Multimodal characterization of male drivers' affect-relevant and workload-related responses in extra-long tunnel...

Extra-long tunnels can impose sustained perceptual, physiological, and regulatory demands on drivers, but affect-relevant and workload-related responses are difficult to assess continuously in naturalistic driving because subjective ratings are retrospective and cannot capture fine-grained spatial variation.

This study developed a multimodal, label-free framework to characterize latent affect-relevant and workload-related response patterns during extra-long tunnel driving. A naturalistic driving experiment was conducted in a 4.9 km tunnel, during which physiological responses, driving behavior, and gaze measures were collected and organized as a third-order tensor (driver × indicator × segment). CANDECOMP/PARAFAC tensor decomposition was then applied to extract the dominant latent structure without subjective labels.

The selected rank-1 solution showed an association with route familiarity. Heart rate and frontal gaze measures contributed most strongly to the latent structure, with additional contributions from vehicle speed and electrodermal activity, suggesting a coupling among autonomic activation, visual attention regulation, and behavioral adaptation. Spatially, elevated factor loadings were observed near tunnel entrances, exits, and several interior segments, indicating that the response patterns observed in male drivers are not limited to transition zones but may also emerge during prolonged tunnel exposure.

The study provides exploratory multimodal evidence that may support the interpretation of affect-relevant and workload-related response structures and the identification of spatially differentiated response patterns that may inform further tunnel safety investigation.