3.1. The Study Setting
In general, two study methods are applicable for motion sickness studies concerning autonomous driving: field experiments with real vehicles and studies in a driving simulator.
In field experiments, the participant is passenger of a real vehicle in a realistic road environment or on a test track. Naturally, the experimenter has no full control of the dynamic events happening and the experiences the participants make in a field study, resulting in reductions of internal validity. However, there are different ways to control for this. First of all, the driving style of the used vehicle should be standardized. In future, this can be realized by using automated functions that perform driving manoeuvers in the same way with high reliability. Until these automated vehicles are commonly available for this kind of study, human drivers need to drive the test vehicles. High levels of realism for future automated vehicles can be achieved by Wizard-Of-Oz settings, in which the automation is simulated by a human driver, e.g., [42
]. It is necessary that human drivers are instructed or trained towards a specified and thus reproducible driving style [45
]. In the presented setting (cf. Chapter 1), we used assistance systems like adaptive cruise control (ACC) and lane keeping for standardization. The trained drivers learned how to perform lane change maneuvers with the necessary step of actions (for example setting indicator before moving steering wheel, changing lanes in six to seven seconds). Along with that, there should be a low number of experimental drivers in order to avoid inter-individual differences in driving style between experimenters. Finally, after completing data collection, it is recommended to analyze the dynamic driving data to identify any conspicuousness within the actual realized driving behavior. If possible, the data can be systematically compared to the vehicle dynamics measured (1) within the same study to check for internal validity and (2) in other settings or situations in order to check for external validity.
However, even if the vehicle dynamics are kept as standardized as possible, external factors like traffic or weather conditions cannot be kept constant or manipulated consciously. However, these aspects can affect motion sickness: A high traffic density can lead to an increased number of braking and overtaking maneuvers due to slower vehicles. This driving behavior can lead to stronger symptoms of motion sickness. In contrast, a low traffic density enables homogenous driving with less accelerations and decelerations, which reduces the probability for motion sickness.
In contrast to field studies, driving simulators enable conducting studies in a highly controlled environment. They are used since the 1960s to investigate driving performance and behavior and are classified into three categories [46
High-level simulators incorporate a motion system and full vehicle cabs;
Mid-level simulators are static simulators with a full vehicle cab;
Low-level simulators are built around simple components such as game controllers and computer monitors.
As most researchers attribute motion sickness in vehicles to contradictory impressions between the vestibular system (which perceives motion) and the visual system (which perceives no motion, e.g., while reading a book), the use of a high-level simulator with a motion system is recommended. In mid-level and low-level simulators, in contrast, only visual induced motion sickness can be investigated. Basically, research questions concerning countermeasures or physiological correlates are conceivable in these simulators. However, it remains unclear how the results of these studies in simulators without motion system would be applicable for automated driving.
The most common motion platform of high-level simulators is a hexapod which provides motion in six degrees of freedom (x, y, z, roll, pitch, yaw). Compared to travelling in a real vehicle, longitudinal and lateral accelerations are different. The feeling for realistic accelerations is generated by hacks like tilting the presented scenery. More elaborated simulators mount the hexapod on an x-y table on which the simulation cabin is moved to produce more realistic accelerations. According to Carsten and Jamson [47
], however, even a large motion system is not capable to provide realistic accelerations in special driving situations like negotiating a long curve.
Probably the most important benefit of driving simulation is the ability to create repeatable scenarios which are tailored to a certain research question. Depending on the research question, motion-sickness provoking scenarios with many strong lateral and longitudinal accelerations are possible as well as more homogenous driving scenarios with few accelerations only (e.g., highway scenarios). Additionally, the researcher is free in the selection of the driving behavior of the autonomous vehicle: each imaginable driving style is feasible even if this driving behavior is not possible in a real autonomous vehicle yet. Another benefit of driving simulation is the availability of data: the simulator provides all data that would be provided by a real test vehicle (e.g., velocity, acceleration) as well as data of the traffic environment (e.g., surrounding traffic, road geometry). Besides, the participant’s behavior (e.g., head movement, glance behavior) and physiological data can be monitored and recorded in a simple way: the laboratory conditions make video recordings easier due to constant light conditions and physiological data recording more precise due to less disturbing artifacts of the environment (e.g., temperature, humidity).
On the other hand, there are disadvantages of driving simulation. Some participants of simulator studies suffer simulator sickness, which is a subtype of motion sickness in simulated environments. The phenomenon occurs in all types of simulators—it also appears in fixed-base simulators without motion system due to visual stimuli only. Similar to motion sickness, it is caused by a mismatch between the visual perception and the vestibular sensation of acceleration and deceleration [14
]. For a motion sickness study, this means that the results for motion sickness can be confounded with simulator sickness. For studies regarding prevalence or development of motion sickness it is recommended to exclude participants who have shown symptoms of simulator sickness in previous studies in order to diminish this artifact. However, as simulator sickness and motion sickness are related and show similar symptoms due to similar reasons, it is possible that some countermeasures are effective against both symptoms. Therefore, it has to be discussed if participants with simulator sickness problems are allowed in a study concerning motion sickness countermeasures. However, this issue has to be decided for each countermeasure or research question separately.
An important issue of driving simulation is the validity. A distinction that has been made on simulator validity is between absolute and relative validity [49
]. Relative validity exists when effects in the simulator and under the same road conditions are in the same order and direction. In contrast, absolute validity is present when the numerical values are about equal in both systems. A lot of validation studies were carried out in various simulators. They compared various parameters of the driver’s behavior (e.g., velocity, lateral displacement, braking behavior, gaze direction) between driving in a simulator and driving in a real vehicle. In most cases, the studies showed that relative validity exists while absolute validity was only rarely verified [50
]. However, these results do not provide evidence that validity is given for motion sickness studies. In a motion sickness study, behavior of a driver is not relevant—moreover, the occupants’ visual and vestibular perceptions are important.
To the authors’ knowledge, there have not yet been studies comparing an occupant’s motion sickness in a driving simulator to his/her motion sickness in a real vehicle. Therefore, we conducted the study design as described above.
The results showed that the progress of motion sickness was comparable in both conditions. After a general rise at the beginning of the run (approx. first 12 min), the sickness ratings increased more slowly in the second and last third. Compared to real driving, self-reported motion sickness was slightly higher in the simulation compared to the real vehicle (Figure 1
). However, the maximum sickness values during the runs do not differ (Wilcoxon signed-rank test: Z
= 1.40, p
= 0.162). The sessions of n
= 3 drivers had to be aborted due to high sickness ratings in the simulator. In the field study, the run of n
= 1 driver was terminated before the end of the test course. According to the symptom questionnaire, most symptoms occurred in a similar frequency and intensity in both runs (Figure 2
left). However, three symptoms differed significantly concerning their intensity: in the driving simulator, participants had higher general discomfort, more difficulties concerning focusing, and increased appetite (Figure 2
right). In a final interview after both runs, the participants stated that the motion sickness symptoms were more distinct in the driving simulator compared to the real vehicle (t
(23) = 5.65, p
These results indicate that relative validity is given for the high-level simulator of the WIVW GmbH concerning motion sickness as the progression during the runs was comparable and the occurrence of frequent symptoms was similar. In contrast, absolute validity cannot be verified, as some of the self-reported symptoms were more distinct in the simulator.
The recommendation for the most appropriate study setting depends on the research questions: A field experiment offers the highest validity and should be used for studies which investigate the prevalence and the development of motion sickness. In this case, a conduction on public roads should be selected. The realistic test track could represent a highway, rural road or inner-city track. Previous studies used driving on highways and inner city roads to identify if and how strong motion sickness occurs. In these studies, the participants performed different tasks in the vehicle [51
In contrast, in case the research question covers the investigation of countermeasures avoiding or reducing the symptoms of motion sickness, it is crucial to choose a test setting that causes motion sickness in the participants quickly and with a high probability. In the vehicle context, this setting was mainly realized on test tracks, on which high provoking maneuvers were driven by the experimenters (e.g., driving in the shape of an eight, or constant stop and go). Other researchers made use of placing the participant rearwards in a vehicle driving on urban roads [51
]. Within this setting, a comparison between a baseline trial and a repetition of the same condition with potential countermeasures allows to investigate the effectiveness in avoiding symptoms. In particular, considering the efforts put into these kinds of participant studies, an efficient and reliable creation of provoking situations needs to be considered in the study design. Besides, a simulator study using a motion sickness provoking scenario can also be conducted when investigating countermeasures. A requirement for this option is the validity of the driving simulator. The presented study shows that a high level driving simulator without x-y table can also offer relative validity—however, as driving simulators are very different this has to be tested for each simulator individually.
3.2. The Participant’s Task
In general, automated driving will enable the driver to engage in various non-driving related activities. In motion sickness research, one relevant research question refers to specifically examining the different non-driving related tasks (NDRTs) for their potential to cause motion sickness. In respective investigations, subjects could either be free to engage in realistic everyday NDRTs of their choice or be presented with a specific NDRT. While many standardized tasks exist in the context of manual driving, such standardization is widely missing in the context of automated driving. Therefore, it would be desirable to also evaluate secondary tasks that cover certain groups of conceivable NDRTs in the future. For our setting, we chose a naturalistic NDRT. Based on previous research [5
], it can be expected that the use case watching a video in an automated vehicle has some external validity.
Concerning other research questions such as the evaluation of countermeasures, it may also be relevant to induce motion sickness in a targeted manner or to investigate an extreme scenario. In this case, NDRTs that are characterized by highly limited peripheral and external vision of motion are required as hints about the vehicle’s future motions can counteract motion sickness [53
]. Therefore, a mainly visual NDRT should be presented in a way that assures gazing away from the road scene. To ensure standardization of the amount of peripheral vision across participants, visual material should be presented at a fixed location, e.g., by means of displays instead of providing handheld devices such as tablets. Naturally, fixed display positions also lead to more standardized participant movements. Since peripheral vision can be manipulated by both display position and size [7
], to prevent the participant from using peripheral vision, a visual NDRT could be presented at a downward angle or on a large display. Further, to promote continuous task engagement, it is recommended to choose an NDRT that is difficult to interrupt or provides instructions and incentives for subjects to focus on the task and refrain from road glances (e.g., concentrating on visual tasks like reading or watching a movie during the drive increases the risk of motion sickness). Artificial, standardized NDRTs can therefore be suitable for this. Please note that engaging in a visual NDRT may cause visual problems such as strained eyes or blurred vision, which cannot be differentiated from symptoms of motion sickness. To better control for this, visual task characteristics and the duration of the task engagement may be considered. Similarly, fatigue may occur due to the experimental session’s duration or as a motion sickness symptom. Further research should examine both the relationship between motion sickness and fatigue as well as methods to control for confounding effects.
For the selection of an adequate task for empirical studies on motion sickness, classifications of NDRTs, provide relevant dimensions such as the primary modality, the locality, the possibility of road glances, the need for sustained attention, and incentives to continue the task [56
]. In addition, the presented material should be controlled for emotionality of content when motion sickness is measured using physiological correlates. Therefore, in our study, subjects watched a movie on a display positioned below the central information display. We further instructed subjects to refrain from road glances. The videos contained documentaries, which were interesting but not emotionally arousing. Other examples for such tasks may be reading a text or answering a quiz that is presented visually. Finally, participant posture should be considered in motion sickness studies given that the risk of motion sickness is also higher when the passenger is sitting on a rearward facing seat compared to a forward facing seat [52
]. Moreover, for postures facing in the driving direction, a regular driving posture may increase the risk of motion sickness compared to a reclined posture [57
3.3. Sample and Recruitment
In order to investigate motion sickness in autonomous vehicles a participant study is recommended. The requirements for the recruitment depend on the study’s research question.
For a large variety of research questions, it is necessary that a significant part of the sample suffers from motion sickness during the study. For example, the effect of a countermeasure for motion sickness during travelling can only be demonstrated when a control condition in a between- or within-subjects design exists in which motion sickness occurs. In contrast, people who are not susceptible to motion sickness do not need countermeasures and are not relevant for the study question. It is only possible to identify physiological correlates of motion sickness when the participants have phases with and without motion sickness. Therefore, the selection of participants is crucial for the study’s success as not all people are susceptible to motion sickness. This consideration leads to the next question regarding participants’ recruitment: how to identify participants who are susceptible to motion sickness?
A common instrument to predict motion sickness susceptibility is the MSSQ (Motion Sickness Susceptibility Questionnaire) [14
]. This tool queries how often several means of transport (e.g., cars, busses, airplanes) and amusement rides (e.g., carousels, rollercoasters) were used in the past and how often sickness occurred. The answers result in a motion sickness susceptibility score. However, the results of our own study indicate that the MSSQ total score is not appropriate to identify subjects who are susceptible to motion sickness while travelling in a car. There was no significant correlation (Spearman r
(24) = 0.266; p
= 0.210) between the MSSQ total score and the suffered motion sickness (measured via a misery scale according to [9
]) in a real driving study on the Autobahn in which the N
= 24 participants were passengers and had to watch a video during the drive (see Figure 3
left). The MSSQ probably covers too many means of transport—respondents with no motion sickness problems in cars can also achieve high MSSQ scores when having symptoms, for instance, in trains and airplanes. In contrast, respondents who compensate for their motion sickness in real driving situations might reach lower MSSQ scores than would be intended: people who know that they are susceptible to motion sickness might not engage in NDRTs in provoking situations and therefore did not experience any severe motion sickness in the past years.
However, the more specific MSSQ item “Over the last 10 years, how often you felt sick or nauseated in cars?” also showed no significant correlation (Spearman r
(24) = 0.212; p
= 0.319) to the suffered motion sickness in the study (see Figure 3
right). The question is very inaccurate as it does not differ between driving in an urban or rural area or on a highway. In addition, it summarizes travelling in a car while reading or texting on the back seat as well as being a co-driver who is attentive to the traffic situation. As the prevalence depends on the individual threshold to motion stimulation and varies under different situations [59
], a curvy rural road can lead to symptoms for some people while other people suffer from motion sickness in urban scenarios only. Therefore, it is recommended to use a highly specific question with the exact test scenario as a screening question for the participants’ recruitment (e.g., “Do you get symptoms of motion sickness as a co-driver while reading on the Autobahn?”).
Concerning other research questions, a more common sample is required. A representative sample is necessary to investigate the prevalence of motion sickness. The sample should be representative concerning all aspects which can affect the prevalence of motion sickness, e.g., age [60
] and gender [27
3.5. Data Analyses in General
Due to ethical reasons (see Chapter 4), participants must be able to terminate participation at any stage of the study. Furthermore, the experimenter has to terminate the session in cases of conspicuous suffering of the participant. Therefore, a researcher has to expect dropouts during the conduction of a motion sickness study. In driving studies concerning other topics (e.g., acceptance of a new driver assistance system) these dropout participants are often replaced by other participants so that each condition consists of a sufficient and equal number of data, which facilitates the statistical analysis. In a motion sickness study, however, the occurrence of a dropout is very important as it indicates that motion sickness was too distinct.
Concerning post-study questionnaires (e.g., MSQ), dropouts are not a problem for data analysis as all participants—regardless of cancelling or completing the session—can fill it out. However, all data collected during the runs are sensitive to dropouts during the session. On the one hand, this influences the statistical data analysis and might necessitate the usage of tests which can handle dropouts and missing data. On the other hand, however, researchers can use dropout rates as dependent variables, investigating which conditions caused how many people to abort the trials due to sickness. Furthermore, dropouts enable time-based parameters describing the progress of motion sickness: How long does it take until the dropouts occur? Does this time differ between the test conditions? Therefore, researchers should not see dropouts as a problem (like in other research issues), but rather as an increase of information.
In general, time-based parameters describing the progress of motion sickness are important for motion sickness studies: if a continuous online assessment of motion sickness is conducted (e.g., via FMS, MISC, or symptom-specific Likert scales), it is possible to use parameters which define the time until a participant reaches a specific symptom (e.g., “time to nausea” or “time to sweating”). These data are helpful for the description of motion sickness and the effect of countermeasures.