Related Experiment Video
Updated: Sep 17, 2026

Eye Tracking During A Complex Aviation Task For Insights Into Information Processing
Published on: April 4, 2025
From signal to construct evidence: multimodal consumer-wearable sensing in real general-aviation flight
Rongbing Xu1,2, Shi Cao1,2, Michael Barnett-Cowan2,3
1Department of Systems Design Engineering, University of Waterloo, Waterloo, ON, Canada.
Abstract:
We asked how far multimodal consumer-wearable signals recorded during real general-aviation flight can be treated as construct-relevant measurement rather than as records of motion, posture, or task structure. Forty pilots flew five standardized maneuvers in a Cessna 172 while frontal electroencephalography (EEG and AF7/AF8), cardiac activity, electrodermal activity (EDA), eye movements, skin temperature, and flight dynamics were recorded, with segments paired with pilot self-reports and the four in-air maneuvers additionally paired with instructor competency ratings. We assembled three nested tiers of evidence, namely, known-groups sensitivity, within-person convergent evidence, and criterion-related association, and then tested prediction in held-out pilots. Signal quality supported inference, with trial-level availability from 85% to 97%. Self-reported stress varied across maneuvers (H = 61.1), but because maneuver order was largely fixed this variation is confounded with elapsed time, fatigue, physical dynamics, and sensor settling, and therefore cannot establish that any signal is specific to stress. Within-person convergence was strongest for cardiac (mean heart rate r rm =.57) and oculomotor (saccade and fixation rates r rm =.39) channels, with phasic EDA and frontal theta adding weaker support. Criterion-tier associations were re-examined with leave-one-pilot-out and wild cluster bootstrap inference: Kinematic and electrodermal associations held, whereas every frontal theta/alpha association proved to rest on individual pilots and did not survive. In cross-validation with tuning nested inside each training fold, sensor features did not exceed a maneuver-only baseline across pilots. Low-cost cardiac and oculomotor channels are the most promising basis for within-person stress monitoring, while pilot-general prediction will require individual calibration, stronger reference measures, and larger samples.