Keywords

1 Introduction

Human factors play an increasingly important role in the evolution of automobile development. The growing integration of vehicle automation, from assistance systems to autonomous driving, reshapes the traditional driver's role. Drivers are transitioning into more passive roles, resembling passengers, as vehicles assume greater control. Consequently, the conventional perception of the driving experience undergoes a major transformation [1].

To enhance the acceptance of novel driving functionalities, assessing the subjective driver experience is crucial. Recognizing that self-reports may be subject to distortion or inaccuracy, it is useful to employ objective measures based on physiological parameters. Some studies have previously identified correlations between (dis-)comfort or trust and physiological responses in automated driving functions. Common physiological measures for assessing the driver’s mood and emotions are electroencephalography (EEG), electrocardiography (ECG), photoplethysmography (PPG), heart rate (HR), heart rate variability (HRV), electrodermal activity (EDA) and eye tracking [2].

Given the widespread popularity of smartwatches, they offer a convenient option for capturing physiological data in real-life scenarios. One notable advantage is the ability to acquire data in a naturalistic setting. This paper explores the adequacy of smartwatch accuracy in monitoring the driver's state while experiencing an automated driving function.

2 Related Work

This paper focuses on the physiological parameters of electrodermal activity and photoplethysmography. EDA describes changes in the bioelectrical properties of the skin. The eccrine sweat glands in the skin are controlled by the sympathetic branch of the autonomic nervous system and respond to psychological stimuli. PPG sensors use an optical technique to detect changes in blood volume within the microvascular bed of tissue. Cardiovascular activity, specifically heart rate, can also be considered an arousal indicator. But unlike EDA, heart rate is regulated by both the sympathetic and parasympathetic nervous systems, making it less clear as an indicator of emotional arousal [2, 3].

Previous studies have examined the relationship between physiological data and subjective ratings. Dillen et al. [4] observed a correlation between self-reported comfort and skin conductance in their study. They demonstrated that the type of driving event and the interaction with other road users influence all physiological responses. They asserted that electrodermal activity could predict comfort and anxiety. In a study by Beggiato et al. [5], the correlation between self-reported discomfort and physiological parameters was investigated using a wearable device for measuring. Unlike other studies, no correlation was found between skin conductance level (SCL) and discomfort. However, HR was identified as a significant parameter. The study reported that HR consistently decreased during discomfort periods and returned to prior levels afterward. It was concluded that specific uncomfortable situations affect physiological parameters like HR, whereas situations with moderate to low reported discomfort did not show any changes.

This study employs the Empatica EmbracePlus, a medical wearable device designed to monitor and analyse a range of physiological parameters, including SCL and PPG. The application of the device has been demonstrated in other studies before. Menghini et al. [6] identified significant discrepancies in skin conductance measurements. Several participants exhibited flat and nonresponsive SC before the recording began. They suggested that the E4, a comparable model to the Empatica EmbracePlus, could be reliably used for measuring average HR in healthy adults. This implication also extends to HRV measures, but only under static and stationary conditions, indicating that the quality of the measurements is highly dependent on motion artefacts. This aligns with the findings of Milstein et al. [7].

The device was also used in studies relating to driver monitoring. Gruden et al. [8] utilised an Empatica E4 wristband and investigated the device's accuracy during manual driving. They identified substantial standard errors and elevated deviations in SCL and HRV measurements resulting from hand movements while steering. In a related study, Stephenson et al. [9] investigated the impact of unexpected events during autonomous driving. They found that elevated electrodermal activity persisted after such events, although they did not identify any statistically significant differences in heart rate.

3 Study Description

3.1 Participants

Before the study begins, participants are required to complete a preliminary questionnaire designed to provide a more detailed characterization of the test subject pool. There was a total of 10 participants recruited for this study (33% females) who were between 20 and 35 years old (M = 28.6, SD = 3.2). Six participants have a professional connection to automated driving. Most of the participants use a car at least three times a week or more.

3.2 Experiment Equipment

The driving simulation system used for the study was the Dynamic Vehicle Road Simulator from the Institute of Automotive Engineering, as shown in Fig. 1 (left). This setup features a modified Volkswagen Golf 7 vehicle, cut off behind the front seats. The cabin is mounted on a HEXaDRIVE motion platform from Simtec Systems GmbH (Braunschweig, Germany). Five 48-inch curved monitors provide a 180° visualisation of driving scenarios. The software utilised for the simulation is IPG CarMaker. The mockup is fully functional, including feedback from an electric power steering unit.

The Empatica EmbracePlus (Empatica Srl, Milan, Italy) smartwatch was selected for the study. This device is equipped with a ventral electrodermal activity sensor, which samples at 4 Hz, and a 4-channel multi-wavelength photoplethysmography sensor with a sampling frequency of 64 Hz. The internal Empatica software calculates the systolic peaks. The watch wristband was placed on the participants’ non-dominant wrist and fastened as tightly as was comfortable for them. The wearable also collects acceleration data via a high-precision 3D microelectromechanical accelerometer, which monitors wrist movements. The sampling frequency of the accelerometer is 64 Hz.

3.3 Procedure

The entire procedure takes approximately one hour. The first step involves a detailed briefing of the test subjects, during which they are informed about the study's objective and instructed on how to use the driving simulator. Following this, a reference measurement for the physiological parameter is recorded. A familiarisation drive is conducted, lasting approximately ten minutes, to acclimate the test subjects to the driving simulator.

Fig. 1.
figure 1

Dynamic Vehicle Road Simulator (DVRS) of the Institute of Automotive Engineering (left) and scenario (1) with a crossing cyclist (right)

Full size image

The actual test drive consists of four separate drives, each covering a 4-km route through an urban environment. These drives are conducted automatically. Each drive incorporates a potentially uncomfortable scenario: (1) a cyclist crosses the road from a parking space, (2) a vehicle pulls out of a parking spot, (3) a skateboarder crosses the road, and (4) a vehicle exits from a ramp. These events are strategically positioned at various points along the route to introduce an element of surprise. Participants are not informed about these specific situations during the initial briefing of the study.

In addition to measuring physiological data, we also assess the participants’ subjective perceptions. We utilise an adapted questionnaire from Morra et al. [10] to identify their reactions to the test events. Therefore, we employ a five-point scale. Furthermore, we assess the participants’ subjective perceptions of comfort and safety for the overall drive using a five-point scale also.

4 Results

The results of the subjective ratings indicate that all participants perceived the four situations of the test drive as dangerous (M = 4.44, SD = 0.71) and surprising (M = 4.27, SD = 0.65). Conversely, the ride was generally perceived as comfortable and safe (M = 4.55, SD = 0.49).

Although the subjective ratings suggest a physiological reaction, the results indicate that there are problems in the correct measurement of the data from the Empatica EmbracePlus. Nine participants showed a flat and non-responsive SCL measurement. Only one participant showed reactions to the events, as shown in Fig. 2. However, the reactions were not significant either. In summary, the Empatica EmbracePlus was unable to provide reliable measurements. Given that the device has previously demonstrated that the quality of the data is highly sensitive to wrist movement, we also analysed the data from the accelerometer. As the entire drive was automatic, there was no movement at the wrist and it did not differ between scenarios.

A possible reason for the results could be that the participants had to adjust the wristband themselves, so it may not have been worn tightly enough. In some cases, the wearable was placed on dry skin. However, we also placed a few drops of warm water with the other participants and this did not make a significant difference to the results [11].

Fig. 2.
figure 2

Skin conductance level over all participants for scenario (1) (left) and the reaction to crossing cyclist from the only participant with a reaction (right)

Full size image

Furthermore, SCL measurements are typically obtained from the distal or intermediate phalanges of the ring and index fingers, which are areas with a greater density of active eccrine sweat glands [11]. The EmbracePlus, like many wearables, employs wrist sensors to measure skin conductance. Given that the wrist is less responsive to skin conductance, an underestimation of parameters could be anticipated [12].

The results for heart rate also present challenges. For the calculation, we utilized the internal Empatica software, which identifies the systolic peaks in the PPG signal. The number of peaks detected in a 10-s sliding window was multiplied by six to obtain the HR in beats per minute (BPM). We applied z-standardisation to the data to account for the individual variability of physiological responses. We examined two periods: 30 s before the stop and 30 s after the stop, as this was chosen in other research, as in [9]. The exemplary results are shown in Table 1.

Table 1. Results of the z-standardized HR before and after scenario (1) and (3)
Full size table

It is not possible to detect a significant difference between the two periods. The HR over the entire trial is subject to considerable fluctuations. This is reflected in the high values for the standard deviation.

5 Conclusion

The results presented in this study investigate the physiological parameters measured by the Empatica EmbracePlus smartwatch. Despite the participants reporting that the situations during the study were perceived as dangerous, the SCL showed nearly no reaction. The same applies to HR, which demonstrated considerable variability. It is important to note that the sample size of 10 test subjects is relatively small, and that a larger pool of test subjects could provide a more reliable database. For future research, it would be beneficial to conduct a study utilising additional EDA and ECG sensors in order to rule out the possibility that the participants have no physiological reaction to the events.