Cognitive Context Detection for Adaptive Automation

Abstract

An important research challenge in Human Machine Systems (HMS) is to create machines that are able to better understand human behavior so that the overall efficiency of the HMS can be enhanced through increased productivity and reduced safety risk. The research question posed in this paper is the following: Can an understanding of physiological behavior of humans be combined with statistical machine learning theory to develop predictive models that are able to accurately predict the cognitive difficulty experienced by humans? In this paper, we answer this question in the affirmative by demonstrating the use of two physiological measurements, pupil dilation and eye-gaze patterns, as indices of cognitive workload. Specifically, we demonstrate the possibility of cognitive context detection through machine learning and classification using eye-tracking data from NRL’s Supervisory Control Operations User Testbed (SCOUTTM), a flexible simulation environment that represents the tasks that a future UAS operator would engage in, while controlling multiple UAS.