Supervising is not the same as driving: the influence of the interaction between driving modality and time-on-driving on stationary gaze entropy during long, monotonous drive

Abstract

Background: The implementation of Driver Monitoring Systems (DMS) in automated vehicles is still an open challenge. Such DMS based on eye-tracking technologies seem to be the most promising tool to track drivers’ states, including their fitness-to-drive. Recent studies have proposed gaze entropy as a reliable measure to be used as a DMS index, in relation to driver engagement. Here, we compared the influence of driving modality (i.e., manual [MD] vs. automated driving [AD]) and time-on-driving on stationary gaze entropy over a 3-hour driving session. Methods: Twenty-six experienced drivers underwent a 180-minute simulated, early-morning, driving session. Each driving session consisted of two 90-minute blocks of both MD and AD. After the first 90 minutes, without resting, the driving modality changed from MD to AD or vice-versa (the order was balanced across participants). We recorded stationary gaze entropy using a 4-camera remote eye-tracker (120 Hz). Driving performance and subjective ratings of alertness were also collected. Results: Stationary gaze entropy was influenced by the interaction between driving modality and time-on-driving. Gaze entropy clearly differentiated between driving modalities, with the AD inducing a less stereotyped (i.e., more random) visual exploration. However, this difference was less clear during the last block of driving, when AD and MD induced a more homogenous visual exploration. Post-hoc comparisons indicated that time-on-driving did not affect gaze entropy over the whole 180-minute driving. Performance and perceived alertness confirmed the effectiveness of the overall experimental manipulation. Conclusions: Our results seems to confirm the need to develop DMS based on complementary sources of information. In our study, gaze entropy clearly differentiated between driving modality, while it was not influenced by the time-on-driving (i.e., fatigue). These findings have the potential to improve road safety by providing accurate measurements of driver engagement and might be key to assess future driver-vehicle interactions.