A Visual-Vestibular Model to Predict Motion Sickness for Linear and Angular Motion

Abstract

Objective

This study proposed a model to predict passenger motion sickness under the presence of a visual-vestibular conflict and assessed its performance with respect to previously recorded experimental data.

Background

While several models have been shown useful to predict motion sickness under repetitive motion, improvements are still desired in terms of predicting motion sickness in realistic driving conditions. There remains a need for a model that considers angular and linear visual-vestibular motion inputs in three dimensions to improve prediction of passenger motion sickness.

Method

The model combined the subjective vertical conflict theory and human motion perception models. The proposed model integrates visual and vestibular sensed 6 DoF motion signals in a novel architecture.

Results

Model prediction results were compared to motion sickness data obtained from studies conducted in motion simulators as well as on-road vehicle testing, yielding trends that are congruent with observed results in both cases.

Conclusion

The model demonstrated the ability to predict trends in motion sickness response for conditions in which a passenger performs a task on a handheld device versus facing forward looking ahead under realistic driving conditions. However, further analysis across a larger population is necessary to better assess the model’s performance.

Application

The proposed model can be used as a tool to predict motion sickness under different levels of visual-vestibular conflict. This can be leveraged to design interventions capable of mitigating passenger motion sickness. Further, this model can provide insights that aid in the development of passenger experiences inside autonomous vehicles.

Keywords

multisensory integration simulation perception-action autonomous driving usability/acceptance measurement and research motion sickness

INTRODUCTION AND MOTIVATION

Motion sickness (MS) has been extensively studied to assess passenger comfort in ground, aerial, and maritime vehicles. MS is a condition that affects one in every three adults in the United States (MedlinePlus Genetics, 2019). Further, two in every three people globally have reported experiencing MS at some point in their lives (Schmidt et al., 2020). Hence, MS is a widespread problem that impacts the daily lives of people who use transportation systems. While limited in-vehicle data has been collected on this matter, the available studies have suggested that passengers are more susceptible to MS than drivers (Fukuda, 1976; Iskander et al., 2019; Kuiper et al., 2020; Rolnick & Lubow, 1991). With the development of autonomous vehicles (AVs), MS has become a pressing issue given that all occupants will be passengers.

Relieved from the task of driving, AV passengers will be able to engage in productive nondriving-related activities during their commute. However, it has been shown that passenger activities, such as reading or typing on a handheld device, further increase MS (Jones, Le, et al., 2019). The impending adoption of AVs has led to greater emphasis on understanding the mechanisms that lead to MS and identifying strategies to improve passenger comfort and productivity.

The characterization of MS in the literature has been primarily done through data-driven experimental studies using motion platforms/simulators and test vehicles to a lesser extent (Kuiper et al., 2019; Matsangas et al., 2014; McCauley et al., 1976; O’Hanlon & McCauley, 1974). Motion simulator studies, however, are often not representative of realistic driving conditions that lead to MS since they do not fully replicate car maneuvers that a passenger would experience (Bos et al., 2021). For instance, it is difficult for motion simulators to replicate the lateral or centripetal acceleration experienced during a car turn as that would require a large footprint to allow the motion platform seat to reach the required lateral acceleration magnitude. In comparison, MS experimental studies conducted in passenger vehicles driven in realistic driving conditions have several operational challenges including safety considerations, test-track access, environmental and route factors, and cost. However, in-vehicle studies are often more representative of passengers’ experience in naturalistic conditions.

As part of a holistic and systematic approach to parameterizing MS, researchers have also developed models to simulate the human response to certain motion and visual inputs. These models are based on theories that attempt to explain the physiological mechanisms that lead to MS (Bos & Bles, 1998; Kamiji et al., 2007; Khalid et al., 2011; Oman, 1982; Reason, 1978; Salter et al., 2020; Telban & Cardullo, 2001). MS prediction models have the potential to be integrated with passenger and vehicle dynamics models to determine how a given route, driving profile, and vehicle model can lead to passenger discomfort, making it an effective tool. MS prediction models also have the potential to efficiently identify and evaluate strategies to mitigate MS.

One of the key deficiencies of MS models in the literature is the inability to accurately predict MS at frequencies below 0.1 Hz. Models tend to underestimate the level of MS experiences at this frequency range. One possible explanation for this behavior is that few models consider the contributions of the visual system to MS, focusing primarily on the vestibular system (Salter et al., 2020; Telban & Cardullo, 2001). It has been shown that the visual contributions to MS play a significant role at low frequencies (<0.2 Hz) (Diels & Howarth, 2013; Duh et al., 2016). While a few visual-vestibular MS models have been proposed in the literature (Braccesi & Cianetti, 2011; Wada et al., 2020), there is no established framework within these models for handling both linear specific force and angular velocity visual-vestibular stimuli in three dimensions. These types of inputs are widely present in MS triggering events, such as in the case of a passenger looking at screen inside a moving vehicle. It has been shown increased obstruction of a passenger’s view of the outside world leads to increased MS (Kuiper et al., 2018). Hence, taking into account the contributions of the visual field of view of a passenger has the potential of improving MS modeling prediction. This poses a limitation on the existing models regarding the extent to which MS can be accurately predicted for realistic driving conditions. To address this, we propose an alternative MS model capable of handling three-dimensional linear and angular visual-vestibular motion stimuli. This accounts for motion that is representative of what occurs in a moving vehicle and further improves MS prediction over the whole frequency range (Telban & Cardullo, 2001; Zacharias, 1977). The proposed model is able to achieve this though a novel framework that draws both from a subjective vertical conflict (SVC) MS model and a visual-vestibular human motion perception model. While the human motion perception model provides a way to estimate the visual and vestibular stimuli as a unisensory perceived signal, the SVC model provides a way to quantify the accumulation of this perceived signal as a MS score. In this paper, this model is described and the model results are compared with experimental MS responses to realistic driving and task performance conditions gathered by Jones et al. (Jones, Le, et al., 2019).

LITERATURE REVIEW

According to the sensory conflict theory, MS arises when the brain receives conflicting inputs from the visual, vestibular, and proprioceptive systems (Bos & Bles, 2002; Claremont, 1931; Hill, 1936; Reason & Brand, 1975). In particular, the two most prevalent types of sensory conflicts are the canal-otolith conflict (Bos & Bles, 2002) and visual-vestibular conflict (Reason, 1978). Given observations that subjects without a functioning labyrinthine system do not suffer from MS, the canal-otolith conflict is considered to be a causative factor of MS, while the other integrated sensory conflicts (such as visual-vestibular) are considered to be modulating factors of MS. The sensory conflict theory has been shown to have large congruence with experimental results. Further, the canal-otolith conflict specifically has been used to create MS models in the literature due to its practical implementation in a mathematical framework. However, there is no consensus as to how the visual and proprioceptive systems can be integrated within the framework of this theory’s mechanism.

A subset of the sensory conflict theory, the SVC theory states that MS arises due to a difference between the subjective vertical predicted based on previous experience and the vertical as perceived by the sensory systems (Bles et al., 1998; Bos & Bles, 1998; Khalid et al., 2011; Oman, 1989). The subjective vertical is the result of an internal model, which behaves as a learning mechanism within the central nervous system. The internal model attempts to duplicate the dynamics of the sensory system to minimize the discrepancy between the subjective vertical and the sensed vertical. The estimation of the subjective vertical is a result of a neural filtering process that attempts to differentiate inertial and gravitational accelerations, which creates inherit lag in the system (Bles et al., 1998). This causes the subjective vertical to deviate from the sensed vertical, ultimately resulting in MS. The SVC theory is can be used for MS model development as the sensed and subjective vertical can be quantified in terms of its direction and magnitude, allowing for a correlation between the conflict and a quantitative MS score prediction (Bos & Bles, 1998; Kamiji et al., 2007). While this theory can be expanded to include any sensory system that is able to perceive motion, the vestibular and visual systems are often considered the most relevant (Bos & Bles, 2002).

While other MS theories are prominent in the field, such as the postural instability theory (Fukuda, 1976), the theories described above are the most relevant for the purposes of the model presented in this paper. MS theories have led to the development of various models that attempt to predict MS given certain visual, vestibular, and proprioceptive inputs. To provide a quantitative output of MS, these models rely on numeric scales that have been previously developed in experimental studies. Some scales standardize MS reporting based on the presence of fixed, explicit MS symptoms. Other scales use self-reported symptoms as part of quantifying MS. Hence, each of these scales cannot fully capture the MS experienced by subjects. Consequently, the assessment of MS models will have different limitations depending on the scale being used.

For example, the commonly used Motion Sickness Incidence Index (MSI) scale has been defined as the percentage of subjects that vomit under a sinusoidal motion with certain magnitude and frequency (ISO 2631-1, 1997). The applicability of this scale to studying MS in AVs is potentially limited as passengers are expected to show a wide range of symptoms prior to vomiting. Another frequently used scale for MS is the MSQ scale, which is based on a questionnaire that assesses symptoms such as headache, fatigue, nausea, stomach awareness, blurred vision, cold sweating and vertigo (KELLOGG et al., 1965). Another commonly used scale is the MSSQ scale, which uses a questionnaire that can account for the subject’s susceptibility to MS (REASON, 1968). Both the MSQ and MSSQ scales are limited by the symptoms they encompass as well as the accuracy of self-reported symptoms. Other scales such as FMS and MISC (Reuten et al., 2021) are prone to similar limitations. (Jones et al., 2019) suggested a new rating scale that captures the spectrum of MS symptoms and physiological indicators during realistic driving conditions, making it pertinent to studies that involve MS prediction applied to AVs. In this scale, participants were asked to rate their MS from ‘0’ (no MS at all) to ‘10’ (need to stop the vehicle). They were also instructed to describe in their own words and rate any MS sensations they experience throughout the testing protocol. The descriptions provided were processed and standardized to the MS score. This allows for subjects to express their MS beyond a predetermined list of symptoms while taking into account their own subjective rating and perception of symptoms. This allows for the assessment of a wide array of symptoms and intensities, which can be advantageous under realistic driving conditions, where multiple symptoms might be present. The model proposed in this paper is compared to the results in the scale proposed by Jones et al.

Key physiological MS models presented in the literature are summarized in Table 1; the table includes the model inputs and outputs, their limitations, the nature of visual information and the number of degrees of freedom (DoF). The DoFs are defined in this table as the number of independent visual-vestibular signals along a given direction (i.e., vertical visual angular velocity and vertical vestibular angular velocity are counted as 1 DoF). It is important to note that most models’ performance have been compared with experimental results often obtained from studies conducted on motion platforms/simulators that do not represent realistic vehicle motion.

TABLE 1:

Summary of Motion Sickness Theoretical Models in Literature

Number	Authors	Year	Type of Model	Inputs (I)/Outputs (O)	Assumptions/Limitations	Degrees of Freedom (DoF)	Visual Information
1	O’Hanlon & McCauley et al. (Mathematical HFR Model) (O’Hanlon & McCauley, 1974)	1974	Vestibular	I: Sensed specific force of vertical motion O: MSI	Limited to specific force only, and vestibular influence	1 DoF	None
2	Oman et al. (Sensory Conflict Theory) (Oman, 1982, 1989)	1982, 1989	Visual-vestibular	I: Sensed motion by visual and vestibular sensors O: Subjective discomfort	Heuristic model for sensory dynamics, only outputs subjective discomfort. First model to propose internal observer	Not defined	Not defined
3	Bos et al. (Subjective Vertical Conflict Theory) (Bos & Bles, 1998)	1998	Vestibular	I: Sensed specific force of vertical motion O: MSI	Only vestibular model, no influence of visual sensory system considered	1 DoF	None
4	Atsumi et al. (Atsumi et al., 2002)	2002	Others (vehicle motion)	I: Vertical specific force, roll and pitch of vehicle O: Motion sickness level	Output is not in a standard motion sickness scale, correlation model with no physiological modeling	3 DoF	None
5	Kamiji et al. (Kamiji et al., 2007)	2007	Vestibular	I: Sensed motion by vestibular sensors (specific force and angular velocity) O: MSI	Only vestibular model, no influence of visual sensory system considered	6 DoF	None
6	Khalid et al. (Subjective Vertical Horizontal Conflict Theory) (Khalid et al., 2011)	2011	Vestibular	I: Sensed motion by vestibular sensors (specific force and angular velocity) O: Weighted MSI, combination of horizontal and vertical MSI	Only vestibular model, no influence of visual sensory system considered, designed for contemporary ships	6 DoF	None
7	Braccesi et al. (UNIPG SeMo) (Braccesi & Cianetti, 2011)	2011	Visual-vestibular	I: Sensed motion by vestibular sensors (specific force), visually perceived linear acceleration O: MSI	Only acceleration, no influence of angular velocity modeled. Uses all acceleration conflicts to predict motion sickness	3 DoF	Linear acceleration of visual flow
8	Salter et al. (Salter et al., 2020)	2020	Others (vehicle motion, passenger gaze and motion)	I: Field of view, motion of passenger head O: MISC	Data correlation that uses vehicle motion, passenger gaze, and passenger motion. Motion passenger head predicted using multibody dynamics	3 DoFs are used for specific force. Visual DoF is undefined per this paper’s definition	Weightage of elements in visual content
9	Wada et al. (Wada et al., 2020)	2020	Visual-vestibular	I: Sensed motion by vestibular sensors (specific force and angular velocity), visually perceived angular velocity O: MSI	Uses camera images to predict visually perceived motion, builds on Bos et al	6 DoF	Angular velocity of visual flow
10	Wada et al. (Wada, 2021)	2021	Vestibular	I: Sensed motion by vestibular sensors (specific force) O: MSI	Only vestibular model, no influence of visual sensory system considered	3DoF	None

The SVC theory has been extensively used in the literature to develop models to predict MS due to the efficiency of computing the conflict between the sensed and subjective verticals. The initial model proposed by Bos et al. (Bos & Bles, 1998) and later adapted by Kamiji et al. (2007) showed congruence between simulation results and the experiments performed by (McCauley et al., 1976). However, when Kamiji attempted to apply the model to replicate Griffins et al. (Donohew & Griffin, 2004) data, it was observed that the simulation results underestimate MS when the input lateral acceleration frequency was below 0.1 Hz. This can be explained by the fact that Griffin’s dataset was collected under the presence of a visual-vestibular conflict as the subjects were enclosed in a moving cabin and exposed to sinusoidal motion with no view of the external environment. Hence, a possible explanation for the deviation between Kamiji’s model results and Griffin’s data is that Kamiji’s model doesn’t account for the visual contribution to MS. The basis for this claim is that the visual component contributes to MS especially at the low frequencies (<0.2 Hz) at which Kamiji’s model was inaccurate (Duh et al., 2016). Braccesi et al. (Braccesi & Cianetti, 2011) attempted to adapt Bos’ SVC model by considering both visual and vestibular components to predict MS. However, Braccesi’s model is only able to handle linear acceleration visual-vestibular inputs, and not angular velocity. Since angular motion is expected in a moving vehicle, it is important for models to capture both linear and angular motion in all three dimensions. Wada et al. made further adaptations to Kamiji’s model to optimize the model parameters (Inoue et al., 2023) and to account for the visual contribution of angular velocity (Wada et al., 2020). However, Wada’s model still lacks the ability to handle linear visual inputs, making it not applicable in situations where there is presence of a linear visual-vestibular conflict. Hence, there is still a lack for a physiological model that captures visual-vestibular linear and angular motion stimuli in three dimensions to predict MS over the entire relevant frequency range (.01–0.5 Hz). In addition to physiological models, data-driven models can also be used to determine the correlation between motion dynamics and MS (e.g., neural networks (Saruchi et al., 2019)). However, the scarcity in MS data available in the literature limits the extent to which these models can be applied as they require an extensive amount of data to develop.

Based on the identified gaps in the models presented in the literature, MS models would benefit from an integration of the visual-vestibular interaction framework to allow for an improvement in MS prediction. To do so, it is necessary to establish an architecture that allows for the integration between the visual and vestibular stimuli. Bos et al. proposed a theoretical framework for the integration of visual-vestibular interaction and SVC models (Bos et al., 2008). In Bos’ framework, the interaction between the visual and vestibular systems results in the sensed body states, which can be compared to the output of the internal model in order to calculate MS.

Figure 1 shows a proposed simplified framework that allows for the integration of the visual-vestibular signals within the context of SVC models. Braccesi showed that human perception models are suitable for modeling the visual-vestibular interactions by assuming that the sensed body states are effectively perceived motion (ref). The perceived motion as a result of human perception modeling can be compared to the expected motion resulting from the internal model in order to yield a quantifiable measure of MS. However, Braccesi’s model only account for linear motion, limiting its applicability to situations in which angular velocity plays a significant role, such as when traveling in a vehicle.

Figure 1.

A generalized framework for the integration of visual-vestibular signals into a subjective vertical conflict model by using human motion perception.

Zacharias (1977) proposed a human motion perception model for angular velocity and Telban et al. (Telban & Cardullo, 2001) expanded on this work by proposing a human motion perception model for linear velocity. For the purposes of the model presented in this paper, the human motion perception model proposed by Telban et al. is used.

Based on the gaps identified in the literature and on the framework for the integration between human motion perception and the SVC theory, in this paper, we propose a model that

(1) Combines a 3D SVC model with a human motion perception model;

(2) Accounts for visual-vestibular linear acceleration/specific force and angular velocity motion stimuli in three dimensions;

(3) Outputs a quantitative MS rating that is congruent with results from previous studies conducted on motion platforms/simulator over the whole frequency range for angular velocity and linear specific force sinusoidal inputs;

(4) Predicts MS for realistic driving and task conditions and accounts for a wide range of MS symptoms in congruence with experimental results.

PROPOSED MODEL

In this work, we have developed a visual-vestibular motion sickness (VVMS) model to predict the motion sickness index (MSI) as a function of visual and vestibular inputs. The general framework established in the previous section is used to combine a 3D SVC model with a motion perception model. It should be noted that while the model output is in MSI, this output can be rescaled appropriately to yield accurate results in other scales as discussed in the next section. This model expands on the previously developed VVMS model proposed in Jalgaonkar et al. (2021) by accounting for the linear component of the visual input. The model combines the three-dimensional vestibular SVC model proposed by Kamiji with the visual-vestibular perception model proposed by Telban et al.

The model’s ability to capture visual-vestibular linear and angular motion in three directions makes it applicable to passenger activities, such as interacting with a handheld device or facing forward and looking out the vehicle window. It also makes it applicable to other MS triggering events such as seasickness and cybersickness. However, comparison of the model results with real world data has only been performed with respect to motion simulators and car sickness data.

For the purposes of the model presented, it is assumed that the visual and vestibular coordinate frames are collocated and coaligned as shown in Figure 2. It is assumed that such reference frame is placed as the midpoint of the line connecting the two ears, intersecting the Frankfurt plane. This assumption allows for a unified reference frame for all sensing systems considered. The model framework is shown in Figure 3 and its details are discussed next.

Figure 2.

Visual-vestibular coordinate frame.

Figure 3.

(a) Visual-vestibular motion sickness (VVMS) model architecture. (b) Detailed view of the ‘G' block. (c) Detailed view of the ‘LP' block

The inputs to the model are a, ω, a_vis, and ω_vis, which are vectors with three DoF defining the inertial linear acceleration of the head with respect to the world reference frame, the angular velocity of the head with respect to the world reference frame, the visual perception of inertial linear acceleration with respect to the visual background and the visual perception of angular velocity with respect to the visual background, respectively. These inputs are defined in terms of the coordinate frame proposed in Figure 2, so any data collected on the world reference frame must be resolved appropriately. This resolution is performed through a coordinate frame transformation based on the resting posture of subjects before the trial has started. It has been proposed in the literature that the visual system can sense linear velocity and not acceleration (Zacharias, 1977). To account for this, Telban (Telban & Cardullo, 2001) proposes that the visual inertial velocity can be differentiated to obtain visual inertial acceleration. While this is a mathematical tool, it does not represent the true cognitive processes associated with the visual signals. In addition to visual velocity, Bos et al. (2008) proposed that the visual system is able to detect the orientation of gravity. In this model, it is assumed that the visual system is able to estimate the true direction and magnitude of gravity. This allows for the quantification of the difference between visual and vestibular signals as part of the motion perception modeling.

Limited work has been done to quantify the visual input in unstructured environments, such as the inside of a vehicle cabin. Wada et al., (2020) proposed that the visual angular velocity input can be estimated using the optical flow of a video through the Farneback method. For the purposes of testing the VVMS model, only two conditions of visual angular velocity inputs are considered: (i) the visual inertial input is equal to zero, which is like the case of a passenger reading inside a moving vehicle; and (ii) the visual inertial input is identical to the vestibular input, similar to the case of a front seat passenger inside a moving vehicle with an unobstructed view of the exterior surroundings.

This paper focuses on the addition of the OTO_v-v and SCC_v-v blocks to the model described in Jalgaonkar et al. (2021). These blocks are subsystems of signals and transfer functions that are further elaborated in Figures 4 and 5.

Figure 4.

Visual-vestibular human angular motion perception model. Detailed view of the SCC_v-v block in Figure 3(a).

Figure 5.

Visual-vestibular human linear motion perception model. Detailed view of the OTO_v-v block in Figure 3(a).

In order to build the model shown in Figure 3, the generally accepted sensory conflict theory is restated in terms of a conflict between a vertical as perceived by the sensory organs and the subjective vertical as determined on the basis of previous experience. Second, this concept is integrated with estimation theory by the use of an internal model.

Physically the otolith cannot differentiate between gravitational acceleration and inertial acceleration. Thus, the otolith senses the combined gravito-inertial acceleration (GIA), as shown below.

f = a + g

\frac{d g}{d t} = - ω \times g

where ‘ a ’ is the inertial acceleration and g is the gravitational acceleration. The G block in Figure 3(b) accounts for the change in the direction of gravity with respect to the head reference frame and the relationship is captured by taking a derivative of E‘ g ’ with respect to time, as shown in the equation above. The LP block (see Figure 3(c)) is defined according to an otolith canal model in which the perceived subjective vertical can be estimated through a low pass filter with the following relationship.

\frac{d \hat{v}}{d t} = \frac{1}{τ} (\hat{f} - \hat{v}) - \hat{ω} \times \hat{v}

Here, τ is the low pass filter time constant. The first term on the right hand side gives the estimate of gravity by otolith afferents only, while the second term gives the change of gravity due to rotation only.

The sensory system model is defined by the architecture that encompasses the otolith visual-vestibular interaction model (OTO_v-v), the semicircular canal visual-vestibular interaction model (SCC_v-v) and the otolith canal interaction model (LP). The outputs of the sensory model are the human perceived dynamics with respect to the world reference frame, consisting of the perceived gravito-inertial acceleration (GIA), perceived angular velocity and perceived vertical, which are defined as $\hat{f}$ , $\hat{ω}$ , and $\hat{v}$ , respectively.

The SCC_v-v model follows the angular human motion perception architecture proposed by Telban (Telban & Cardullo, 2001), shown in Figure 4, to estimate the perceived angular velocity because of both visual and vestibular input signals. Similarly, the OTO_v-v model follows the linear human motion perception architecture proposed by Telban, shown in Figure 5, to estimate the perceived GIA because of both visual and vestibular linear input signals.

The SCC_v-v block shown in Figure 4 accounts for the coupling of visual and vestibular angular velocity. The input visual signals for both linear acceleration and angular velocity have an associated time delay defined by τ_delay, which accounts for the delays of the visual receptors, the neural transmissions and the processing during motion perception The angular velocity visual signal is passed through a sensory internal model of the semicircular canals $\bar{{S C C}_{v i s}}$ to produce a predicted vestibular signal that can be compared to the actual vestibular signal. It is assumed that the $\bar{{S C C}_{v i s}}$ transfer function is a perfect copy of the SCC transfer function, which was defined experimentally by Haslwanter (Haslwanter et al., 2000), such that

{ω_{c u e}}_{i} = \frac{τ_{d} τ_{a} s^{2}}{(τ_{d} s + 1) (τ_{a} s + 1)} ω_{i} (i = x, y, z)

where

τ_{a}

and

τ_{d}

are second order high pass filter time constants. The absolute value of the difference between the vestibular and visual cues passes through an adaptation operator to allow for long-term resolution of steady-state conflict. This adaptation operator is defined as a first order high pass filter with time constant

τ_{w}

and its output is a conflict signal

ω_{e r r}

. The optokinetic gain K is computed as a function of

ω_{e r r}

using a modified cosine bell function with conflict threshold

ε

, shown in Figure 4. The difference between the visual and vestibular cues is multiplied by the optokinetic gain and an optokinetic influence transfer function, which consists of a low pass filter with time constant

τ_{v a}

. This optokinetic influence represents the gradual build-up of self-velocity, or vection, and its output is added to the vestibular cue to yield the perceived angular velocity.

The OTO_v-v block accounts for the coupling of visual linear specific force and vestibular linear specific force. These pass through an integrator in order to yield visual and vestibular linear velocity. The remaining structure of the OTO_v-v block is like that of the SCC_v-v block, but with different model parameters,

{τ_{w}}^{'}

{τ_{v a}}^{'}

, and

ε^{'}

. Further, it is assumed that the otolith transfer function is a unity gain as proposed by Bos et al. (Bos & Bles, 1998). The visual-vestibular perceived velocity can then be differentiated to yield perceived specific force. Thus, output of the OTO_v-v block is the perceived GIA. Table 2 shows the nominal model parameter values from Kamiji et al. and Telban et al. (Kamiji et al., 2007; Telban & Cardullo, 2001).

TABLE 2:

VVMS Nominal Model Parameters

Vestibular System		Visual System
Parameter	Value	Parameter	Value
K_ω	0.8	$τ_{w}$	8s
K_a	0.1	$ε$	1.6deg/s
K_ωc	5	$τ_{v a}$	1.6s
K_vc	5	$τ_{d e l a y}$	90s
K_ac	1	${τ_{w}}^{'}$	1s
P	85	$ε^{'}$	0.2 m/s
$τ_{I}$	720s	${τ_{v a}}^{'}$	1.6s
B	0.5 m/s²
$τ_{d}$	7 s
$τ_{a}$	190s
OTO	1
$τ$	5s

RESULTS AND DISCUSSION

To assess the performance of the proposed model with respect to MS triggering events in controlled environments, we compared our model results with data collected by Donohew and Griffin (2004) for lateral oscillations. In this experiment, subjects could not see the external environment, meaning there was a visual-vestibular conflict. To enable the comparison between the model results and Griffin’s data, it was necessary to find a common scale for the two, which was achieved by normalizing MS score by the highest observed score.

The VVMS model was exercised based on two conditions, one in which the visual input was identical to the vestibular input (simulating no sensory conflict) and one in which the visual inertial input was equal to zero (simulating sensory conflict). Note that under the second condition, there is still a visual component of gravity present as explained in the previous section. The two model results were compared to Griffin’s data as a function of frequency of lateral oscillation, as shown in Figure 6. Griffin normalized the data such that the highest score observed would correspond to a rating of 1. The model replicated this standardization by normalizing the highest MS score under the same conditions to 1. It is possible to observe that the model response for the conflict condition had larger agreement with respect to the experimental data below 0.2 Hz. This is aligned with the expected behavior as visual stimuli causes higher MS in the low frequency region. The model results for the conflict condition were slightly worse in the frequency region above 0.2 Hz. When considering the entire frequency region (.01–0.8 Hz), the model performed better under the conflict condition. This is expected as subjects of the Griffin dataset were exposed to MS triggering motion under a visual-vestibular conflict condition. This supports the notion that the integration of linear human motion perception and SVC models leads to an improvement in MS prediction.

Figure 6.

Normalized MS with respect to frequency of lateral oscillation for different model results and experimental data presented by Griffin et al. (Donohew & Griffin, 2004). The grey line represents the no conflict condition and the black line represents the conflict condition.

By comparing model results to the Griffin dataset, it was shown that the VVMS model improves MS prediction for prescribed motion profiles generated by motion platforms/simulators. However, motion simulators are not able to capture the conditions a passenger would experience in a moving vehicle To further compare the model results with realistic driving experimental data, 8 subjects’ data were used from the experiments conducted by Jones et al. (Jones, Le, et al., 2019).

Jones et al. gathered MS data from passengers during test-track and on-road driving conditions (Jones, Le, et al., 2019). The test-track scripted route was completed at two different acceleration levels on the Mcity closed test facility. The on-road scripted routes were completed on local Ann Arbor roads (urban route) and Michigan highways (highway route). The on-road routes were approximately 55 min in duration, while the Mcity route was approximately 20min in duration. For each, MS ratings were obtained during every 1-minute of the in-vehicle exposure. The dataset included a total of eight subjects with two subjects from each route condition. These conditions consisted of (i) an urban route, (ii) a highway route, (iii) a low acceleration level of a scripted route and (iv) a moderate acceleration level of a scripted route. Conditions (iii) and (iv) were performed at the Mcity test facility, while conditions (i) and (ii) were performed on Michigan roads. Each subject was exposed to the same route twice, once in which they were instructed to face forward with an unconstrained head posture (no-task), and once in which they performed a task on a handheld device (task). Subjects were asked to self-report the frequency at which they become motion sick between ‘never’, ‘rarely’, ‘sometimes’ and ‘frequently’. The following is the occurrence of self-reported MS frequency for each corresponding test condition: (i) one ‘rarely’, one ‘sometimes’ (ii) one never, one ‘sometimes’, (iii) one ‘never’, one ‘frequently’, (iv) two ‘sometimes’.

The passengers’ head motion was recorded using IMUs that provided angular velocity and linear acceleration in 3 directions. The IMU data was used as an input to the model to replicate the sensory information of the otolith and semicircular canals. To account for the orientation of the IMU mounted on the passenger’s head, the rotation matrix between the passenger’s initial head acceleration and the acceleration of gravity was determined based on accelerometer data. The collected IMU data was then multiplied by the rotation transformation matrix that aligned the passenger’s initial head orientation with the world reference frame. Both linear acceleration and angular velocities were acquired at 100 Hz and filtered using a Gaussian Filter with a moving window of 80 data samples.

During the no-task condition, passengers adopted an unconstrained head posture and gaze, for which the visual inputs were assumed to be identical to the vestibular inputs. This assumption is based on a naturalistic passenger behavior study conducted by Reed et al. that showed front row passengers directed their gaze outside the windows of a vehicle close to 68% of the time (Reed et al., 2020). For the task condition, passengers performed visual-based tasks on a handheld device, for which the visual inertial input was assumed to be zero. It is assumed that the passenger views the device as an approximate static object, despite movements on the peripherals of the field of view. Notice that the act of performing a task in itself may be a confounding factor (Bos, 2015; Sepich et al., 2022). Future work may include improve estimates of the visual input using head-mounted cameras or other visual tracking methods.

Given that the SVC model was originally built to reflect results congruent with experimental data acquired using the MSI scale, it was necessary to change certain model parameters to account for the Jones et al. (Jones, Le, et al., 2019) scale. While there is no direct correspondence that can be traced between the two scales, parameter tuning can provide results that approximate the observed response. The parameter ‘P’ was modified to a value of 10 to reflect the maximum possible MS rating of the Jones scale. In addition, the parameter τ_I was significant in the replication of such results as it dictates the rate at which MS accumulates over time. In order to determine the optimal τ_I that fits the experimental data, the following optimization problem was solved

\min_{τ_{I} \in R} \sum_{j = 1}^{M} \sum_{i = 1}^{N} ‖ {M S}_{m o d e l}_{i, j} - {M S}_{e x p e r i m e n t a l}_{i, j} ‖

where j corresponds to the subject instance, M is the number of subjects assessed, i corresponds to the instance of input IMU data, N is the number of IMU data points in the experimental runs, MS_model is the model output, and MS_experimental is the moving average of the subject’s MS response with a window of 5 minutes. The result of this procedure yielded a value of τ_I equal to 216 s.

Preliminary observations suggested that the model was underestimating MS for the conditions in which the passenger performed a task. One hypothesis for this is that the model doesn’t account for the cognitive resources associated with completing the task. It has been shown that performing a task in a moving vehicle increases MS response, often causing the passenger to cease the task at hand (Jones, Le, et al., 2019). We hypothesize that passenger task performance contributes to MS not only through the introduction of a visual-vestibular conflict, but also through the addition of cognitive processing requirements. Previous studies have found a relationship between cognitive load and cybersickness (Bos, 2015; Jones, Le, et al., 2019; Sepich et al., 2022), but there is a lack of passenger vehicles studies on this matter. The proposed hypothesis is also aligned with the multiple resource theory, which states that mental operation performance degrades when the available attention resources are insufficient (Matsangas et al., 2014; Sjörs Dahlman et al., 2014). Inconsistencies across studies warrants further studies to understand the impact of task workload on MS. An alternative hypothesis for the model’s underestimation of MS is that the assumption of zero visual inertial input during task performance doesn’t effectively capture the passenger’s gaze movement. To address this discrepancy, we decreased the value of the learning feedback gains of the internal model. Modifying the value of K_ac, K_vc, K_wc to 50% of the original values yielded results with higher congruence when compared to the experimental data for task conditions.

With these model adaptations, the predicted MS response was compared to the observed MS response to assess the model’s performance. The subject’s self-reported MS data in the previously discussed Jones et al. scale (Jones, Ebert, et al., 2019), along with the model predicted outputs for the Mcity subjects are shown in Figure 7. Each panel corresponds to a different subject with the red lines representing experimental data and the blue line representing model prediction. The dashed lines correspond to the condition in which the subject was not performing a task, while the solid line corresponds to the condition in which the subject was performing a task. The subject’s self-reported MS data, along with the model predicted outputs for the on-road subjects are shown in Figure 8. Among the eight subjects, five were male and three were female and their ages ranged from 21 to 71 years old. Tables 3 and 4 show the sum of residuals between model prediction and interpolated experimental data for each test condition defined as the following.

r e s i d u a l = \sum_{j = 1}^{M} \sum_{i = 1}^{N} ‖ {M S}_{m o d e l}_{i, j} - {M S}_{e x p e r i m e n t a l}_{i, j} ‖

Figure 7.

Experimentally measured and model predicted responses for passenger motion sickness during experimental runs performed at Mcity under low and moderate vehicle acceleration profiles from Jones et al. (Jones, Le, et al., 2019). (a) Subject 1 exposed to moderate acceleration. (b) Subject 2 exposed to moderate acceleration. (c) Subject 3 exposed to low acceleration. (d) Subject 4 exposed to low acceleration

Figure 8.

Experimentally measured and model predicted responses for passenger motion sickness during experimental runs performed during on-road conditions (Jones, Le, et al., 2019). (a) Subject 5 exposed to the urban route. (b) Subject 6 exposed to the urban route. (c) Subject 7 exposed to the highway route. (d) Subject 8 exposed to the highway route.

TABLE 3:

Sum of Residuals Between MS Model Predictions and Experimental Data Corresponding to Figure 7 (Scale of 10⁴)

		A	B
Low acceleration	Task	.3087	2.0764
Low acceleration	No task	.8363	8.2187
		C	D
Moderate acceleration	Task	6.8589	92.28
Moderate acceleration	No task	.70012	17.762

TABLE 4:

Sum of Residuals Between MS Model Predictions and Experimental Data Corresponding to Figure 8 (Scale of 10⁴)

		A	B
Urban route	Task	2.4482	12.598
Urban route	No task	2.2239	18.835
		C	D
Highway route	Task	17.653	35.884
Highway route	No task	29.836	1.2817

As observed in Figures 7 and 8, the model has shown agreement with the trends observed for the experimental data. The same model parameters were used on the Mcity routes and on-road routes. Further, the same model parameters were used across different subjects. Despite changes in route and task conditions and individual subject differences, the model was able to capture the general trends observed for MS response. Both Mcity routes and road routes showed comparable results in terms of the sum of residuals (within 1 order of magnitude) between model prediction and experimental data, indicating robustness of the model across different test paths. Further, both task and no-task conditions showed comparable results in terms of the sum of residuals (within 1 order of magnitude), showing robustness of the model across different task test conditions. While this result indicates generalizability of the model, further analysis encompassing more subjects is still required. Due to the limited number of subjects available, a statistical analysis of the model results does not allow for conclusive results. This warrants further data collection and statistical analysis of the same.

Importantly, individual subjects have different levels of MS susceptibility and subsequent MS responses, which can cause deviation between model prediction and observed response. This can be observed for instance in Figures 7(d) and 8(d), where the subjects did not become motion sick despite exposure to routes that caused other subjects to become motion sick. To accurately estimate MS for each subject, individual characteristics must be taken into account. The proposed model doesn’t account for other sources of variability that may influence passenger MS response. These include, for instance, a passenger’s MS susceptibility, age, health condition (i.e., vestibular disease and migraine), physical (i.e., nutrition, sleep) and mental (i.e., stress) states, among the other factors associated with MS. Subjects’ behavior and posture during the test conditions are also hypothesized to influence the manifestation of MS symptoms. In the future, the model can be further parameterized and tuned to represent the variance in the population and provide the more accurate MS prediction given a subject’s MS susceptibility. A model that accounts for population heterogeneity will enable the identification and evaluation of MS mitigation interventions that can potentially be implemented to improve passenger comfort on an individual basis. While the model parameters can be tuned to fit individual subjects’ data, the goal of this study was to create a model that is generalizable across multiple subjects. With this, it’s possible to use the MS model to provide an initial estimate of individuals who will become motion sick for a given acceleration exposure and task condition. This allows for the simulation of MS mitigation interventions that can potentially be implemented to improve passenger comfort.

Finally, the resolution of acceleration and angular velocity reference frames based on the subject’s initial resting head posture introduces uncertainty as this head frame might not be perfectly aligned with their vestibular and/or visual systems. While these systems’ orientations are fixed with respect to the head, the resolution of IMU data on their frames might introduce alignment errors. Despite these limitations, the model was able to replicate the general trend observed for passenger MS response.

While the SVC model proposed by Kamiji et al. was originally built to provide congruent results with motion simulator data in the MSI scale, we showed that the integrated SVC and human motion perception model proposed in this paper can be tuned to yield congruent results in different scales and driving and task conditions. This indicates that the proposed model is a fundamental step towards designing MS mitigation interventions that attempt to modulate the subject motion dynamics. In the future, the model’s effectiveness on assessing MS mitigation can be further investigated to determine the extent to which the model can be used to design MS intervention systems.

CONCLUSIONS AND FUTURE WORK

A model to predict MS (VVMS) due to linear acceleration and angular velocity visual-vestibular stimuli in three dimensions was proposed along with a comparison of the model results with respect to experimental datasets. The model showed an improvement in MS prediction with respect to motion simulator MS triggering events under a visual-vestibular conflict condition. The model also showed congruent results when compared to realistic driving and task conditions from dataset captured from a larger investigation conducted by Jones et al. (Jones, Le, et al., 2019). In the future, the model can be further investigated in terms of its effectiveness in informing MS mitigating interventions. This would allow for the use of this model to design motion modulating interventions to improve passenger comfort. Further, characterization of the visual inputs can be implemented to predict MS under multiple in-vehicle scenarios. For instance, the model can be used for designing passenger interactions and experience inside AVs. As part of future research, the VVMS model can be combined with multibody dynamics simulation models which can simulate vehicle and passenger motion dynamics thereby allowing for the creation of an end-to-end MS prediction that can simulate vehicle route, vehicle motion, passenger motion, and passenger performing tasks in the vehicle.

KEY POINTS

We proposed a model that combines a 3D SVC model and a human motion perception model to provide MS prediction for visual-vestibular linear acceleration and angular velocity motion stimuli in three dimensions.

The model was able to improve prediction of MS over the whole frequency range for angular velocity and linear acceleration sinusoidal inputs when compared to results from previous studies conducted on motion platforms/simulators.

Model predictions presented similar trends to the on-road experimental data for both task and no-task conditions. The model predicted higher MS ratings for passengers performing tasks inside a moving vehicle, in line with what was observed in the experimental data.

The model can be used to design MS mitigating interventions as well as improve passenger experience in AVs.

ORCID iD

Monica L. H. Jones https://orcid.org/0000-0001-8690-9829

Footnotes

Daniel Sousa Schulman, PhD Candidate, Department of Mechanical Engineering, University of Michigan. Highest degree obtained: BS, Mechanical Engineering, 2019, University of Minnesota.

Nishant Jalgaonkar, PhD Candidate, Department of Mechanical Engineering, University of Michigan. Highest degree obtained: MSE, Mechanical Engineering, 2019, University of Michigan.

Sneha Ojha, Undergraduate Researcher, University of Michigan. Highest degree obtained: BS (in progress), Mechanical Engineering, University of Michigan.

Ana Valles, Graduate Researcher, University of Michigan. Highest degree obtained: MS, Mechanical Engineering, 2022, University of Michigan.

Monica L.H. Jones, Associate Research Scientist, University of Michigan Transportation Research Institute, University of Michigan. Highest degree obtained: PhD, Industrial Engineering, 2011, University of Michigan.

Shorya Awtar, Professor, Department of Mechanical Engineering, University of Michigan. Highest degree obtained: PhD, Mechanical Engineering, 2004, Massachusetts Institute of Technology.

References

Atsumi

Tokunaga

Kanamori

Sugawara

Yasuda

Inagaki

(2002). Evaluation of vehicle motion sickness due to vehicle vibration. JSAE Review, 23(3), 341–346. https://doi.org/10.1016/S0389-4304(02)00197-2

Bles

Bos

J. E.

de Graaf

Groen

Wertheim

A. H.

(1998). Motion sickness: Only one provocative conflict? Brain Research Bulletin, 47(5), 481–487. https://doi.org/10.1016/S0361-9230(98)00115-4

Bos

J. E.

(2015). Less sickness with more motion and/or mental distraction. Journal of Vestibular Research: Equilibrium & Orientation, 25(1), 23–33. https://doi.org/10.3233/VES-150541

Bos

J. E.

Bles

(1998). Modelling motion sickness and subjective vertical mismatch detailed for vertical motions. Brain Research Bulletin, 47(5), 537–542. https://doi.org/10.1016/s0361-9230(98)00088-4

Bos

J. E.

Bles

(2002). Theoretical considerations on canal–otolith interaction and an observer model. Biological Cybernetics, 86(3), 191–207. https://doi.org/10.1007/s00422-001-0289-7

Bos

J. E.

Bles

Groen

E. L.

(2008). A theory on visually induced motion sickness. Displays, 29(2), 47–57. https://doi.org/10.1016/j.displa.2007.09.002

Bos

J. E.

Nooij

S. A. E.

Souman

J. L.

(2021). (Im)possibilities of studying carsickness in a driving simulator. In 20th Driving Simulation & Virtual Reality Conference & Exhibition (DSC 2021 EUROPE VR) (pp. 59–63). Driving Simulation Association.

Braccesi

Cianetti

(2011). Motion sickness. Part I: Development of a model for predicting motion sickness incidence. International Journal of Human Factors Modelling and Simulation, 2(3), 163–187. https://doi.org/10.1504/IJHFMS.2011.044492

Claremont

(1931). The psychology of sea-sickness. Psyche, 11, 86–90. https://cir.nii.ac.jp/crid/1571417125726615936

10.

Diels

Howarth

P. A.

(2013). Frequency characteristics of visually induced motion sickness. Human Factors, 55(3), 595–604. https://doi.org/10.1177/0018720812469046

11.

Donohew

B. E.

Griffin

M. J.

(2004). Motion sickness: Effect of the frequency of lateral oscillation. Aviation Space & Environmental Medicine, 75(8), 649–656. https://www.ingentaconnect.com/content/asma/asem/2004/00000075/00000008/art00001

12.

Duh

H. B.-L.

Parker

D. E.

Philips

J. O.

Furness

T. A.

(2016). “Conflicting” motion cues to the visual and vestibular self- motion systems around 0.06 Hz evoke simulator sickness. Human Factors, 46(1), 142–153. https://doi.org/10.1518/hfes.46.1.142.30384

13.

Fukuda

(1976). Postural behaviour and motion sickness. Acta Oto-Laryngologica, 81(3–4), 237–241. https://doi.org/10.3109/00016487609119955

14.

Haslwanter

Jaeger

Mayr

Fetter

(2000). Three-dimensional eye-movement responses to off-vertical axis rotations in humans. Experimental Brain Research, 134(1), 96–106. https://doi.org/10.1007/s002210000418

15.

Hill

(1936). The care of the sea-sick. British Medical Journal, 2(3955), 802–807. https://doi.org/10.1136/bmj.2.3955.802

16.

Inoue

Liu

Wada

(2023). Revisiting motion sickness models based on SVC theory considering motion perception. (No. 0148–7191). SAE International.

17.

Iskander

Attia

Saleh

Nahavandi

Abobakr

Mohamed

Asadi

Khosravi

Lim

C. P.

Hossny

(2019). From car sickness to autonomous car sickness: A review. Transportation Research Part F: Traffic Psychology and Behaviour, 62, 716–726. https://doi.org/10.1016/j.trf.2019.02.020

18.

ISO 2631-1. (1997). Mechanical vibration and shock—Evaluation of human exposure to whole-body vibration.

19.

Jalgaonkar

Sousa Schulman

Ojha

Awtar

(2021). A visual-vestibular model to predict motion sickness response in passengers of autonomous vehicles. SAE International Journal of Advances and Current Practices in Mobility, 3(5), 2421–2432. https://doi.org/10.4271/2021-01-0104

20.

Jones

M. L. H.

Ebert

Reed

(2019a). Sensations associated with motion sickness response during passenger vehicle operations on a test track. SAE International Journal of Advances and Current Practices in Mobility, 1(4), 1398–1403. https://doi.org/10.4271/2019-01-0687

21.

Jones

M. L. H.

V. C.

Ebert

S. M.

Sienko

K. H.

Reed

M. P.

Sayer

J. R.

(2019b). Motion sickness in passenger vehicles during test track operations. Ergonomics, 62(10), 1357–1371. https://doi.org/10.1080/00140139.2019.1632938

22.

Kamiji

Kurata

Wada

Doi

(2007). Modeling and validation of carsickness mechanism. In SICE Annual Conference 2007 (pp. 1138–1143). IEEE. https://doi.org/10.1109/SICE.2007.4421156

23.

Kellogg

R. S.

Kennedy

R. S.

Graybiel

(1965). Motion sickness symptomatology of labyrinthine defective and normal subjects during zero gravity maneuvers. Aerospace Medicine, 36, 315–318. https://pubmed.ncbi.nlm.nih.gov/14288892/

24.

Khalid

Turan

Bos

J. E.

(2011). Theory of a subjective vertical–horizontal conflict physiological motion sickness model for contemporary ships. Journal of Marine Science and Technology, 16(2), 214–225. https://doi.org/10.1007/s00773-010-0113-y

25.

Kuiper

O. X.

Bos

J. E.

Diels

(2018). Looking forward: In-vehicle auxiliary display positioning affects carsickness. Applied Ergonomics, 68, 169–175. https://doi.org/10.1016/j.apergo.2017.11.002

26.

Kuiper

O. X.

Bos

J. E.

Diels

Cammaerts

(2019). Moving base driving simulators’ potential for carsickness research. Applied Ergonomics, 81, 102889. https://doi.org/10.1016/j.apergo.2019.102889

27.

Kuiper

O. X.

Bos

J. E.

Schmidt

E. A.

Diels

Wolter

(2020). Knowing what’s coming: Unpredictable motion causes more motion sickness. Human Factors, 62(8), 1339–1348. https://doi.org/10.1177/0018720819876139

28.

Matsangas

McCauley

M. E.

Becker

(2014). The effect of mild motion sickness and sopite syndrome on multitasking cognitive performance. Human Factors, 56(6), 1124–1135. https://doi.org/10.1177/0018720814522484

29.

McCauley

M. E.

Royal

J.W.

Wylie

C.D.

O’Hanlon

J.F.

Mackie

R.R.

, (1976). Motion sickness incidence: Exploratory studies of habituation, pitch and roll, and the refinement of a mathematical model. Human Factors Research Inc. Technical Report 1733-2, Task No. NR 105-841, April 1976. https://repository.tudelft.nl/islandora/object/uuid%3Ae191c507-e377-4ccb-9b81-64326a7e8680

30.

MedlinePlus Genetics . (2019, October 29). Motion sickness. Genetics Home Reference. https://ghr.nlm.nih.gov/condition/motion-sickness

31.

O’Hanlon

J. F.

McCauley

M. E.

(1974). Motion sickness incidence as a function of the frequency and acceleration of vertical sinusoidal motion. Aerospace Medicine, 45(4), 366–369.

32.

Oman

C. M.

(1982). A heuristic mathematical model for the dynamics of sensory conflict and motion sickness. Acta Oto-Laryngologica, 94(sup392), 4–44. https://doi.org/10.3109/00016488209108197

33.

Oman

C. M.

(1989, July 1). Sensory conflict in motion sickness: An observer theory approach. https://ntrs.nasa.gov/search.jsp?R=19900013641

34.

Reason

J. T.

(1968). Relations between motion sickness susceptibility, the spiral after-effect and loudness estimation. British Journal of Psychology, 59(4), 385–393. https://doi.org/10.1111/j.2044-8295.1968.tb01153.x

35.

Reason

J. T.

(1978). Motion sickness adaptation: A neural mismatch model. Journal of the Royal Society of Medicine, 71(11), 819–829. https://doi.org/10.1177/014107687807101109

36.

Reason

J. T.

Brand

J. J.

(1975). Motion sickness. Academic Press.

37.

Reed

M. P.

Ebert

S. M.

Jones

M. L. H.

(2020). Naturalistic passenger behavior: Posture and activities, University of Michigan, Ann Arbor, transportation research Institute. (University of Michigan, Ann Arbor, Transportation Research Institute; University of Michigan, Ann Arbor, Transportation Research Institute). University of Michigan, Ann Arbor, Transportation Research Institute. https://deepblue.lib.umich.edu/handle/2027.42/154855

38.

Reuten

A J C

Nooij

S A E

Bos

J E

Smeets

J B J

(2021). How feelings of unpleasantness develop during the progression of motion sickness symptoms. Exp Brain Res, 1432-1106239(12), 3615–3624. DOI:10.1007/s00221-021-06226-1

34595572

39.

Rolnick

Lubow

R. E.

(1991). Why is the driver rarely motion sick? The role of controllability in motion sickness. Ergonomics, 34(7), 867–879. https://doi.org/10.1080/00140139108964831

40.

Salter

Diels

Herriotts

Kanarachos

Thake

(2020). Model to predict motion sickness within autonomous vehicles. Proceedings of the Institution of Mechanical Engineers - Part D: Journal of Automobile Engineering, 234(5), 1330–1345. https://doi.org/10.1177/0954407019879785

41.

Saruchi

S. A.

Mohammed Ariff

M. H.

Zamzuri

Hassan

Wahid

(2019). Artificial neural network for modelling of the correlation between lateral acceleration and head movement in a motion sickness study. IET Intelligent Transport Systems, 13(2), 340–346. https://doi.org/10.1049/iet-its.2018.5264

42.

Schmidt

E. A.

Kuiper

O. X.

Wolter

Diels

Bos

J. E.

(2020). An international survey on the incidence and modulating factors of carsickness. Transportation Research Part F: Traffic Psychology and Behaviour, 71, 76–87. https://doi.org/10.1016/j.trf.2020.03.012

43.

Sepich

N. C.

Jasper

Fieffer

Gilbert

S. B.

Dorneich

M. C.

Kelly

J. W.

(2022). The impact of task workload on cybersickness. Frontiers in Virtual Reality, 3. https://doi.org/10.3389/frvir.2022.943409

44.

Sjörs Dahlman

Dahlman

Lundgren

Gerdle

(2014). Effects of motion sickness on encoding and retrieval performance and on psychophysiological responses. Journal of Ergonomics, 04(01). https://doi.org/10.4172/2167-7948.1000156

45.

Telban

Cardullo

(2001). An integrated model of human motion perception with visual-vestibular interaction. In AIAA Modeling and Simulation Technologies Conference and Exhibit. American Institute of Aeronautics and Astronautics. https://doi.org/10.2514/6.2001-4249

46.

Wada

(2021). Computational model of motion sickness describing the effects of learning exogenous motion dynamics. Frontiers in Systems Neuroscience, 15, 634604. https://doi.org/10.3389/fnsys.2021.634604

47.

Wada

Kawano

Okafuji

Takamatsu

Makita

(2020). A Computational Model of Motion Sickness Considering Visual and Vestibular Information. In 2020 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 1758–1763. https://doi.org/10.1109/SMC42975.2020.9283350

48.

Zacharias

(1977). Motion sensation dependence on visual and vestibular cues. [Thesis. Massachusetts Institute of Technology]. https://dspace.mit.edu/handle/1721.1/16420