Gaze Entropy Metrics for Mental Workload Estimation are Heterogenous During Hands-Off Level 2 Automation

Goodridge, Courtney Michael; Gonçalves, Rafael C.; Arabian, Ali; Horrobin, Anthony; Solernou, Albert; Lee, Yee Thung; Lee, Yee Mun; Madigan, Ruth; Merat, Natasha · 2023 · Crossref

DOI: 10.31234/osf.io/g2z5w

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of estimating driver mental workload during hands-off Level 2 automated driving, a critical requirement for developing reliable Driver Monitoring Systems (DMS). While previous research has utilized gaze-based metrics to detect workload, it has predominantly focused on mean differences, ignoring the inherent variability within and between individuals. The authors argue that for DMS to be valid and reliable across diverse populations, it is essential to model this heterogeneity. Specifically, the paper investigates two Information Theoretic gaze metrics: Stationary Gaze Entropy ($H_s$), which measures the predictability of fixation locations, and Gaze Transition Entropy ($H_t$), which measures the predictability of the sequence of fixations. The study employed a 2x2 repeated-measures design with 38 participants (mean age 38.81) in a motion-based driving simulator. Participants completed two 35-minute drives on a simulated UK motorway. In one drive, they performed a verbal 2-back cognitive task during hands-off automated driving to induce high mental workload; in the other, no secondary task was present. Both drives included critical takeover events with varying time-to-collision (3s and 5s). Gaze data were recorded at 60 Hz. The analysis utilized a Bayesian distributional multilevel modeling approach, which allowed the researchers to model both the population mean ($\mu$) and the standard deviation ($\sigma$) of the gaze entropy metrics as functions of workload and individual characteristics, rather than assuming homogeneity of variance. The results indicated that Stationary Gaze Entropy was a reliable indicator of mental workload, with 92% of the predicted population showing a decrease in $H_s$ when performing the 2-back task. In contrast, Gaze Transition Entropy exhibited substantial heterogeneity; only 66% of the population was predicted to show similar decreases in $H_t$ under high workload. Furthermore, age emerged as a strong predictor of this heterogeneity, influencing the average causal effect of high mental workload on eye movements. These findings suggest that while $H_s$ consistently reflects workload changes, $H_t$ varies significantly among individuals. The significance of this work lies in its shift from point-estimate analyses to distributional modeling, highlighting that individual variability is not merely noise but a theoretically relevant feature of gaze control. The findings imply that future DMS designs must account for inter-individual differences, particularly regarding age, to ensure that workload detection metrics are both valid (accurate) and reliable (consistent) for a wide range of drivers. Ignoring this heterogeneity could compromise the safety benefits of automated vehicle functions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• workload measurement
• mental demand
• gaze based attention detection
• cognitive
• temporal
• situational awareness

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: behavioral performance data, physiological data
• Theoretical Contribution: theory or model