Pilot Skill Level and Workload Prediction for Sliding-Scale Autonomy

Abstract

An emerging topic in human-computer interaction research involves optimal collaboration between humans and machines to achieve a particular goal. One approach to such a goal involves sliding-scale autonomy, in which a machine dynamically adjusts between different levels of autonomy based on a variety of measurements. In this paper, we propose a system to predict skill level and workload for aircraft pilots using machine learning algorithms. Our proposed system uses the pilot's heart rate variability and flight control data, including pilot inputs such as throttle and aileron, and flight sensor data such as latitude and longitude. We conduct a user study on 15 pilots, each flying the same 5 pre-defined routes on a flight simulator. Our results indicate that the flight control data alone are sufficient to provide a near-perfect classification of a pilot's skill level into expert or novice. On the other hand, predicting mental workload is much more difficult, and a combination of flight control and heart rate data is required to obtain an accurate estimate of mental workload. Our findings provide the first step towards a sliding-scale autonomous system for aviation.