Toward Real-time Assessment of Workload: A Bayesian Inference Approach

Abstract

Workload management is of critical concern in teleoperation of unmanned vehicles, because high workload can lead to sub-optimal task performance and can harm human operators’ long-term well-being. In the present study, we conducted a human-in-the-loop experiment, where the human operator teleoperated a simulated High Mobility Multipurpose Wheeled Vehicle (HMMWV) and performed a secondary visual search task. We measured participants’ gaze trajectory and pupil size, based on which their workload level was estimated. We proposed and tested a Bayesian inference (BI) model for assessing workload in real time. Results show that the BI model can achieve an encouraging 0.69 F1 score, 0.70 precision, and 0.69 recall.