Research indicates that the incidence of motion sickness is higher in females than in males (Schmidt et al., 2020). Therefore, this study recruited 10 participants (6 females and 4 males) through a paid recruitment process, with an average age of 23.7 years (standard deviation = 4.53). All experimental procedures and contents of this study were approved by the Ethics Committee of Chongqing University of Arts and Sciences (Approval No.: CQWL202540), and the handling of participants’ data complied with the Declaration of Helsinki. The screening process was carried out in three rounds: the first round aimed to ensure the physical health of each participant, excluding those with cold symptoms, a history of vestibular dysfunction, or otolith disorder. The second round involved a subjective test using the Motion Sickness Susceptibility Questionnaire (MSSQ), requiring a total score of ≥6 to confirm that participants were susceptible to motion sickness. The third round excluded participants who had abnormal routines, consumed alcohol, or engaged in vigorous physical activity within the past 24 h. After obtaining the participants’ consent, the research team signed informed consent forms with the participants, informing them of the experimental content and the tasks they were required to complete during the study.

To ensure the safety of the experiment and the authenticity of the road scenarios, the trial site was selected on the roads of the university town. The test route includes typical straight sections, one-way turns, and S-shaped curves to meet the needs of the experimental testing. The total distance of the journey is 13 kilometers, with the driving speed at the task points set to 40 km/h, and the estimated driving time is approximately 40 min (as shown in Figure 2). Before the experiment begins, drivers need to familiarize themselves with the site in advance to ensure smooth and steady transitions between different road scenarios.

The experimental scenario setup (as shown in Figure 3) involves a 5-seat automatic electric vehicle. The participant is seated in the right side of the second row. The required equipment includes:

The experiment is divided into three phases: the preparation phase, the experimental implementation phase, and the completion phase. In the preparation phase, the lead investigator first verifies the participants’ basic information, then provides a detailed explanation of the experiment’s specific content and procedures, and assists the participants in wearing the experimental equipment.

During the experimental implementation phase, the vehicle will sequentially travel to six task points. Before and after each task point, the lead investigator will collect participants’ MISC subjective motion sickness ratings through a question-and-answer format and will mark the task points in the multi-channel physiological monitoring system. At the beginning of the experiment, the vehicle will proceed to the starting point of the road, where participants will adjust to a calm state and gather 60 s of physiological data as a baseline. Following this, the vehicle will successively visit the six task points. Throughout the process, objective data will be continuously recorded and categorized based on task points and silent state markers; time stamping will be used to precisely capture data for different road conditions. At the end of Task 6, participants will collect 60 s of final state data (as shown in Figure 4). After the experiment concludes, the lead investigator will assist the participants in removing the experimental equipment and conduct in-depth interviews.

Data preprocessing was conducted on the raw subjective data to obtain valid analytical data: First, different road condition segments were precisely extracted using trial time alignment methodology; this was followed by data cleaning procedures, including the removal of outliers (Based on the Interquartile Range (IQR) method for outlier detection, with thresholds set at Q1–1.5 × IQR and Q3 + 1.5 × IQR, data points exceeding this range are flagged as outliers). Missing values are filled using the mean interpolation method based on adjacent data points. Finally, the data was standardized using Z-score standardization. After the above process, 80 sets of valid subjective data were obtained. The mean, maximum, minimum, and standard deviation of the subjective data for each task phase represent the signal values of that phase. The calculation process of the signal values is as follows Equations (1–4):

In the formula: n is the number of data points, M i is the ith data point.

The collected raw physiological signals are often contaminated by noise and artifacts, necessitating preprocessing. In this study, a multimodal physiological signal processing approach was employed for the specific preprocessing of galvanic skin response (GSR), heart rate (HR), and respiratory (RESP) signals. The GSR signal was decomposed into multiple frequency sub-bands using discrete wavelet transform (DWT), followed by soft thresholding for denoising, and the processed sub-bands were ultimately reconstructed into a complete signal through wavelet inverse transform. The HR signal underwent high-pass filtering with a cutoff frequency of 0.5 Hz to eliminate low-frequency noise and extract time-domain feature data. Additionally, the RESP signal was processed with a band-pass filter at a cutoff frequency of 0.1 Hz to extract the time-domain waveform.

Based on the six task nodes in the experimental process, the physiological signals collected synchronously were segmented, and feature parameters for each frequency band were extracted. For the galvanic skin response signal, the mean skin conductance level ( GS R mean ) and the standard deviation of skin conductance level ( GS R SD ) were extracted to represent the GSR values for that phase. For the heart rate signal, the average heart rate ( H R mean ) and the root mean square of successive differences (RMSSD) of the RR intervals were extracted to represent the heart rate values for that phase. For the respiratory signal, the mean respiratory frequency ( RES P mean ) and the standard deviation of respiratory frequency ( RES P SD ) were extracted to represent the respiratory signal values for that phase. The difference between the calm phase and the task phase is used to calculate the signal change value. The calculation processes for each signal value are as follows Equations (5–10):

In the formula: GS R mean represents the mean skin conductance level during a specific phase, GSR i denotes the skin conductance value at the ith phase, and n represents the number of sampling points.

In the formula: GS R SD represents the standard deviation of skin conductance level during a specific phase.

In the formula: H R mean represents the average heart rate during a specific phase, HR i denotes the heart rate value at the ith phase, and N indicates the number of sampling points.

In the formula: RMSSD represents the heart rate variability based on the root mean square of successive differences. RR i indicates the RR interval at the ith moment, which is the time interval between heartbeats. R R i − 1 denotes the RR interval at the (i − 1)th moment, representing the duration of the previous heartbeat cycle. RR interval refers to the time interval between two successive R waves on an electrocardiogram.

In the formula: RSE P mean represents the mean respiratory rate for a given phase, and RESP i represents the respiratory rate value for the ith phase.

In the formula: RES P SD represents the standard deviation of the respiratory rate for a given phase.

A comprehensive dataset integrating subjective data and objective physiological indicators was constructed. Subsequently, a mapping model between subjective data and objective physiological indicators was developed to enable the quantitative assessment of motion sickness severity across different road environments. The specific formula is Equations (11, 12):

Where D ( t ) is the degree of motion sickness, F represents the mapping function, G ( t ) is the electrodermal activity data, H ( t ) is the heart rate data, R ( t ) is the respiratory rate, and t is the task phase time.

The selected test route comprised six typical road conditions. Participants subjectively ranked their susceptibility to motion sickness for each condition, with Rank 1 indicating the most motion sickness-inducing scenario and Rank 6 the least. Results are presented in Table 1. Road Condition 4 was identified as the most motion sickness-inducing scenario, ranked first by 50% of participants. Neither Condition 1 nor Condition 6 received any first-place rankings. This indicates that single-direction turns (R1and R6) are less likely to induce passenger motion sickness compared to other typical road conditions.

Based on the subjective ranking results of participants’ motion sickness susceptibility across six typical road conditions (Table 1), this study further quantified the induction effect weights of different road types on motion sickness symptoms. The calculation formula is as follows:

In the formula: S For motion sickness composite score. P i : The proportion of frequency at which a specific road scenario was rated as “most likely to induce motion sickness” (ranked 1) by occupants. (Exact percentage values are provided in Table 1.) W j : denotes the weighting factor, with rank-based weights assigned as follows: Rank 1 (most motion sickness-inducing) = 5 points, Rank 2 = 4 points, Rank 3 = 3 points, Rank 4 = 2 points, Rank 5 = 1 point, Rank 6 (least motion sickness-inducing) = 0 points. Normalization Factor 2: Convert the comprehensive score to a 0–10 scale, aligning with the MISC scale range to enhance result comparability. The final motion sickness comprehensive score S ranges from 0 to 10, where higher scores indicate greater likelihood of the road condition inducing motion sickness symptoms in occupants.

This study employs the Kruskal-Wallis non-parametric test to analyze the statistical differences in physiological signal distributions among different levels of motion sickness. Furthermore, it utilizes partial η² (Partial Eta Squared, η²_p) to quantify the explanatory power of each physiological indicator regarding motion sickness severity, thereby assessing their sensitivity. Finally, Pearson correlation analysis is used to validate the consistency between subjective motion sickness ratings and objective physiological changes.

A logistic regression prediction model was established to validate the effectiveness of physiological signal changes in predicting motion sickness severity induced by different road conditions. In the predictive model, physiological data indicators were used as independent variables, while the subjective MISC scale score (0–10 points) was converted into a binary dependent variable based on the motion sickness threshold (MISC > 1), categorizing 0–1 points as “asymptomatic” and 2–10 points as “motion sickness discomfort.” The model’s predictive performance was objectively evaluated by calculating the Area Under the Receiver Operating Characteristic Curve (AUC).

Based on the variation data of MISC motion sickness levels across different road scenarios (as shown in Figure 5), a one-way ANOVA was conducted to examine the effect of road type on passengers’ subjective motion sickness levels. Prior to data analysis, the Kolmogorov–Smirnov test was performed to assess normality (p > 0.05), followed by a homogeneity of variances test. The results indicated that the mean (W = 0.455, p = 0.768), maximum (W = 0.529, p = 0.715), minimum (W = 0.508, p = 0.730), and standard deviation (W = 0.537, p = 0.704) of the subjective motion sickness levels all met the criteria for homogeneity of variances (p > 0.05). With confirmation that the data satisfied the assumptions of normality and homogeneity of variances, the one-way ANOVA results demonstrated significant statistical differences in the effects of different road scenarios on passengers’ motion sickness levels (p < 0.05).

The comprehensive calculation results of the induced effects of six types of road conditions on motion sickness are shown in Figure 6. The results indicate that the motion sickness inducing intensity of different roads ranks as follows: R4 > R5 > R3 > R2 > R1 > R6. Among them, R4 has a significantly higher comprehensive motion sickness score (M = 8.4) than the other roads, suggesting that R4 exerts the strongest stimulation on the vestibular system of passengers and is the most likely to induce motion sickness symptoms. In contrast, R6 shows the lowest comprehensive motion sickness score, indicating that R6 has relatively minimal sensitivity to inducing motion sickness.

In the objective data analysis, we examined the significance and sensitivity of physiological information, including skin conductance, heart rate, and respiratory frequency, among participants, with the results shown in Table 2. Specifically, the indicators GS R mean (H = 23.89, p < 0.01, η² = 0.070), GS R SD (H = 10.19, p < 0.05, η² = 0.029), H R mean (H = 16.25, p < 0.05, η² = 0.048), RMSSD (H = 16.61, p < 0.05, η² = 0.049), and RES P mean (H = 21.456, p < 0.05, η² = 0.063) all exhibited significant differences. In contrast, RES P SD (H = 3.098, p > 0.05, η² = 0.009) did not reach statistical significance, and its very low effect size indicates that this parameter has poor sensitivity to motion sickness responses. Among these indicators, GS R mean (η² = 0.070) showed the largest effect size, demonstrating a significant correlation with subjective data and indicating a notable sensitivity to motion sickness. Conversely, although GS R SD (η² < 0.04) achieved statistical significance, its small effect size results in relatively limited sensitivity for distinguishing levels of motion sickness. Therefore, this paper will primarily analyze the physiological characteristics of the four parameters: GS R mean , H R mean , RMSSD, and RES P mean .

The monitoring and data analysis of various physiological indicators reveal significant differences in the induction of motion sickness symptoms across different road conditions. As shown in Figure 7A, the GS R mean is significantly higher on road R4 compared to other road conditions, while the values for R1 and R6 are relatively low. The analysis of H R mean , presented in Figure 7B, indicates that the heart rate values for R4 and R5 are higher, with R4 reaching its peak average heart rate. The results for the RMSSD indicator, depicted in Figure 7C, show a contrasting trend, with higher RMSSD values for R1 and R6, while R4 has the lowest value. Additionally, the RES P mean indicator in Figure 7D shows larger values for roads R4 and R5, with R6 exhibiting the lowest value.

The statistical analysis of the objective data indicates that GS R mean , H R mean , and RES P mean are positively correlated with subjective motion sickness ratings, while the RMSSD indicator shows a negative correlation. However, all indicators consistently demonstrate that road R4 has the highest sensitivity for inducing motion sickness. In contrast, the motion sickness induced by a left turn (R6) and a right turn (R1) is relatively low, with a smaller range of difference in values.

In Pearson correlation analysis, the absolute value of the correlation coefficient (|ρ|) approaching 1 indicates a strong linear association between variables. Table 3 presents the full-sample Pearson correlation analysis results between physiological indicators and subjective motion sickness severity (MISC), demonstrating significant statistical associations (p < 0.05) between MISC and galvanic skin response mean ( GS R mean ), respiratory rate mean ( RES P mean ), heart rate mean ( H R mean ), and heart rate variability (RMSSD). Specifically, GS R mean (p = 0.685), RES P mean (p = 0.629), and H R mean (p = 0.432) exhibited significant positive correlations with MISC, with GS R mean demonstrating the strongest association. Conversely, the parasympathetic nervous activity marker RMSSD (p = −0.498) showed a significant negative correlation with MISC. The above results clearly confirm that different dimensions of the autonomic nervous system exhibit significantly differentiated response patterns in motion sickness reactions.

This study developed a logistic regression model to predict motion sickness severity (MISC) using autonomic physiological indicators: galvanic skin response ( GS R mean ), heart rate ( H R mean ), heart rate variability (RMSSD), and respiratory rate ( RES P mean ). The model identified significant positive associations between elevated MISC levels and increases in GS R mean (p < 0.05), H R mean (p < 0.05), and RES P mean (p < 0.05), while reduced RMSSD (p < 0.05) correlated negatively with symptom severity. Univariate predictive performance analysis demonstrated that GS R mean achieved the highest discriminative power (AUC = 0.8125) (as shown in Figure 8A), significantly surpassing H R mean (AUC = 0.6375) (as shown in Figure 8B), RMSSD (AUC = 0.650) (as shown in Figure 8C), and RES P mean (AUC = 0.750) (as shown in Figure 8D), with H R mean and RMSSD showing limited standalone predictive utility (AUC < 0.7). Further variable importance analysis confirmed that GSR had the greatest contribution to the model, with its predictive value clearly superior to that of heart rate and respiratory indicators. Therefore, when GS R mean is used as an input for the model, the prediction performance is optimal, achieving an accuracy of 81.25% (as shown in Table 4).