Behavioral Cloning of Student Pilots with Modular Neural Networks

This paper investigates how behavioral cloning can be used to decrease training time for students learning to y on simulators. The challenges presented to each student must be tailored to their unique learning experiences. This requires an intelligent training regime that exploits a model of each student that predicts where the student's performance will be de cient. Here we show that cloning the behavior of student pilots with a modular neural network results in the automatic decomposition of the behavior into sets of skills. This decomposition may provide a means for identifying when certain skills are acquired by students and which skills are decient. This information may then be used to decrease training time by altering the sequence of simulation experiences to just those that the student needs.