Can Artificial Intelligence Improve Training of Uncrewed Aerial Systems Operators?

Abstract

Human operator error remains a persistent safety concern in Uncrewed Aerial Systems (UAS) operations, with elevated cognitive load (CL) identified as a key contributing factor. This highlights the importance of reliable, real-time Artificial Intelligence (AI)-assisted, particularly Machine Learning (ML)-based, CL assessment during operator training, which may support future adaptive training approaches to improve trainee performance and operational safety. This study examines ML models for reliable real-time operator CL classification using multimodal physiological signals under non-random validation strategies designed to reflect real-world training conditions, where models must generalize to unseen data. Among the models examined, the Convolutional Neural Network (CNN) achieved the highest classification accuracy (75.2%), demonstrating the feasibility of real-time CL classification when temporal and individual dependencies are preserved. These findings provide a methodological basis for future ML-assisted adaptive UAS operator training, where continuous monitoring of trainee cognitive state may inform training adjustments to improve training effectiveness and operational safety.