Gaze entropy metrics for mental workload estimation are heterogenous during hands-off level 2 automation

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Psychology

Background:

• Increasing vehicle automation leads drivers to non-driving tasks, necessitating reliable Driver Monitoring Systems (DMS).
• Estimating driver mental workload is crucial for safety but remains challenging.
• Existing research often overlooks individual variability in workload metrics.

Purpose of the Study:

• To investigate the utility of Information Theoretical gaze metrics for quantifying mental workload in automated driving.
• To analyze within- and between-participant variability in gaze metrics under different workload conditions.
• To explore the influence of factors like age on workload-related gaze patterns.

Main Methods:

• Employed a Bayesian distributional modeling approach to analyze gaze data from 38 drivers in hands-off Level 2 automated driving.
• Compared gaze metrics (Stationary Gaze Entropy, Gaze Transition Entropy) during a secondary cognitive task (2-back) versus a no-task condition.
• Assessed individual differences and the impact of age on gaze metric responses to mental workload.

Main Results:

• Stationary Gaze Entropy effectively indicated mental workload, with 92% of participants predicted to show a decrease during the 2-back task.
• Gaze Transition Entropy exhibited significant heterogeneity, with only 66% predicted to show similar decreases.
• Age was identified as a significant predictor of individual differences in gaze responses to workload.

Conclusions:

• Information Theoretic gaze metrics, particularly Stationary Gaze Entropy, show promise for DMS in estimating mental workload.
• Addressing the heterogeneity in gaze responses is critical for developing valid and reliable DMS.
• Future DMS design must account for individual variability to ensure driver safety in automated vehicles.