APTIMA INC — Department of Defense SBIR Phase I: Rapid development of expert models would enable the expansion of computer-based training t

Description

Rapid development of expert models would enable the expansion of computer-based training to new training domains, but current methods for defining expert models are labor intensive, require large amounts of expert time, and often require the experts to understand aspects of the underlying technology. To reduce the amount of expert interaction required to develop a model, Aptima proposes EMULATE: Expert Models Using (Machine) Learning to Accelerate Training system Engineering (EMULATE). EMULATE"s Bayesian Inverse Reinforcement Learning (IRL) algorithms will observe experts and learn to mimic their behavior by working backward from observations of expert actions which are often fragmentary or incomplete to a representation of expert goals, preferences, and beliefs. EMULATE will feature expert model learning and data collection software modules that will integrate with simulation games and other types of intelligent tutoring system learning environments to provide feedback to students about how their behavior compares to that of experts. EMULATE will build on Aptima"s Performance Measurement Engine to collect expert data from multiple training environments and demonstrate its ability to rapidly learn expert models in three complex training domains, including multi-UAS route planning.