Driver takeover performance regulation and control method based on trust state

Abstract

The invention relates to a driver takeover performance regulation and control method based on a trust state, and belongs to the technical field of intelligent driving. The method comprises the steps that S1, takeover performance classification is carried out, S11, qualitative analysis is carried out on takeover data, specifically, the performance of a driver in the takeover process is evaluated from the four indexes of takeover time, minimum collision time, collision probability and braking application; s12, determining the takeover performance of the driver, specifically, calculating the performance score of each takeover link by adopting a grey relational degree calculation method and taking the four indexes as input; and S2: takeover performance prediction: on the basis of real-time trust measurement, effectively predicting the takeover performance of the driver under different trust and physiological levels by using a machine learning algorithm. According to the invention, the takeover performance can be accurately predicted for timely performance regulation and control so as to obtain the optimal takeover performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent driving and relates to a driver takeover performance control method based on trust status. Background Art

[0002] At present, my country's automobile industry is in the transition stage from L2 to L3 autonomous driving levels. The autonomous driving systems of some vehicles have reached L4 level under specific scenarios, demonstrating potential technological breakthroughs and application prospects.

[0003] However, with the popularization of autonomous driving technology, the process of the driver switching between autonomous vehicles and manual driving, that is, the takeover process, has become a matter of great concern. This process is actually the driver's transition from a passive observation state or a semi-active participation state to an active driving state, involving the driver's restoration of awareness of the driving environment and re-taking control of the vehicle control. The driver's takeover performance, that is, the ability and efficiency demonstrated in this process, directly reflects the safety and effectiveness of the human-machine collaborative system. Good takeover performance means that the driver can quickly and accurately regain control of the vehicle, thereby reducing the risk of traffic accidents.

[0004] In recent years, many scholars have conducted research on autonomous driving takeover behavior. Takeover request time is an important factor affecting takeover behavior, usually defined as the collision time between the vehicle and the obstacle ahead when the takeover request warning prompt signal of the autonomous driving system is issued. Takeover time is one of the important takeover performance indicators, usually defined as the time interval from the takeover request time to the driver turning off the autonomous driving mode. The shutdown method of autonomous driving is mainly achieved through buttons on the steering wheel, braking and steering (setting the brake pedal or steering wheel angle to exceed a certain threshold).

[0005] There are many factors that affect the driver's takeover performance, among which the level of trust between humans and machines is considered to be an important factor. The driver's trust in the autonomous driving system will directly affect its performance during the takeover process. Therefore, in-depth research on the relationship between driver trust and takeover performance and accurate prediction of the driver's takeover performance at different trust levels are of great significance for improving the safety and reliability of the autonomous driving system. Summary of the invention

[0006] In view of this, an object of the present invention is to provide a driver takeover performance control method based on trust state, which accurately predicts the takeover performance and performs timely performance control to obtain the best takeover performance.

[0007] A driver takeover performance control method based on trust state specifically includes the following steps:

[0008] S1: Takeover performance classification;

[0009] S11: Qualitative analysis of takeover data, specifically including: evaluating the driver's performance during the takeover process based on four indicators: takeover time, minimum collision time, collision probability and brake application;

[0010] S12: clarify the driver's takeover performance, specifically including: using the grey correlation calculation method, taking the above four indicators as input, and calculating the performance score of each takeover link;

[0011] S2: Takeover performance forecast;

[0012] Based on real-time measurement of trust, a machine learning algorithm is used to effectively predict the driver's takeover performance under different trust and physiological levels.

[0013] Further, in step S11, the Take-Over Time (TOT) is a measure of the time from the start of takeover to the actual control of the vehicle, which reflects the driver's reaction speed to the system limit and is also an important indicator for evaluating the driver's reaction ability; the Minimum Collision Time (TTC) is used to measure the shortest time required for the vehicle to collide during the takeover process, thereby evaluating the quality of the takeover; especially when approaching the system limit, TTC will gradually decrease after the takeover is started until the driver changes lanes or significantly slows down. The collision probability is a measure of the probability of a collision within the takeover range, which plays an important role in evaluating the risk of the takeover process; brake application is used to indicate whether the driver has used the brakes. There is a significant difference between drivers who apply the brakes in the takeover situation and those who do not.

[0014] Further, step S12 specifically includes the following steps:

[0015] S121: For each indicator, calculate the grey correlation degree between it and other indicators;

[0016] First, the threshold method is used to make the indicator dimensionless and eliminate the dimension effect:

[0017]

[0018] Among them, y i is the evaluation value after index conversion, x i is the actual value, x0 is the threshold of the indicator; then calculate the grey correlation coefficient ξ i (k), using the formula:

[0019]

[0020] Δ i (k)=|m(k)-n i(k)|

[0021] Where k is the kth value in the i-th column, m(k) is the kth value in the reference sequence, and n i (k) is the kth value in the i-th column of data, Δ i (k) is m(k) and n i (k), which indicates the difference between the reference sequence and the actual sequence at the kth position; ρ is the resolution coefficient, which takes a value of 0<ρ<1;

[0022] Then calculate the correlation r i , the calculation formula is as follows:

[0023]

[0024] Where n is the total number of values ​​in the i-th column;

[0025] S122: Determine the relative importance of each indicator to the driving takeover performance according to the calculated grey correlation degree;

[0026] The weight is determined according to the size of the grey correlation degree. The greater the correlation degree, the higher the weight of the indicator:

[0027]

[0028] Among them, w i is the weight of the ith indicator, m is the number of indicators, x ii is the average correlation between the ith indicator and other indicators;

[0029] S123: Based on the determined weights, the performance of each indicator is comprehensively considered to obtain an evaluation result of the driving takeover performance;

[0030] Using the weighted comprehensive scoring method, the value of each indicator is multiplied by the corresponding weight, and then the sum is obtained to obtain the comprehensive score:

[0031]

[0032] Among them, P is the comprehensive score, w i is the weight of the ith indicator, x i is the value of the ith indicator;

[0033] S124: The cumulative frequency distribution method is used to quantitatively grade the takeover segment performance value, and the corresponding index values ​​of 80% and 50% of the cumulative frequencies are used as thresholds to divide the driver's takeover performance into three categories: good, average and poor.

[0034] Further, step S2 specifically includes the following steps:

[0035] S21: Trust and physiological signal data processing;

[0036] Using the driver's trust status, skin electrical signals and heart rate signals as input and the takeover performance level as output, a takeover performance prediction model is constructed; trust characteristics include high trust, moderate trust and lack of trust;

[0037] S22: The random forest algorithm is used as the classification model of the driving performance prediction model; 10-fold cross validation is used to evaluate the performance of the model, the grid search algorithm is used to adjust the parameters, and the optimal model after the grid search is saved; the gini coefficient is selected as the evaluation indicator, and the remaining parameters are selected as the default settings.

[0038] Further, in step S21, for the skin electrical signal, the skin electrical signal is first preprocessed as necessary, and then the preprocessed skin electrical signal is constructed into a mixed matrix, in which each row represents an observation time point, and each column represents a measurement of a skin electrical signal; the specific steps are: dividing the skin electrical signal into N time windows, and the signal in each time window is considered to be an independent observation; for each time window, the corresponding skin electrical signal data is put into the mixed matrix as a row; using the fastICA algorithm to decompose the mixed matrix, the skin electrical signal is decomposed into a phase component and a tension component; according to the results of fastICA, a suitable principal component is selected as the phase component, and then the phase component is used for further feature extraction.

[0039] Further, in step S21, for the heart rate signal, the RR variability signal (RRV) is first extracted from the ECG, and the RR interval is the time difference between consecutive R peaks. Then, the collected RR interval data is preprocessed as necessary to remove abnormal values; the RR interval data is converted into heart rate data, usually by calculating the heart rate by calculating the number of beats per minute (BPM):

[0040]

[0041] Among them, HR represents heart rate (heart beats per minute), RR represents RR interval; finally, the heart rate variability RRSSD is calculated using the following formula:

[0042]

[0043] Among them, RR i is the i-th RR interval, K represents the total number of RR intervals, and since each calculation is based on the difference between two adjacent RR intervals, the differences of K-1 adjacent intervals are calculated.

[0044] Further, in step S22, random forest randomly selects m predictor variables from p predictor variables based on the bootstrap sampling method, and finds the best partition among them, dividing a node into two child nodes, and at the same time votes on the constructed tree to select the category with the most votes; the classification decision formula is:

[0045]

[0046] in, represents a combined classification model; is the prediction of the jth tree for the response variable x; y represents the output variable; I(·) is the indicator function.

[0047] The beneficial effect of the present invention is that during the human-machine co-driving stage, the driver is no longer required to actively monitor the driving environment. However, as a backup for automation, when the automation system reaches its operating limit, the driver needs to take over control of the vehicle. The present invention will clarify the driver's takeover performance through qualitative analysis of takeover data, and use machine learning algorithms to effectively predict the driver's takeover performance under different trust and physiological levels based on real-time measurement of trust, and accurately predict the takeover performance to conduct timely performance regulation to obtain the best takeover performance.

[0048] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0050] Figure 1 The present invention is a flow chart of the driver takeover performance control method based on the trust state. DETAILED DESCRIPTION

[0051] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0052] See also Figure 1 The present invention provides a driver takeover performance control method based on trust status, which clarifies the driver's takeover performance through qualitative analysis of takeover data, and uses a machine learning algorithm to effectively predict the driver's takeover performance under different trust and physiological levels based on real-time measurement of trust. The method is divided into two modules, a takeover performance classification module and a takeover performance prediction module.

[0053] 1. Take over the performance classification module

[0054] The driver's performance during the takeover process is evaluated based on four indicators: takeover time, minimum collision time, brake application, and collision probability. First, the takeover time (TOT) measures the time from the start of takeover to the actual control of the vehicle. This reflects the driver's reaction speed to the system limit and is also an important indicator for evaluating the driver's reaction ability. Secondly, the minimum collision time (TTC) is used to measure the shortest time required for the vehicle to collide during the takeover process, thereby evaluating the quality of the takeover. Especially when approaching the system limit, TTC will gradually decrease after the takeover is started until the driver changes lanes or slows down significantly. The collision probability measures the probability of a collision within the takeover range and plays an important role in evaluating the risk of the takeover process. Finally, brake application indicates whether the driver used the brakes. There is a significant difference between drivers who applied the brakes and those who did not apply them in the takeover situation.

[0055] In order to quantitatively evaluate the takeover performance, the grey correlation calculation method is used, taking the above four indicators as input to calculate the performance score of each takeover link. First, for each indicator, the grey correlation between it and other indicators is calculated. This method takes into account the degree of mutual correlation between indicators, so as to reflect the degree of influence of indicators on driving takeover performance.

[0056] Different indicators are not comparable due to different dimensions. First, the threshold method is used to make the indicators dimensionless to eliminate the dimension effect:

[0057]

[0058] Among them, y i is the evaluation value after index conversion, x i is the actual value, and x0 is the threshold value of the indicator. Then calculate the grey correlation coefficient using the formula:

[0059]

[0060] Δ i (k)=|y(k)-x i (k)|

[0061] Among them, x i is the data in the ith column, k is the kth value in the ith column, ρ is the resolution coefficient, which takes a value of 0<ρ<1, and is 0.5 here. Then calculate the correlation:

[0062]

[0063] According to the calculated grey correlation degree, the relative importance of each indicator to the driving takeover performance is determined. The weight can be determined according to the size of the grey correlation degree. The greater the correlation degree, the higher the weight of the indicator.

[0064]

[0065] Among them, w i is the weight of the ith indicator, m is the number of indicators, x ii is the average correlation between the ith indicator and other indicators. According to the determined weights, the performance of each indicator is comprehensively considered to obtain the evaluation result of the driving takeover performance. Using the weighted comprehensive scoring method, the value of each indicator is multiplied by the corresponding weight, and then the sum is obtained to obtain the comprehensive score:

[0066]

[0067] Among them, P is the comprehensive score, w i is the weight of the ith indicator, x i is the value of the ith indicator. Finally, the cumulative frequency distribution method is used to quantify the takeover performance value. The corresponding indicator values ​​of 80% and 50% cumulative frequencies are used as thresholds to divide the driver's takeover performance into three categories: good, average, and poor.

[0068] 2. Take over performance prediction module

[0069] 1) Trust and physiological signal data processing

[0070] This method uses the driver's trust status, skin electrical signals and heart rate signals as inputs and takes the takeover performance level as output to build a takeover performance prediction model. The trust characteristics are presented in three results: high trust, moderate trust and lack of trust.

[0071] For the electrical skin signal, the necessary preprocessing is first performed on the electrical skin signal, such as noise removal and filtering, to ensure the quality and accuracy of the signal. The preprocessed electrical skin signal is then constructed into a mixed matrix, in which each row represents an observation time point and each column represents a measurement of the electrical skin signal. The specific steps are to divide the electrical skin signal into N time windows, and the signal in each time window is considered to be an independent observation. For each time window, the corresponding electrical skin signal data is put into the mixed matrix as a row. Let x ij(x_{ij}) is the element of the i-th row (time window) and j-th column (skin electrical signal channel) in the mixing matrix X, then:

[0072]

[0073] The fastICA algorithm is used to decompose the mixed matrix and decompose the skin electrical signal into phase component and tension component. According to the results of fastICA, the appropriate principal component is selected as the phase component, and then the phase component is used for further feature extraction.

[0074] For the heart rate signal, first extract the RR variability signal (RRV) from the ECG. The RR interval is the time difference between R consecutive peaks. Then, perform necessary preprocessing on the collected RR interval data to remove abnormal values. The RR interval data is converted into heart rate data, usually by calculating the heart rate by taking the number of beats per minute (BPM):

[0075]

[0076] Among them, HR represents heart rate (heart beats per minute), and RR represents RR interval. Finally, the following formula is used to calculate heart rate variability:

[0077]

[0078] 2) Machine Learning Algorithms

[0079] This method uses the random forest algorithm as the classification model of the driving performance prediction model. The core idea of ​​the random forest algorithm is "ensemble learning", which is to integrate decision trees with a fixed number and different features to make it have higher prediction performance and robustness. The advantage of the random forest algorithm is that it has strong adaptability to the disturbance of the data set and has high robustness to bad facts in many ways. In addition, it is also an algorithm that can handle high-dimensional features because the random forest algorithm can adaptively reduce dependence on unimportant features. Based on the self-service sampling method, the random forest randomly selects m predictor variables from p predictor variables, and finds the best partition from them, dividing a node into two sub-nodes, and voting on the constructed tree at the same time to select the category with the most votes. The classification decision formula is:

[0080]

[0081] in, represents a combined classification model; is the prediction of the jth tree for the response variable x; y represents the output variable; I() is the indicative function.

[0082] A 10-fold cross validation was used to evaluate the performance of the model, and a grid search algorithm was used to adjust the parameters. The optimal model after the grid search was saved. The gini coefficient was selected as the evaluation indicator, and the rest of the parameters were set to default.

[0083] 3) Multi-classification evaluation

[0084] The accuracy, precision, recall and F1 value of each category are used, and then the macro-average and micro-average of the above indicators are calculated to evaluate the multi-classification model.

[0085]

[0086]

[0087] Among them, TP means predicting a positive sample as a positive sample, that is, the prediction is correct; FP means predicting a negative sample as a positive sample, that is, the prediction is wrong; FN means predicting a positive sample as a negative sample, that is, the prediction is wrong; TN means predicting a negative sample as a negative sample, that is, the prediction is correct.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A driver takeover performance control method based on trust status, characterized in that: The method specifically comprises the following steps: S1: Takeover performance classification; S11: Qualitative analysis of takeover data, specifically including: evaluating the driver's performance during the takeover process based on four indicators: takeover time, minimum collision time, collision probability and brake application; S12: clarify the driver's takeover performance, specifically including: using the grey correlation calculation method, taking the above four indicators as input, and calculating the performance score of each takeover link; S2: Takeover performance forecast; Based on real-time measurement of trust, a machine learning algorithm is used to predict the driver's takeover performance under different trust and physiological levels.

2. The driver takeover performance control method according to claim 1, characterized in that: In step S11, the takeover time is a measure of the time from the start of takeover to the actual control of the vehicle; the minimum collision time is used to measure the shortest time required for the vehicle to collide during the takeover process, thereby evaluating the quality of takeover; the collision probability is a measure of the probability of a collision within the takeover range; and the brake application is used to indicate whether the driver has used the brakes.

3. The driver takeover performance control method according to claim 1, characterized in that: Step S12 specifically includes the following steps: S121: For each indicator, calculate the grey correlation degree between it and other indicators; First, the threshold method is used to make the indicator dimensionless and eliminate the dimension effect: Among them, y i is the evaluation value after index conversion, x i is the actual value, x0 is the threshold of the indicator; then calculate the grey correlation coefficient ξ i (k), using the formula: △ i (k)=|m(k)-n i (k)| Where k is the kth value in the i-th column, m(k) is the kth value in the reference sequence, and n i (k) is the kth value in the i-th column of data, Δ i (k) is m(k) and n i (k), which indicates the difference between the reference sequence and the actual sequence at the kth position; ρ is the resolution coefficient, which takes a value of 0<ρ<1; Then calculate the correlation r i , the calculation formula is as follows: Where n is the total number of values ​​in the i-th column; S122: Determine the relative importance of each indicator to the driving takeover performance according to the calculated grey correlation degree; The weight is determined according to the size of the grey correlation degree. The greater the correlation degree, the higher the weight of the indicator: Among them, w i is the weight of the ith indicator, m is the number of indicators, x ii is the average correlation between the ith indicator and other indicators; S123: Based on the determined weights, the performance of each indicator is comprehensively considered to obtain an evaluation result of the driving takeover performance; Using the weighted comprehensive scoring method, the value of each indicator is multiplied by the corresponding weight, and then the sum is obtained to obtain the comprehensive score: Among them, P is the comprehensive score, w i is the weight of the ith indicator, x i is the value of the ith indicator; S124: The cumulative frequency distribution method is used to quantitatively grade the takeover segment performance value, and the corresponding index values ​​of 80% and 50% of the cumulative frequencies are used as thresholds to divide the driver's takeover performance into three categories: good, average and poor.

4. The driver takeover performance control method according to claim 3, characterized in that: Step S2 specifically includes the following steps: S21: Trust and physiological signal data processing; Using the driver's trust status, skin electrical signals and heart rate signals as input and the takeover performance level as output, a takeover performance prediction model is constructed; trust characteristics include high trust, moderate trust and lack of trust; S22: The random forest algorithm is used as the classification model of the driving performance prediction model; 10-fold cross validation is used to evaluate the performance of the model, the grid search algorithm is used to adjust the parameters, and the optimal model after the grid search is saved; the gini coefficient is selected as the evaluation indicator, and the remaining parameters are selected as the default settings.

5. The driver takeover performance control method according to claim 4, characterized in that: In step S21, for the skin electrical signal, the skin electrical signal is first preprocessed, and then the preprocessed skin electrical signal is constructed into a mixed matrix, in which each row represents an observation time point, and each column represents a measurement of a skin electrical signal; the specific steps are: dividing the skin electrical signal into N time windows, and the signal in each time window is considered to be an independent observation; for each time window, the corresponding skin electrical signal data is put into the mixed matrix as a row; using the fastICA algorithm to decompose the mixed matrix, the skin electrical signal is decomposed into a phase component and a tension component; according to the results of fastICA, a suitable principal component is selected as the phase component, and then the phase component is used for further feature extraction.

6. The driver takeover performance control method according to claim 4, characterized in that: In step S21, for the heart rate signal, the RR variability signal is first extracted from the ECG, where the RR interval is the time difference between R consecutive peaks, and then the collected RR interval data is preprocessed to remove abnormal values; the RR interval data is converted into heart rate data, and the heart rate is calculated by calculating the number of heart beats per minute: Among them, HR represents heart rate and RR represents RR interval; finally, the heart rate variability RRSSD is calculated using the following formula: Among them, RR i is the i-th RR interval, and K represents the total number of RR intervals.

7. The driver takeover performance control method according to claim 4, characterized in that: In step S22, random forest randomly selects m predictor variables from p predictor variables based on the bootstrap sampling method, and finds the best partition among them, dividing a node into two child nodes, and at the same time votes on the constructed tree to select the category with the most votes; the classification decision formula is: in, represents a combined classification model; is the prediction of the jth tree for the response variable x; y represents the output variable; I(·) is the indicator function.

Images