With the gradual improvement of vehicle automation, researchers have found that drivers free from driving tasks are more prone to under mental workload conditions11. De Winter et al.2 summarised 32 empirical studies on drivers’ mental workload. They established that the mental workload of drivers under automated driving is reduced by 20.8% compared to manual driving. In both automatic and manual driving mode, the complex driving scenario has a cumulative effect on mental workload3. The driver needs to monitor the road continuously and the complex driving conditions still need high task requirements4. Therefore, this study quantifies the co-variation relationship between task demand and mental workload by creating traffic scenarios with different complexity levels.
Young et al.5 further improved the relationship between task demand, mental workload, and operational performance based on previous studies. They divided three regions by the two red lines. The region to the left of the red line has low task requirements and arousal levels, and the operator has the remaining cognitive resources, called the reserve capacity region. The region on the right having higher task requirements is called the overload region. The region between the two red lines represents the optimal load region. These three regions are of great significance for driver workload monitoring and accident prediction.
In summary, we have selected two factors: driving mode (automated driving and manual driving) and driving scenario (monotonic condition, engaging condition), and put forward the following hypothesis:
Driver’s mental workload in automated driving under monotonic conditions is in the reserve capacity region.
Mental workload is closely related to driving fatigue. Hancock and Desmond6 distinguished task-related fatigue into active and passive fatigue. Active fatigue is caused by tasks that require continuous coordination of perceptual activity. The fatigue caused by tasks requiring few perceptual activities and long-term monotonous reactions is called passive fatigue. May and Baldwin7 further perfected this view and proposed that the key to dividing both fatigues depends on the mental workload. The fatigue induced by the underload was passive, and the fatigue caused by the overload was active. Previous research has unclear standards for the effectiveness of passive and active fatigue, and researchers generally classify them by creating driving scenarios with different complexity levels8; however the complexity of the condition is not the only criterion for defining active and passive fatigue. Oron-Gilad et al9 believed that even monotonic conditions can induce active fatigue in novice and sleep-deprived drivers. Therefore, a simple road scenario cannot be used to determine the type of fatigue induced in this study. Owing to the increasing trend of alpha wave power in monotonic and complex scenarios10,11,12, it is difficult to judge the type of fatigue using EEG. According to May and Baldwin7, the key to distinguishing between active and passive fatigue is the mental workload. We can prove that passive fatigue induction was successful only when the experimental participants had a lower mental workload with accumulated fatigue.
This study measures the driver’s mental workload and degree of fatigue simultaneously and tests the effectiveness of passive fatigue induction by comparing these two variables. Therefore, we proposed the following hypothesis:
Compared with other driving conditions, the mental workload of drivers in automated driving under monotonous conditions should be the lowest, but the degree of fatigue is significantly increased, indicating that they are in a passive fatigue state.
Various types of fatigue have various causes and intervention methods. Under automated driving, a mental workload that is too low can quickly induce passive fatigue in the driver, making it more difficult for them to take over the vehicle’s control13. Körber et al.8 found that drivers experienced fatigue in about 42 min based on the eye movement index, indicating that it took a specific time for drivers to develop a low mental workload from fatigue. Vogelpohl et al.14 proved this by judging the time of fatigue in the condition of automatic driving by facial expression in 15–35 min. In this study, we hoped to collect the driver’s EEG indicators and detection-response task (DRT) performance based on previous studies to clarify the precise time of passive fatigue. Consequently, we stated the following:
Drivers’ mental workload under reserve capacity region of approximately 40 min will induce the driver to experience passive fatigue.
During automation, the speed of drivers with passive fatigue became slower with regard to responding to takeover requests14,15,16, and more frequent involvement in non-driving-related tasks occurs17,18. These indicators can sensitively reflect fatigue in practice; however, the driver’s state is not sufficient to measure fatigue if it is to be monitored accurately and in real time. Researchers favour EEG indicators because of their accuracy and real-time performance19. Lal et al.20 used the average EEG activity of waking state participants as the benchmark. They analysed the characteristics of changes in the participants’ EEG in different fatigue stages and concluded that when the driver was fatigued, the delta and theta activity increased. In a simulated driving scenario, Jagannath and Balasubramanian11 found that as the test participants’ fatigue increased, alpha increased significantly, and theta decreased significantly; however, to accurately define a driver’s passive fatigue state, a clearer definition standard is required. Thus, we compared and analysed the time-domain characteristics of EEG signals in different states and selected the sample entropy that characterises the signal’s complexity as a passive indicator. Further, we used the receiver operating characteristic curve (ROC) analysis method to determine the driver’s passive fatigue discrimination threshold, based on the entropy of the EEG samples. Thus, we proposed the following hypothesis.
Based on the ROC curve method, the critical threshold of the EEG samples’ entropy of the driver's passive fatigue and waking state can be calculated.