Conference Agenda

The assessment of Mental Workload (MWL) is a key re-

quirement in Industry 5.0 environments, where operators engage in in-

creasingly complex cognitive tasks. This paper presents a real-time, fully

causal eye-tracking-based framework for continuous MWL feature gener-

ation, relying exclusively on gaze and pupillometry signals from a wear-

able device. The framework integrates subject-specific baseline normal-

isation, online artefact handling, and sliding-window aggregation into a

modular pipeline under strict causality constraints, without offline post-

processing or additional sensors. Eleven features capturing pupil dynam-

ics, baseline-relative variations, temporal derivatives, and spatial gaze de-

scriptors are generated at 1 Hz. Evaluated on N = 13 quasi-independent

recordings through a modified n-back paradigm with five progressively

demanding sessions, the pipeline achieved a median feature acceptance

rate of 99.1% (IQR 99.1-99.2%) across 65 sessions. A Friedman test on

N = 12 subjects (independent observations) revealed a significant ses-

sion effect (χ2(4) = 14.53, p = 0.006, Kendall’s W = 0.30), reflecting

sensitivity to progressive cognitive demand. The proposed framework

provides a scalable and non-invasive approach for MWL feature gener-

ation, combining literature-grounded causal integration with wearable-

only deployability, supporting integration into adaptive human-machine

systems within Industry 5.0 contexts.