An automatic driving control method based on passenger motion sickness perception and comfort

Abstract

The application discloses an automatic driving control method based on passenger motion sickness perception and comfort, collects and pre-processes vehicle dynamic performance data, passenger physiological signals and subjective scores, extracts dynamic characteristics such as linear acceleration, angular velocity, jerk and omnidirectional acceleration, and physiological characteristics such as electrocardiogram and skin electricity, and carries out feature fusion and standardization; a passenger motion sickness perception and comfort score prediction model is constructed; individualized motion sickness threshold is constructed based on model output results, and motion sickness risk score is calculated in combination with future short-time vehicle motion characteristics, so that prospective estimation of motion sickness risk is realized; according to sensitivity typing and risk level, a hierarchical disposal strategy is executed, and vehicle driving style, in-vehicle sound and light environment and voice prompt are dynamically adjusted. The automatic driving control method based on passenger motion sickness perception and comfort can realize real-time identification and active intervention of passenger motion sickness state, and improve the ride comfort and man-machine cooperation level of the automatic driving vehicle.

Description

Technical Field

[0001] This invention belongs to the fields of autonomous driving technology and human-computer interaction, and relates to an autonomous driving control method based on occupant motion sickness perception and comfort. Background Technology

[0002] With the rapid development of intelligent connected vehicles and autonomous driving technologies, vehicle control methods are gradually shifting from manual driving to intelligent decision-making and automatic control. In an autonomous driving environment, occupants no longer directly participate in driving operations, focusing more on non-driving tasks such as reading, working, or resting. However, due to the inconsistency between dynamic stimuli such as vehicle acceleration, angular velocity, and jerk and the occupants' visual and vestibular perception, occupants are highly susceptible to motion sickness, severely impacting their riding experience and safety. Therefore, there is an urgent need for a comprehensive control method for motion sickness perception and comfort of autonomous driving occupants that can integrate and process multi-source signals, possess individualized recognition capabilities, and enable forward-looking risk prediction and tiered intervention.

[0003] Current research has attempted to analyze motion sickness responses in occupants based on visually induced experiments or changes in electroencephalogram (EEG) signals. However, these methods are often limited to specific experimental scenarios, are complex to operate, lack versatility, are difficult to integrate with real-time driving behavior, and lack the ability to fuse and analyze multimodal data. Furthermore, motion sickness risk assessments are mostly static results, lacking forward-looking predictions of future driving trajectories and risk stratification mechanisms.

[0004] The patent application CN119902620B discloses a method for subjective intervention of motion sickness in virtual reality based on perception theory. Although it can identify early signs of motion sickness and make visual perception adjustments in virtual reality experience by collecting head movement, eye movement and physiological data, this method is mainly applicable to virtual environment scenarios and cannot be applied to motion sickness monitoring and control of passengers in real vehicles with autonomous driving. It only focuses on the conflict between vision and proprioception, and does not establish a coupling model between vehicle dynamic stimuli and passenger physiological reactions, and lacks individualized identification of differences in motion sickness sensitivity. The intervention strategy is simplistic, relying only on visual display adjustments, and cannot achieve multi-dimensional dynamic intervention of driving behavior, in-vehicle sound, light and temperature environment, etc. At the same time, the method is a reactive adjustment after the fact, lacking the ability to predict risks based on future trajectories, and the algorithm is complex and difficult to meet the real-time and embedded requirements of in-vehicle systems.

[0005] In light of the aforementioned background, in order to overcome the shortcomings of existing evaluation methods, it is necessary to propose an autonomous driving control method based on occupant motion sickness perception and comfort, so as to improve the accuracy of evaluation and the practicality of control. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an autonomous driving control method based on occupant motion sickness perception and comfort.

[0007] The technical solution of this invention:

[0008] An autonomous driving control method based on occupant motion sickness perception and comfort includes the following steps:

[0009] S1. Collect target data;

[0010] S11. Determine test conditions: Clarify the test environment and conditions, including vehicle type, road conditions, and driving speed;

[0011] S12. Sensor Deployment: Sensors are deployed on the vehicle and the occupants' bodies. These sensors include a triaxial accelerometer, skin conductance acquisition electrodes and a skin conductance meter, and electrocardiogram (ECG) acquisition electrodes and an ECG meter. The computer then... Interface device connects to vehicle Interface; among which, the triaxial accelerometer is used to collect vehicle dynamic performance data, and the computer... Interface device connects to vehicle Interface reading Driving behavior parameters in the bus, skin conductance acquisition electrodes and skin conductance meters, and electrocardiogram acquisition electrodes and electrocardiogram meters are used to collect physiological indicators of the human body;

[0012] Vehicle dynamic performance data is collected by a three-axis accelerometer installed under the passenger seat. The three-axis accelerometer collects the dynamic acceleration of the vehicle along the X-axis in the direction of travel, the Y-axis perpendicular to the direction of travel on the horizontal plane, and the Z-axis perpendicular to the direction of travel on the vertical plane.

[0013] The device collects physiological indicators from the occupant. Electrodermal sampling electrodes are fixed to the thenar eminence of both hands to measure the occupant's electrodermal signals and are connected to the electrodermal analyzer via wireless transmission. Electrocardiogram (ECG) sampling electrodes are placed on the chest, left arm, and right arm to measure the occupant's ECG signals and are connected to the ECG analyzer via wireless transmission. The electrodermal analyzer and ECG analyzer are connected to a computer via their microcomputer interfaces to transmit the collected electrodermal and ECG signals. The ECG sampling electrodes are three-lead patch electrodes.

[0014] S13. Collect objective data: through vehicles The recorded vehicle dynamic performance data under different conditions are used to evaluate vehicle dynamic performance and its impact on comfort; the physiological index data of the occupants are recorded simultaneously to quantify the physiological response of the human body to vehicle motion.

[0015] S14. Subjective Data Collection: Before the test, collect motion sickness questionnaires from occupants; after the test, collect riding experience questionnaires from occupants; during the test, collect subjective rating data from occupants simultaneously with objective data collection, and evaluate comfort based on their actual feelings. The evaluation scope is... The score is divided into 1 point, with 0.5-point intervals introduced, where 1 point is extremely uncomfortable and 10 points is very comfortable.

[0016] S2, Data Processing and Feature Extraction;

[0017] S21. Data Preprocessing: The vehicle dynamic performance data, occupant physiological index data, and subjective rating data collected in step S1 are used as raw data and systematically preprocessed. First, the subjective rating data is filtered by interval; cross-modal data synchronization is performed using timestamps; low-pass filtering and moving average processing are performed on the physiological index data; and sampling reconstruction and outlier removal are performed on the vehicle dynamic performance data.

[0018] The preprocessing method is as follows: The vehicle dynamic performance data, occupant physiological index data, and subjective rating data collected in step S1 are used as raw data and systematically preprocessed. First, the subjective rating data is filtered by interval, retaining subjective ratings between [3, 8] points; second, timestamps are used for cross-modal data synchronization, and a reference time axis is established based on a unified timestamp. Using linear interpolation for any time interval Align the data above:

[0019]

[0020] in, The set of raw sampling times belonging to triaxial accelerometer, electrocorticometry, and electrocardiogram. , These are two adjacent original sampling times, and ; For the i-th triaxial accelerometer, electrocardiogram, and electrocardiogram at their own original sampling time The original sampling data on; To establish a unified reference timeline The top-aligned data outputs from the i-th triaxial accelerometer, electrodermometer, and electrocardiogram.

[0021] Physiological data were processed using low-pass filtering and moving average. First, the electrocardiogram (ECG) signal was filtered, and a band-pass filter was used to eliminate baseline drift and electromyographic interference. The frequency band was set to 0.5Hz~40Hz. It is expressed as follows:

[0022]

[0023] in, This is the spectrum of the electrocardiogram (ECG) signal; For the response function of an ideal bandpass filter:

[0024]

[0025] in, , ;

[0026] Smoothing of skin electrical signals:

[0027]

[0028] Where 2N+1 is the size of the sliding window, set to 3~9 points; The original skin electrical signal at time t is the first digit before and after the current time. The values ​​at each sampling point Centered on the current moment, the signals from 2N+1 sampling points before and after the current moment are averaged to achieve smoothing of the electrodermal signal.

[0029] The vehicle dynamic performance data is sampled, reconstructed, and outlier removed. First, a first-order low-pass filter is used to suppress high-frequency interference on the vehicle acceleration and jerk signals.

[0030]

[0031] in, This is the filtered output at the current moment; This refers to acceleration data within the vehicle's dynamic performance data. It is a smoothing factor, and Set it to 0.1~0.3; This is the filtered output from the previous time step;

[0032] Next, outlier detection and removal are performed, using the Z-score method to check whether a point is an outlier at each time step:

[0033]

[0034] in, The mean value within the selected sliding window; This represents the standard deviation within the selected sliding window.

[0035] like If so, it is determined to be an outlier and removed;

[0036] Finally, resampling and frequency unification are performed, with the original data at the resampling point. The new value is obtained by linear interpolation:

[0037]

[0038] in, , The raw data consists of vehicle dynamic performance data, occupant physiological indicators, and subjective rating data. Adjacent sampling points, ; , The data values ​​corresponding to the sampling points in the raw data consisting of vehicle dynamic performance data, occupant physiological index data, and subjective rating data.

[0039] S22. Feature Extraction: Extract physiological features describing the motion sickness response of passengers from the collected physiological index data, including feature indicators of heart rate variability and skin conductance response; extract dynamic performance features reflecting the motion sickness stimuli induced by the vehicle from the collected vehicle dynamic performance data, including linear acceleration feature indicators, angular velocity and angular acceleration feature indicators, jerk feature indicators and omnidirectional acceleration composite indicator.

[0040] The feature extraction method is as follows: Feature extraction is performed on the data preprocessed in step S21. First, heart rate variability is extracted, specifically the standard deviation of heart rate variability. Indicators for measuring overall heart rate variability:

[0041]

[0042] in, for The total number of intervals; For the first indivual Interval; For all within this window The average value of the interval; among which, The interval is one period within a window of an electrocardiogram. Wave to the next The time interval between waves;

[0043] The root mean square (RMSSD) of adjacent heart rate variability was extracted as a measure of short-term changes in parasympathetic activity.

[0044]

[0045] Extract the LF / HF ratio, a frequency domain feature of heart rate variability, calculate the frequency domain energy distribution of heart rate variability to represent autonomic balance, and use Fast Fourier Transform to calculate the power spectrum.

[0046]

[0047] in, This refers to the spectral power at low frequencies. This refers to the spectral power at high frequencies;

[0048] Secondly, the characteristics of the skin conductance response were extracted, and the average conductivity of the skin conductance response was extracted to measure the overall tension level.

[0049]

[0050] in, This is the original skin conductance signal; For the observation window length;

[0051] The frequency of skin conductance responses was extracted, and the peak value was detected using a threshold method. The number of responses per second was defined as:

[0052]

[0053] in, for Number of skin conductance responses per second;

[0054] Maximum amplitude of skin conductance response:

[0055]

[0056] in, Peak value; The baseline of conductance before rise;

[0057] Feature extraction is performed on vehicle dynamic performance data. Linear acceleration features include mean, standard deviation, and peak acceleration; angular velocity and angular acceleration features include angular acceleration and yaw frequency; jerk features include jerk and root mean square jerk; and omnidirectional acceleration composite features include total acceleration.

[0058] S23. Feature Fusion and Standardization: The extracted physiological features and vehicle dynamic performance features are concatenated to form a unified feature vector, and the features are standardized by normalization or standard deviation scaling for subsequent modeling.

[0059] The method for feature fusion and standardization is as follows: First, multimodal feature fusion is performed, concatenating feature vectors from different sources along their feature dimensions to form a unified high-dimensional input vector. Let:

[0060] Let m be the features extracted from the electrocardiogram signal;

[0061] Let n be the features extracted from the electrodermal signal;

[0062] Let r be the r features extracted from vehicle acceleration;

[0063] The fused feature vector is then:

[0064]

[0065] Among them, symbols This indicates concatenation based on feature dimensions;

[0066] In step S21, all features have been processed using Z-score normalization before fusion, as shown in the following formula:

[0067]

[0068] in, This is the a-th extracted feature value; Let be the mean of the a-th extracted feature across all samples; Let be the standard deviation of the extracted feature a across all samples; These are the standardized feature values, with a mean of 0 and a standard deviation of 1.

[0069] S3. Construct a prediction model for occupant motion sickness perception and comfort rating:

[0070] S31. Feature integrity check and redundancy control: Perform integrity checks on the vehicle dynamic performance features and occupant physiological features extracted in step S2, and then identify the high linear correlation between different features. If redundancy exists, only retain the features with more direct physical meaning and higher signal stability.

[0071] The method for feature integrity checking and redundancy control is as follows: First, the integrity of the collected vehicle dynamic features and physiological features is checked, and features with a missing rate of more than 20% or variance close to zero are removed.

[0072] The Pearson correlation coefficient is used to identify high linear correlations between different features. The Pearson correlation coefficient between any two features is defined as follows:

[0073]

[0074] in, , Let p-th and q-th fused features be the sequence of values ​​among all features; , The value of the corresponding feature in the k-th sample; , denoted as the sample mean of the feature; u is the sample size; The correlation coefficient represents the degree of linear correlation between two features. ;

[0075] when When it is assumed that the two features are highly correlated or redundant, the system retains only the feature with the more direct physical meaning and higher signal stability.

[0076] S32. Modeling and Prediction: Input the vehicle's dynamic performance characteristics and the occupants' physiological characteristics into the constructed occupant motion sickness perception and comfort score prediction model based on a gradient boosting framework that builds multiple decision trees.

[0077] Motion sickness sensitivity levels are identified using the LightGBM classification model. Multiple decision trees are constructed based on a gradient boosting framework, and the prediction function is expressed as follows:

[0078]

[0079] in, To predict motion sickness sensitivity labels; x is a sample feature vector that integrates physiological and vehicle dynamics features; Z represents the set of classification labels: {low sensitivity, medium sensitivity, high sensitivity}; Z is the number of base learner decision trees in the LightGBM classification model. Output the score of the z-th tree in category c;

[0080] The LightGBM classification model uses the cross-entropy loss function for classification problems, and the loss is defined as follows:

[0081]

[0082] in, To determine whether sample v belongs to category c, and ; This represents the predicted probability of sample v belonging to class c by the LightGBM classification model.

[0083] Training efficiency is improved by constructing gradient histograms and using a leaf-based tree growth strategy; for each round of newly added tree models... The optimization objective is:

[0084]

[0085] in, The learning rate controls the step speed of the LightGBM classification model; , is the prediction score of the LightGBM classification model for the input sample x in the w-th iteration, and is the cumulative output within the model;

[0086] Motion sickness score prediction was performed using an XGBoost regression model. The XGBoost regression model uses multiple CART trees in an additive structure to stack the predicted values.

[0087]

[0088] in, To predict motion sickness scores, the domain is defined as [1, 10]; S is the number of base learner regression trees in the XGBoost regression model; This represents the prediction result of the s-th regression tree; The set of all possible regression tree functions;

[0089] Using mean squared error as the regression loss function:

[0090]

[0091] in, This represents the true motion sickness score for the g-th sample. These are the predicted values ​​from the XGBoost regression model;

[0092] The objective function of the XGBoost regression model is:

[0093]

[0094] in, Let be the objective function of the XGBoost regression model in the t-th iteration, which includes a prediction error term and a model complexity regularization term. In the prediction error term, For real labels, This is the value predicted in the previous round. Output the current tree. This is the loss function used to calculate the error between the predicted value and the true value;

[0095] The model complexity regularization term is:

[0096]

[0097] Where T is the current number of leaf nodes; The weights are those of the leaf nodes; These are the regularization parameters for the XGBoost regression model.

[0098] S4. Comfort evaluation results output and intelligent control strategy:

[0099] S41. Motion Sickness Sensitivity Classification: Based on the motion sickness sensitivity labels output by the occupant motion sickness perception and comfort score prediction model constructed in step S3, determine the occupant motion sickness sensitivity type.

[0100] S42. Motion sickness risk prediction: Combine the future trajectory data output by the trajectory planning module of the autonomous vehicle to calculate the normalized ratio of the root mean square of lateral acceleration, jerk and omnidirectional acceleration, and use the future short-term motion sickness risk score level output by the multi-decision tree occupant motion sickness perception and comfort score prediction model as the input signal for triggering the hierarchical improvement strategy.

[0101] The method for predicting motion sickness risk is as follows: First, the trajectory planning module of the autonomous vehicle is invoked to obtain candidate trajectory data within a time window of 3-8 seconds in the future, and representative vehicle kinematic features, including lateral acceleration, are extracted. accelerometer and root mean square of omnidirectional acceleration The corresponding individualization thresholds are respectively , , Its value is derived from the vehicle kinematics characteristics corresponding to when the subjective score drops to 6 points. This threshold is dynamically updated during operation based on the output of the XGBoost regression model. To unify the scale, the feature normalization ratio is calculated first.

[0102]

[0103] The system simultaneously reads the current comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3. And define discomfort as:

[0104]

[0105] in, , The larger the value, the more likely the occupant is already showing signs of discomfort.

[0106] The future short-term motion sickness risk score is defined as:

[0107]

[0108] in, This represents a score indicating the risk of short-term motion sickness in the future; a higher value indicates a higher risk. , , , These are the weighting coefficients, and ;

[0109] When omnidirectional acceleration features are not used , , When it contains omnidirectional acceleration characteristics, , ;

[0110] When R < 0.6, it is considered low risk, indicating that the probability of discomfort caused by the future trajectory is low; when 0.6 ≤ R < 1.0, it is considered medium risk, indicating that mild discomfort may occur in the future; when R ≥ 1.0, it is considered high risk, indicating that a significant motion sickness reaction is very likely to be induced in the next few seconds.

[0111] S43. Layered improvement strategy: Select different vehicle control and cabin environment adjustment strategies based on the occupant's motion sickness sensitivity type and future short-term motion sickness risk score level.

[0112] The method of the stratified improvement strategy is as follows: Based on the sensitivity label and comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3, determine the current susceptibility level of the occupant, assess the motion sickness risk by combining the future short-term motion sickness risk score, and implement the corresponding vehicle and environmental control strategies according to the two-dimensional stratification rules.

[0113] The vertical basic stratification is as follows: for low-sensitivity occupants, the vehicle control constraints are relatively loose, with main voice prompts and mild smoothing; for moderately sensitive occupants, the vehicle is moderately constrained, and environmental adjustments for wind, temperature, and light are increased; for highly sensitive occupants, the vehicle strictly constrains jerk and lateral acceleration, and multimodal measures are used to alleviate motion sickness.

[0114] Lateral motion sickness risk stratification is as follows: for low-risk cases with R < 0.6, the vehicle maintains the current strategy and continues to monitor; for medium-risk cases with 0.6 ≤ R < 1.0, the vehicle implements a mild smoothing or environmental pre-adjustment strategy in advance; for high-risk cases with R ≥ 1.0, the vehicle immediately activates the corresponding level of conservative driving mode and multimodal intervention.

[0115] When the comfort score is less than or equal to 3, or when a significant physiological trigger is detected, the root mean square difference between adjacent heart rate variability decreases by more than or equal to 20%, or the frequency or amplitude of skin conductance responses increases by more than or equal to 50%, the high-risk strategy is directly implemented regardless of the future short-term motion sickness risk score R level; otherwise, the control level is determined by the combination of sensitivity label and risk score.

[0116] The combined decision matrix of sensitivity labels and risk scores is as follows:

[0117] Low sensitivity + low risk: Maintain original driving style; vehicle only provides voice prompts;

[0118] Low sensitivity + medium risk: jerk≤0.8m / s³, a≤1.2m / s²; vehicle temperature reduced by 1-3℃ or ventilation turned on;

[0119] Low sensitivity + high risk: jerk≤0.6m / s³, a≤1.0m / s²; vehicle HUD guide strips provide advance warning;

[0120] Medium sensitivity + low risk: jerk ≤ 0.8 m / s³; vehicle monitoring of physiological fluctuations;

[0121] Medium sensitivity + medium risk: jerk ≤ 0.6 m / s³; vehicle temperature reduced by 1-3℃, maintain ventilation;

[0122] Medium-sensitive + high-risk: jerk≤0.5m / s³; vehicle HUD guide strip advance warning and voice prompts;

[0123] High sensitivity + low risk: jerk≤0.6m / s³; soft light and constant temperature inside the vehicle;

[0124] High sensitivity + medium risk: jerk≤0.5m / s³; vehicle temperature reduction of 1-3℃, maintaining ventilation, and seat vibration warning;

[0125] High sensitivity + high risk: jerk≤0.4m / s³; conservative driving, HUD guide strips provide advance warnings, voice prompts and gentle cooling, and maintain ventilation.

[0126] S44. Continuous monitoring and dynamic adjustment: After the vehicle control and cabin environment adjustment strategies are implemented, the comfort score is continuously monitored, the motion sickness sensitivity type and the changes in the future short-term motion sickness risk score are determined, and the vehicle control and cabin environment adjustment strategies are dynamically adjusted online according to the tiered improvement strategy. If the future short-term motion sickness risk score decreases within 5 consecutive seconds and the comfort score increases by more than or equal to 1 point, the constraints are gradually relaxed; if the future short-term motion sickness risk score remains high or decreases by more than or equal to 2 points within 2 consecutive seconds, the control is tightened again.

[0127] The beneficial effects of this invention are as follows: This invention combines physiological indicators with vehicle dynamics features, collecting and inputting data based on a fixed driving event window, avoiding feature ambiguity caused by sliding slices and improving the model's response accuracy. The lightweight dual-model structure constructed in this invention uses LightGBM for occupant motion sickness sensitivity identification and XGBoost for subjective comfort rating prediction, exhibiting good generalization ability and interpretability, facilitating subsequent feature tracing and model optimization. This invention introduces a short-term motion sickness risk score, combined with dynamic parameters of the planned trajectory for the next 3-8 seconds, to achieve forward prediction of motion sickness risk, identifying potential risks before discomfort occurs and providing proactive intervention basis for vehicle decision-making. Based on occupant sensitivity labels and risk levels, this invention proposes a multi-level comprehensive intervention scheme, significantly improving ride comfort. Attached Figure Description

[0128] Figure 1 This is a flowchart illustrating the overall process of an autonomous driving control method based on occupant motion sickness perception and comfort according to the present invention.

[0129] Figure 2This invention provides multimodal data for an autonomous driving control method based on occupant motion sickness perception and comfort.

[0130] Figure 3 This is a schematic diagram of a hierarchical improvement strategy for an autonomous driving control method based on occupant motion sickness perception and comfort, according to the present invention. Detailed Implementation

[0131] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0132] like Figure 1 As shown in the figure, this embodiment exemplarily illustrates an autonomous driving control method based on occupant motion sickness perception and comfort, including the following steps:

[0133] 1) Determining Test Conditions: First, the vehicle type should possess typical autonomous driving capabilities and stable power response characteristics. Second, the test road environment should include a mixed traffic scenario encompassing urban arterial roads and expressways, incorporating various typical operating conditions such as lane changing, acceleration, braking, and ramp maneuvers to ensure comprehensive coverage of diverse driving scenarios. Regarding driving speed settings, to avoid weakening the impact of operations at excessively low speeds or uncontrollable data fluctuations at high speeds, the test limits the vehicle's speed to... Runs within range, and through Real-time speed monitoring. In each set scenario, the operation trigger time is calibrated using preset markers, voice prompts, or manual recording by the tester to ensure consistency and timing alignment between behavioral events and collected signals. Furthermore, to avoid environmental factors interfering with the occupants' physiological state, the entire experiment was conducted under relatively constant outdoor temperature and humidity conditions, while ensuring the vehicle interior temperature was maintained at [specific temperature range]. To avoid errors in skin electrical response caused by hot and cold stimuli;

[0134] 2) Sensor Placement: Vehicle dynamic performance data acquisition utilizes a triaxial accelerometer. The sensor is fixed to the bottom of the passenger seat using double-sided tape and a shock-absorbing bracket, ensuring close contact with the rigid connection points of the vehicle structure to avoid interference from the seat's flexible structure on the vibration signals. Physiological data acquisition for occupants utilizes physiological signal acquisition equipment, such as… Multi-channel physiological signal acquisition system, ECG signal uses The module acquires data, with three-lead electrodes attached to the chest, left arm, and right arm; skin electrical signals are transmitted through... The module acquires data, and the electrodes are fixed to the thenar eminence of both hands to improve conductivity and reduce motion artifacts.

[0135] To ensure stable electrode adhesion and excellent signal channel quality, all participants were required to perform a maximal voluntary contraction test before the formal experiment. This involved the participant actively moving both arms slightly, simultaneously opening their palms and then clenching their fists, simulating potential displacement interference during the experiment. During this time, the instrument simultaneously recorded changes in skin conductance and electrocardiogram signals, detecting electrode detachment, signal loss, severe noise, or spurious abnormalities. This process also recorded the peak response of skin conductance under maximum tension, serving as a reference for individual physiological response amplitude and aiding in subsequent data normalization and response sensitivity analysis.

[0136] 3) Collecting objective data: During the experiment, the computer first uses the vehicle... Interface access The system uses a bus to simultaneously extract the dynamic acceleration of the vehicle along the X-axis (direction of travel), the Y-axis (perpendicular to the direction of travel on the horizontal plane), and the Z-axis (perpendicular to the direction of travel on the vertical plane) in typical operating scenarios such as lane changing, acceleration, and overtaking.

[0137] Secondly, during the acquisition experiment, electrocardiogram (ECG) and electrodermal (TE) signals were recorded simultaneously. Fixed time windows were set before and after each driving operation. The ECG extraction window was 10 seconds before and 20 seconds after the driving operation, and the TE signal extraction window was 10 seconds before and after the driving operation.

[0138] 4) Collect subjective data: Before the experiment, collect motion sickness questionnaires from passengers. The questionnaires cover basic information, history of motion sickness, health status, lifestyle habits and other dimensions. They are used to screen subjects and control experimental variables. The time to fill out the questionnaire is controlled within 5 minutes.

[0139] During the test, after each driving operation, occupants were required to subjectively rate their experience based on their personal feelings. The rating system used an equal-interval scoring method from 1 to 10 points, where 1 point represents extreme discomfort and 10 points represent complete comfort, with 0.5-point intervals introduced. The ratings were completed by occupants via tablet interface or voice input and were automatically linked to the operation number, timestamp, and corresponding multimodal data collection. The subjective comfort evaluation table is shown in Table 1.

[0140] Table 1 Subjective Evaluation Table of Comfort

[0141]

[0142] After the experiment, passenger experience questionnaires were collected. The questionnaires were used to assess passengers’ subjective feelings about the vehicle’s movement during the experiment, focusing on motion sickness symptoms, comfort and perception of vehicle dynamics. The Likert scale and open-ended questions were used to facilitate quantitative analysis and qualitative feedback.

[0143] 5) Data processing and feature extraction:

[0144] Data preprocessing: The collected vehicle dynamic performance data, occupant physiological index data, and subjective rating data are used as raw data and systematically preprocessed. First, subjective ratings are filtered by range, retaining those between [3, 8]. Second, timestamps are used for cross-modal data synchronization, and a reference time axis is established based on a unified timestamp. Using linear interpolation for any time interval Align the data above:

[0145]

[0146] in, The set of raw sampling times belonging to triaxial accelerometer, electrocorticometry, and electrocardiogram. , These are two adjacent original sampling times, and ; For the i-th triaxial accelerometer, electrocardiogram, and electrocardiogram at their own original sampling time The original sampling data on; To establish a unified reference timeline The top-aligned data outputs from the i-th triaxial accelerometer, electrodermometer, and electrocardiogram.

[0147] Physiological indicators were processed using low-pass filtering and moving average. First, the electrocardiogram (ECG) signal was filtered using a band-pass filter to eliminate baseline drift and electromyographic interference, with a frequency band set to 0.5–40 Hz. It is expressed as follows:

[0148]

[0149] in, This is the spectrum of the electrocardiogram (ECG) signal; For the response function of an ideal bandpass filter:

[0150]

[0151] in, , ;

[0152] Smoothing of skin electrical signals:

[0153]

[0154] Where 2N+1 is the size of the sliding window, set to 3~9 points; The original skin electrical signal at time t is the first digit before and after the current time. The values ​​at each sampling point Centered on the current moment, the signals from 2N+1 sampling points before and after the current moment are averaged to achieve smoothing of the electrodermal signal.

[0155] The vehicle dynamic performance data is sampled, reconstructed, and outlier removed. First, a first-order low-pass filter is used to suppress high-frequency interference on the vehicle acceleration and jerk signals.

[0156]

[0157] in, This is the filtered output at the current moment; This is the original acceleration value; It is a smoothing factor, and Set it to 0.1~0.3; This is the filtered output from the previous time step;

[0158] Next, outlier detection and removal are performed, using the Z-score method to check whether a point is an outlier at each time step:

[0159]

[0160] in, The mean value within the selected sliding window; This represents the standard deviation within the selected sliding window.

[0161] like If so, it is determined to be an outlier and removed;

[0162] Finally, resampling and frequency unification are performed, with the original data at the resampling point. The new value is obtained by linear interpolation:

[0163]

[0164] in, , For the original data and Adjacent sampling points, ; , The data values ​​corresponding to the sampling points in the raw data composed of vehicle dynamic performance data, occupant physiological index data, and subjective rating data;

[0165] The feature extraction method is as follows: Feature extraction is performed on the preprocessed data. First, heart rate variability is extracted, specifically the standard deviation of heart rate variability. Indicators for measuring overall heart rate variability:

[0166]

[0167] in, for The total number of intervals; For the first indivual Interval; For all within this window The average value of the interval; among which, The interval is one period within a window of an electrocardiogram. Wave to the next The time interval between waves;

[0168] The root mean square (RMSSD) of adjacent heart rate variability was extracted as a measure of short-term changes in parasympathetic activity.

[0169]

[0170] Extract the LF / HF ratio, a frequency domain feature of heart rate variability, calculate the frequency domain energy distribution of heart rate variability to represent autonomic balance, and use Fast Fourier Transform to calculate the power spectrum.

[0171]

[0172] in, This refers to the spectral power at low frequencies. This refers to the spectral power at high frequencies;

[0173] Secondly, the characteristics of the skin conductance response were extracted, and the average conductivity of the skin conductance response was extracted to measure the overall tension level.

[0174]

[0175] in, This is the original skin conductance signal; For the observation window length;

[0176] The frequency of skin conductance responses was extracted, and the peak value was detected using a threshold method. The number of responses per second was defined as:

[0177]

[0178] in, for Number of skin conductance responses per second;

[0179] Maximum amplitude of skin conductance response:

[0180]

[0181] in, Peak value; The baseline of conductance before rise;

[0182] Feature extraction is performed on vehicle dynamic performance data. Linear acceleration features include mean, standard deviation, and peak acceleration; angular velocity and angular acceleration features include angular acceleration and yaw frequency; jerk features include jerk and root mean square jerk; and omnidirectional acceleration composite features include total acceleration.

[0183] The method for feature fusion and standardization is as follows: First, multimodal feature fusion is performed, concatenating feature vectors from different sources along their feature dimensions to form a unified high-dimensional input vector. Let:

[0184] Let m be the features extracted from the electrocardiogram signal;

[0185] Let n be the features extracted from the electrodermal signal;

[0186] Let r be the r features extracted from vehicle acceleration;

[0187] The fused feature vector is then:

[0188]

[0189] Among them, symbols This indicates concatenation based on feature dimensions;

[0190] In step S21, all features have been processed using Z-score normalization before fusion, as shown in the following formula:

[0191]

[0192] in, This is the a-th extracted feature value; Let be the mean of the a-th extracted feature across all samples; Let be the standard deviation of the extracted feature a across all samples; These are the standardized feature values, with a mean of 0 and a standard deviation of 1.

[0193] 6) Construct a prediction model for motion sickness perception and comfort rating of autonomous driving occupants based on multimodal data:

[0194] The method for feature integrity checking and redundancy control is as follows: First, the integrity of the collected vehicle dynamic features and physiological features is checked, and features with a missing rate of more than 20% or variance close to zero are removed.

[0195] The Pearson correlation coefficient is used to identify high linear correlations between different features. The Pearson correlation coefficient between any two features is defined as follows:

[0196]

[0197] in, , Let p-th and q-th fused features be the sequence of values ​​among all features; , The value of the corresponding feature in the k-th sample; , denoted as the sample mean of the feature; u is the sample size; The correlation coefficient represents the degree of linear correlation between two features. ;

[0198] when When it is assumed that the two features are highly correlated or redundant, the system retains only the feature with the more direct physical meaning and higher signal stability.

[0199] The model is constructed as follows: LightGBM classification model is used to identify motion sickness sensitivity levels; multiple decision trees are built based on a gradient boosting framework; and the prediction function is expressed as:

[0200]

[0201] in, To predict motion sickness sensitivity labels; x is a sample feature vector that integrates physiological and vehicle dynamics features; Z represents the set of classification labels: {low sensitivity, medium sensitivity, high sensitivity}; Z is the number of base learner decision trees in the LightGBM classification model. Output the score of the z-th tree in category c;

[0202] The LightGBM classification model uses the cross-entropy loss function for classification problems, and the loss is defined as follows:

[0203]

[0204] in, To determine whether sample v belongs to category c, and ; This represents the predicted probability of sample v belonging to class c by the LightGBM classification model.

[0205] Training efficiency is improved by constructing gradient histograms and using a leaf-based tree growth strategy; for each round of newly added tree models... The optimization objective is:

[0206]

[0207] in, The learning rate controls the step speed of the LightGBM classification model; , is the prediction score of the LightGBM classification model for the input sample x in the w-th iteration, and is the cumulative output within the model;

[0208] Motion sickness score prediction was performed using an XGBoost regression model. The XGBoost regression model uses multiple CART trees in an additive structure to stack the predicted values.

[0209]

[0210] in, To predict motion sickness scores, the domain is defined as [1, 10]; S is the number of base learner regression trees in the XGBoost regression model; This represents the prediction result of the s-th regression tree; The set of all possible regression tree functions;

[0211] Using mean squared error as the regression loss function:

[0212]

[0213] in, This represents the true motion sickness score for the g-th sample. These are the predicted values ​​from the XGBoost regression model;

[0214] The objective function of the XGBoost regression model is:

[0215]

[0216] in, Let be the objective function of the XGBoost regression model in the t-th iteration, which includes a prediction error term and a model complexity regularization term. In the prediction error term, For real labels, This is the value predicted in the previous round. Output the current tree. This is the loss function used to calculate the error between the predicted value and the true value;

[0217] The model complexity regularization term is:

[0218]

[0219] Where T is the current number of leaf nodes; The weights are those of the leaf nodes; These are the regularization parameters for the XGBoost regression model.

[0220] 7) Comfort evaluation results output and intelligent control strategy:

[0221] The method for classifying motion sickness sensitivity is as follows: based on the motion sickness sensitivity labels output by the constructed multi-decision-tree occupant motion sickness perception and comfort score prediction model, the occupant motion sickness sensitivity type is determined.

[0222] The method for predicting motion sickness risk is as follows: First, the trajectory planning module of the autonomous vehicle is invoked to obtain candidate trajectory data within a time window of 3-8 seconds in the future, and representative vehicle kinematic features, including lateral acceleration, are extracted. accelerometer and root mean square of omnidirectional acceleration The corresponding individualization thresholds are respectively , , Its value is derived from the vehicle kinematics characteristics corresponding to when the subjective score drops to 6 points. This threshold is dynamically updated during operation based on the output of the XGBoost regression model. To unify the scale, the feature normalization ratio is calculated first.

[0223]

[0224] The system simultaneously reads the current comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3. And define discomfort as:

[0225]

[0226] in, , The larger the value, the more likely the occupant is already showing signs of discomfort.

[0227] The future short-term motion sickness risk score is defined as:

[0228]

[0229] in, This represents a score indicating the risk of short-term motion sickness in the future; a higher value indicates a higher risk. , , , These are the weighting coefficients, and ;

[0230] When omnidirectional acceleration features are not used , , When it contains omnidirectional acceleration characteristics, , ;

[0231] When R < 0.6, it is considered low risk, indicating that the probability of discomfort caused by the future trajectory is low; when 0.6 ≤ R < 1.0, it is considered medium risk, indicating that mild discomfort may occur in the future; when R ≥ 1.0, it is considered high risk, indicating that a significant motion sickness reaction is very likely to be induced in the next few seconds.

[0232] The tiered improvement strategy is as follows: Based on the sensitivity label and comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3, the occupant's current susceptibility level is determined. The motion sickness risk is assessed by combining the future short-term motion sickness risk score. Corresponding vehicle and environmental control strategies are then executed according to the two-dimensional tiered rules. Autonomous vehicles primarily improve occupant discomfort through three aspects: driving style adjustment, environmental control, and human-machine interaction prompts.

[0233] The vertical basic stratification is as follows: for low-sensitivity occupants, the vehicle control constraints are relatively loose, with main voice prompts and mild smoothing; for moderately sensitive occupants, the vehicle is moderately constrained, and environmental adjustments for wind, temperature, and light are increased; for highly sensitive occupants, the vehicle strictly constrains jerk and lateral acceleration, and multimodal measures are used to alleviate motion sickness.

[0234] Lateral motion sickness risk stratification is as follows: for low-risk cases with R < 0.6, the vehicle maintains the current strategy and continues to monitor; for medium-risk cases with 0.6 ≤ R < 1.0, the vehicle implements a mild smoothing or environmental pre-adjustment strategy in advance; for high-risk cases with R ≥ 1.0, the vehicle immediately activates the corresponding level of conservative driving mode and multimodal intervention.

[0235] When the comfort score is less than or equal to 3, or when a significant physiological trigger is detected, the root mean square difference of adjacent heart rate variability decreases by more than or equal to 20%, or the frequency or amplitude of skin conductance responses increases by more than or equal to 50%, regardless of the future short-term motion sickness risk score R level, the high-risk strategy is directly implemented; otherwise, the control level is determined by the combination of sensitivity label and risk score.

[0236] The combined decision matrix of sensitivity labels and risk scores is as follows:

[0237] Low sensitivity + low risk: Maintain original driving style; vehicle only provides voice prompts;

[0238] Low sensitivity + medium risk: jerk≤0.8m / s³, a≤1.2m / s²; vehicle temperature reduced by 1-3℃ or ventilation turned on;

[0239] Low sensitivity + high risk: jerk≤0.6m / s³, a≤1.0m / s²; vehicle HUD guide strips provide advance warning;

[0240] Medium sensitivity + low risk: jerk ≤ 0.8 m / s³; vehicle monitoring of physiological fluctuations;

[0241] Medium sensitivity + medium risk: jerk ≤ 0.6 m / s³; vehicle temperature reduced by 1-3℃, maintain ventilation;

[0242] Medium-sensitive + high-risk: jerk≤0.5m / s³; vehicle HUD guide strip advance warning and voice prompts;

[0243] High sensitivity + low risk: jerk≤0.6m / s³; soft light and constant temperature inside the vehicle;

[0244] High sensitivity + medium risk: jerk≤0.5m / s³; vehicle temperature reduction of 1-3℃, maintaining ventilation, and seat vibration warning;

[0245] High sensitivity + high risk: jerk≤0.4m / s³; conservative driving, HUD guide strips provide advance warnings, voice prompts and gentle cooling, and maintain ventilation.

[0246] The method for continuous monitoring and dynamic adjustment is as follows: After the vehicle control and cabin environment adjustment strategies are implemented, the comfort score, motion sickness sensitivity type, and changes in the future short-term motion sickness risk score level are continuously monitored. The vehicle control and cabin environment adjustment strategies are dynamically adjusted online according to the tiered improvement strategy. If the future short-term motion sickness risk score decreases within 5 consecutive seconds and the comfort score increases by more than or equal to 1 point, the constraints are gradually relaxed. If the future short-term motion sickness risk score remains high or decreases by more than or equal to 2 points within 2 consecutive seconds, the control is tightened again. Figure 2 This invention provides multimodal data for an autonomous driving control method based on occupant motion sickness perception and comfort. Figure 3 This is a schematic diagram of a hierarchical improvement strategy for an autonomous driving control method based on occupant motion sickness perception and comfort, according to the present invention.

[0247] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. An autonomous driving control method based on occupant motion sickness perception and comfort, characterized in that, The steps are as follows: S1. Collect target data; S11. Determine test conditions: Clarify the test environment and conditions, including vehicle type, road conditions, and driving speed; S12. Sensor Deployment: Sensors are deployed on the vehicle and the occupants' bodies. These sensors include a triaxial accelerometer, skin conductance acquisition electrodes and a skin conductance meter, and electrocardiogram (ECG) acquisition electrodes and an ECG meter. The computer then... Interface device connects to vehicle Interface; among which, the triaxial accelerometer is used to collect vehicle dynamic performance data, and the computer... Interface device connects to vehicle Interface reading Driving behavior parameters in the bus, skin conductance acquisition electrodes and skin conductance meters, and electrocardiogram acquisition electrodes and electrocardiogram meters are used to collect physiological indicators of the human body; S13. Collect objective data: through vehicles The recorded vehicle dynamic performance data under different conditions are used to evaluate vehicle dynamic performance and its impact on comfort; the physiological index data of the occupants are recorded simultaneously to quantify the physiological response of the human body to vehicle motion. S14. Subjective Data Collection: Before the test, collect motion sickness questionnaires from occupants; after the test, collect riding experience questionnaires from occupants; during the test, collect subjective rating data from occupants simultaneously with objective data collection, and evaluate comfort based on their actual feelings. The evaluation scope is... The score is divided into 1 point, with 0.5-point intervals introduced, where 1 point is extremely uncomfortable and 10 points is very comfortable. S2, Data Processing and Feature Extraction; S21. Data Preprocessing: The vehicle dynamic performance data, occupant physiological index data, and subjective rating data collected in step S1 are used as raw data and systematically preprocessed. First, the subjective rating data is filtered by interval; cross-modal data synchronization is performed using timestamps; low-pass filtering and moving average processing are performed on the physiological index data; and sampling reconstruction and outlier removal are performed on the vehicle dynamic performance data. S22. Feature Extraction: Extract physiological features describing the motion sickness response of passengers from the collected physiological index data, including feature indicators of heart rate variability and skin conductance response; extract dynamic performance features reflecting the motion sickness stimuli induced by the vehicle from the collected vehicle dynamic performance data, including linear acceleration feature indicators, angular velocity and angular acceleration feature indicators, jerk feature indicators and omnidirectional acceleration composite indicator. S23. Feature Fusion and Standardization: The extracted physiological features and vehicle dynamic performance features are concatenated to form a unified feature vector, and the features are standardized by normalization or standard deviation scaling for subsequent modeling. S3. Construct a prediction model for occupant motion sickness perception and comfort rating: S31. Feature integrity check and redundancy control: Perform integrity checks on the vehicle dynamic performance features and occupant physiological features extracted in step S2, and then identify the high linear correlation between different features. If redundancy exists, only retain the features with more direct physical meaning and higher signal stability. S32. Modeling and Prediction: Input the vehicle's dynamic performance characteristics and the occupants' physiological characteristics into the constructed occupant motion sickness perception and comfort score prediction model based on a gradient boosting framework that builds multiple decision trees. S4. Comfort evaluation results output and intelligent control strategy: S41. Motion Sickness Sensitivity Classification: Based on the motion sickness sensitivity labels output by the occupant motion sickness perception and comfort score prediction model constructed in step S3, determine the occupant motion sickness sensitivity type. S42. Motion sickness risk prediction: Combine the future trajectory data output by the trajectory planning module of the autonomous vehicle to calculate the normalized ratio of the root mean square of lateral acceleration, jerk and omnidirectional acceleration, and use the future short-term motion sickness risk score level output by the multi-decision tree occupant motion sickness perception and comfort score prediction model as the input signal for triggering the hierarchical improvement strategy. S43. Layered improvement strategy: Select different vehicle control and cabin environment adjustment strategies based on the occupant's motion sickness sensitivity type and future short-term motion sickness risk score level. S44. Continuous monitoring and dynamic adjustment: After the vehicle control and cabin environment adjustment strategies are implemented, the comfort score is continuously monitored, the motion sickness sensitivity type and the changes in the future short-term motion sickness risk score are determined, and the vehicle control and cabin environment adjustment strategies are dynamically adjusted online according to the tiered improvement strategy. If the future short-term motion sickness risk score decreases within 5 consecutive seconds and the comfort score increases by more than or equal to 1 point, the constraints are gradually relaxed; if the future short-term motion sickness risk score remains high or decreases by more than or equal to 2 points within 2 consecutive seconds, the control is tightened again.

2. The autonomous driving control method based on occupant motion sickness perception and comfort according to claim 1, characterized in that, In step S12, Vehicle dynamic performance data is collected by a three-axis accelerometer installed under the passenger seat. The three-axis accelerometer collects the dynamic acceleration of the vehicle along the X-axis in the direction of travel, the Y-axis perpendicular to the direction of travel on the horizontal plane, and the Z-axis perpendicular to the direction of travel on the vertical plane. Physiological indicators of the occupant are collected. The skin conductance collection electrodes are fixed on the thenar eminence of both hands to measure the skin conductance signal of the occupant and are connected to the skin conductance meter through a wireless transmission device. The electrocardiogram collection electrodes are placed on the chest, left arm, and right arm to measure the electrocardiogram signal of the occupant and are connected to the electrocardiogram meter through a wireless transmission device. The electrodermatology and electrocardiogram (ECG) devices are connected to a computer via their microcomputer interfaces, transmitting the collected electrodermatology and ECG signals to the computer. The ECG acquisition electrodes are three-lead patch electrodes.

3. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 2, characterized in that, In step S21, The preprocessing method is as follows: The vehicle dynamic performance data, occupant physiological index data, and subjective rating data collected in step S1 are used as raw data and systematically preprocessed. First, the subjective rating data is filtered by interval, retaining subjective ratings between [3, 8] points; second, timestamps are used for cross-modal data synchronization, and a reference time axis is established based on a unified timestamp. Using linear interpolation for any time interval Align the data above: ； in, The set of raw sampling times belonging to triaxial accelerometer, electrocorticometry, and electrocardiogram. , These are two adjacent original sampling times, and ; For the i-th triaxial accelerometer, electrocardiogram, and electrocardiogram at their own original sampling time The original sampling data on; To establish a unified reference timeline The top-aligned data outputs from the i-th triaxial accelerometer, electrodermometer, and electrocardiogram. Physiological data were processed using low-pass filtering and moving average. First, the electrocardiogram (ECG) signal was filtered, and a band-pass filter was used to eliminate baseline drift and electromyographic interference. The frequency band was set to 0.5Hz~40Hz. It is expressed as follows: ； in, This is the spectrum of the electrocardiogram (ECG) signal; For the response function of an ideal bandpass filter: ； in, , ; Smoothing of skin electrical signals: ； Where 2N+1 is the size of the sliding window, set to 3~9 points; The original skin electrical signal at time t is the first digit before and after the current time. The values ​​at each sampling point Centered on the current moment, the signals from 2N+1 sampling points before and after the current moment are averaged to achieve smoothing of the electrodermal signal. The vehicle dynamic performance data is sampled, reconstructed, and outlier removed. First, a first-order low-pass filter is used to suppress high-frequency interference on the vehicle acceleration and jerk signals. ； in, This is the filtered output at the current moment; This refers to acceleration data within the vehicle's dynamic performance data. It is a smoothing factor, and Set it to 0.1~0.3; This is the filtered output from the previous time step; Next, outlier detection and removal are performed, using the Z-score method to check whether a point is an outlier at each time step: ； in, The mean value within the selected sliding window; This represents the standard deviation within the selected sliding window. like If so, it is determined to be an outlier and removed; Finally, resampling and frequency unification are performed, with the original data at the resampling point. The new value is obtained by linear interpolation: ； in, , The raw data consists of vehicle dynamic performance data, occupant physiological indicators, and subjective rating data. Adjacent sampling points, ; , The data values ​​corresponding to the sampling points in the raw data consisting of vehicle dynamic performance data, occupant physiological index data, and subjective rating data.

4. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 3, characterized in that, In step S22, The feature extraction method is as follows: Feature extraction is performed on the data preprocessed in step S21. First, heart rate variability is extracted, specifically the standard deviation of heart rate variability. Indicators for measuring overall heart rate variability: ； in, for The total number of intervals; For the first indivual Interval; For all within this window The average value of the interval; among which, The interval is one period within a window of an electrocardiogram. Wave to the next The time interval between waves; The root mean square (RMSSD) of adjacent heart rate variability was extracted as a measure of short-term changes in parasympathetic activity. ； Extract the LF / HF ratio, a frequency domain feature of heart rate variability, calculate the frequency domain energy distribution of heart rate variability to represent autonomic balance, and use Fast Fourier Transform to calculate the power spectrum. ； in, This refers to the spectral power at low frequencies. This refers to the spectral power at high frequencies. Secondly, the characteristics of the skin conductance response were extracted, and the average conductivity of the skin conductance response was extracted to measure the overall tension level. ； in, This is the original skin conductance signal; For the observation window length; The frequency of skin conductance responses was extracted, and the peak value was detected using a threshold method. The number of responses per second was defined as: ； in, for Number of skin conductance responses per second; Maximum amplitude of skin conductance response: ； in, Peak value; The baseline of conductance before rise; Feature extraction is performed on vehicle dynamic performance data. Linear acceleration features include mean, standard deviation, and peak acceleration; angular velocity and angular acceleration features include angular acceleration and yaw frequency; jerk features include jerk and root mean square jerk; and omnidirectional acceleration composite features include total acceleration.

5. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 4, characterized in that, In step S23, The method for feature fusion and standardization is as follows: First, multimodal feature fusion is performed, concatenating feature vectors from different sources along their feature dimensions to form a unified high-dimensional input vector. Let: Let m be the features extracted from the electrocardiogram signal; Let n be the features extracted from the electrodermal signal; Let r be the r features extracted from vehicle acceleration; The fused feature vector is then: ； Among them, symbols This indicates that the data is concatenated according to the feature dimension. In step S21, all features have been processed using Z-score normalization before fusion, as shown in the following formula: ； in, This is the a-th extracted feature value; Let a be the mean of the a-th extracted feature across all samples; Let a be the standard deviation of the extracted feature a across all samples; These are the standardized feature values, with a mean of 0 and a standard deviation of 1.

6. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 5, characterized in that, In step S31, The method for feature integrity checking and redundancy control is as follows: First, the integrity of the collected vehicle dynamic features and physiological features is checked, and features with a missing rate of more than 20% or variance close to zero are removed. The Pearson correlation coefficient is used to identify high linear correlations between different features. The Pearson correlation coefficient between any two features is defined as follows: ； in, , Let p-th and q-th fused features be the sequence of values ​​among all features; , This represents the value of the corresponding feature in the k-th sample; , denoted as the sample mean of the feature; u is the sample size; The correlation coefficient represents the degree of linear correlation between two features. ; when When two features are considered to be highly correlated or redundant, the system retains only the feature with the more direct physical meaning and higher signal stability.

7. The autonomous driving control method based on occupant motion sickness perception and comfort according to claim 6, characterized in that, In step S32, Motion sickness sensitivity levels are identified using the LightGBM classification model. Multiple decision trees are constructed based on a gradient boosting framework, and the prediction function is expressed as follows: ； in, To predict motion sickness sensitivity labels; x is a sample feature vector that integrates physiological and vehicle dynamics features; Z represents the set of classification labels: {low sensitivity, medium sensitivity, high sensitivity}; Z is the number of base learner decision trees in the LightGBM classification model. Output the score of the z-th tree in category c; The LightGBM classification model uses the cross-entropy loss function for classification problems, and the loss is defined as follows: ； in, To determine whether sample v belongs to category c, and ; This represents the predicted probability of sample v belonging to class c by the LightGBM classification model. Training efficiency is improved by constructing gradient histograms and using a leaf-based tree growth strategy; for each round of newly added tree models... The optimization objective is: ； in, The learning rate controls the stepping speed of the LightGBM classification model; , is the prediction score of the LightGBM classification model for the input sample x in the w-th iteration, and is the cumulative output within the model; Motion sickness score prediction was performed using an XGBoost regression model. The XGBoost regression model uses multiple CART trees in an additive structure to stack the predicted values. ； in, To predict motion sickness scores, the domain is defined as [1, 10]; S is the number of base learner regression trees in the XGBoost regression model; This represents the prediction result of the s-th regression tree; The set of all possible regression tree functions; Using mean squared error as the regression loss function: ； in, This represents the true motion sickness score for the g-th sample. These are the predicted values ​​from the XGBoost regression model; The objective function of the XGBoost regression model is: ； in, Let be the objective function of the XGBoost regression model in the t-th iteration, which includes a prediction error term and a model complexity regularization term. In the prediction error term, For real labels, This is the value predicted in the previous round. Output the current tree. This is the loss function used to calculate the error between the predicted value and the true value; The model complexity regularization term is: ； Where T is the current number of leaf nodes; The weights are those of the leaf nodes; These are the regularization parameters for the XGBoost regression model.

8. The automatic driving control method based on occupant motion sickness perception and comfort as described in claim 7. Its features are, In step S42, The method for predicting motion sickness risk is as follows: First, the trajectory planning module of the autonomous vehicle is invoked to obtain candidate trajectory data within a time window of 3-8 seconds in the future, and representative vehicle kinematic features, including lateral acceleration, are extracted. accelerometer and root mean square of omnidirectional acceleration The corresponding individualization thresholds are respectively , , Its value is derived from the vehicle kinematics characteristics corresponding to when the subjective score drops to 6 points. This threshold is dynamically updated during operation based on the output of the XGBoost regression model. To unify the scale, the feature normalization ratio is calculated first. ； The system simultaneously reads the current comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3. And define discomfort as: ； in, , The larger the value, the more likely the occupant is already showing signs of discomfort. The future short-term motion sickness risk score is defined as: ； in, This represents a score indicating the risk of short-term motion sickness in the future; a higher value indicates a higher risk. , , , These are the weighting coefficients, and ; When omnidirectional acceleration features are not used , , When it contains omnidirectional acceleration characteristics, , ; When R < 0.6, it is considered low risk, indicating that the probability of discomfort caused by the future trajectory is low; when 0.6 ≤ R < 1.0, it is considered medium risk, indicating that mild discomfort may occur in the future; when R ≥ 1.0, it is considered high risk, indicating that a significant motion sickness reaction is very likely to be induced in the next few seconds.

9. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 8, characterized in that, In step S43, The method of the stratified improvement strategy is as follows: Based on the sensitivity label and comfort score output by the occupant motion sickness perception and comfort score prediction model in step S3, determine the current susceptibility level of the occupant, assess the motion sickness risk by combining the future short-term motion sickness risk score, and implement the corresponding vehicle and environmental control strategies according to the two-dimensional stratification rules. The vertical basic stratification is as follows: for low-sensitivity occupants, the vehicle control constraints are relatively loose, with main voice prompts and mild smoothing; for moderately sensitive occupants, the vehicle is moderately constrained, and environmental adjustments for wind, temperature, and light are increased; for highly sensitive occupants, the vehicle strictly constrains jerk and lateral acceleration, and multimodal measures are used to alleviate motion sickness. Lateral motion sickness risk stratification is as follows: for low-risk cases with R < 0.6, the vehicle maintains the current strategy and continues to monitor; for medium-risk cases with 0.6 ≤ R < 1.0, the vehicle implements a mild smoothing or environmental pre-adjustment strategy in advance; for high-risk cases with R ≥ 1.0, the vehicle immediately activates the corresponding level of conservative driving mode and multimodal intervention. When the comfort score is less than or equal to 3, or when a significant physiological trigger is detected, the root mean square difference between adjacent heart rate variability decreases by more than or equal to 20%, or the frequency or amplitude of skin conductance responses increases by more than or equal to 50%, the high-risk strategy is directly implemented regardless of the future short-term motion sickness risk score R level; otherwise, the control level is determined by the combination of sensitivity label and risk score.

10. The automatic driving control method based on occupant motion sickness perception and comfort according to claim 9, characterized in that, In step S43, The combined decision matrix of sensitivity labels and risk scores is as follows: Low sensitivity + low risk: Maintain original driving style; vehicle only provides voice prompts; Low sensitivity + medium risk: jerk≤0.8m / s³, a≤1.2m / s²; vehicle temperature reduced by 1-3℃ or ventilation turned on; Low sensitivity + high risk: jerk≤0.6m / s³, a≤1.0m / s²; Vehicle HUD directional strips provide advance warning; Medium sensitivity + low risk: jerk ≤ 0.8 m / s³; vehicle monitoring of physiological fluctuations; Medium sensitivity + medium risk: jerk ≤ 0.6 m / s³; vehicle temperature reduced by 1-3℃, maintain ventilation; Medium-sensitive + high-risk: jerk≤0.5m / s³; vehicle HUD guide strip advance warning and voice prompts; High sensitivity + low risk: jerk≤0.6m / s³; soft light and constant temperature inside the vehicle; High sensitivity + medium risk: jerk≤0.5m / s³; vehicle temperature reduction of 1-3℃, maintaining ventilation, and seat vibration warning; High sensitivity + high risk: jerk≤0.4m / s³; conservative driving, HUD guide strips provide advance warnings, voice prompts and gentle cooling, and maintain ventilation.

Images