Developing Models of Expert Performance for Support in an Adaptive Marksmanship Trainer

The U.S. Army’s Engagement Skills Trainer (EST) uses sensors on simulated weapons to collect valuable data about a soldier’s performance during marksmanship exercises. That data is available to an instructor for coaching and remediation purposes. However, experience shows that accessing the data, reviewing the data, and providing feedback to a trainee can be a time consuming process. This environment presents challenges when considering the number of trainees who must complete this training and the limited number of instructors available. This also assumes that instructors are capable of accurately interpreting the data and applying effective remediation. Simulators like the EST are prime candidates for the incorporation of an Intelligent Tutoring System’s (ITS) capabilities. The goals of an ITS are to collect data from a system, make inference on that data as it relates to defined metrics, and to provide formative feedback when data is found to deviate from a specified standard. For this purpose, a system requires models to compare data against. In this paper, we will present the results of the first phase of a study to apply ITS technology to the fundamentals of marksmanship. Models created in this phase will be integrated into an adaptive training system prototype built within the Generalized Intelligent Framework for Tutoring (GIFT) for future experimentation. Data was collected across eight experts from the U.S. Army Marksmanship Unit’s service rifle team as they conducted marksmanship tasks. These models are built around sensor data collected during execution, with each sensor being selected based on their link to the fundamentals of marksmanship. We will review the techniques applied to the data for model construction, trends found in the data that are generalized across each expert, and how the models will be used to diagnose error and trigger remediation.