The arousal paradox in critical task performance in automated driving during sleep inertia using a quasi experimental approach

Abstract

Sleep inertia is the post-awakening transitional state of lowered arousal, characterized by increased low-frequency activity in the electroencephalogram (EEG) and impaired cognition. While some theories consider arousal holistically, recent research questions whether these findings apply to situations requiring immediate critical action post-awakening, such as for pilots, emergency responders, or future drivers of automated vehicles. This study compared self-reported, cortical, and physiological arousal in such a scenario. Twenty-four participants completed four drives in a driving simulator. In three drives, participants were instructed to sleep for 20, 40, and 60 min during automated driving before being prompted to resume control. The sleep stage prior to the takeover request served as a quasi-experimental independent variable. Regression analyses showed that cortical arousal was low following awakenings from N2 or N3, indicated by increased delta, theta, and alpha activity. However, beta activity and heart rate also increased, suggesting elevated physiological arousal. Significant positive correlations were found between delta activity, heart rate and self-reported sleepiness. This "arousal paradox" is not in line with the idea of arousal as a holistic concept. We hypothesize that the heightened physiological response under sleep inertia may be attributed to stress in demanding situations under sleep inertia. We conclude that forced awakenings from N2 or N3 should be avoided. If someone is nevertheless awakened from N2 or N3, they should be given sufficient time between awakening and taking over duties for arousal to normalize.

Keywords: Arousal; Electroencephalography (EEG); Sleep; Sleep inertia; Takeover.

Fig. 1

Cockpit of the high-fidelity driving simulator and visual takeover request which appeared in the cluster display as indicated by the white arrow (photo provided by WIVW, used with permission).

Fig. 2

Schematic procedure of the experimental sessions. One of the three sessions included an additional block with automated and manual drive whose data serves as a baseline.

Fig. 3

Mean distribution of sleep stages across all participants.

Fig. 4

Mean heart rate before, during and after the takeover depending on the last sleep stage. TOR: Takeover request; bpm: beats per minute; Significance of regression model: *p <.05; **p <.01; ***p <.001; n.s. = not significant. a) Segment 0 shorter than 30 s.

Fig. 5

Mean occipital EEG delta activity before, during and after the takeover depending on the last sleep stage. PSD: Power spectral density; TOR: Takeover request; Significance of regression model: *p <.05; **p <.01; ***p <.001; n.s. = not significant. a) Segment 0 shorter than 30 s.

Fig. 6

Mean occipital EEG theta activity before, during and after the takeover depending on the last sleep stage. PSD: Power spectral density; TOR: Takeover request; Significance of regression model: *p <.05; **p <.01; ***p <.001; n.s. = not significant. a) Segment 0 shorter than 30 s.

Fig. 7

Mean occipital EEG alpha activity before, during and after the takeover depending on the last sleep stage. PSD: Power spectral density; TOR: Takeover request; Significance of regression model: *p <.05; **p <.01; ***p <.001; n.s. = not significant. a) Segment 0 shorter than 30 s.

Fig. 8

Mean occipital EEG beta activity before, during and after the takeover depending on the last sleep stage. PSD: Power spectral density; TOR: Takeover request; Significance of regression model: *p <.05; **p <.01; ***p <.001; n.s. = not significant. a) Segment 0 shorter than 30 s.