Automatic driving takeover performance evaluation method and system considering multiple influence factors

Abstract

The invention discloses an automatic driving takeover performance evaluation method and system considering multiple influence factors, and the method comprises the steps: obtaining multi-modal data and influence factor data in the whole process of switching a working mode of an automatic driving system in different traffic takeover scenes through constructing a driving data collection assembly, and obtaining the performance of the automatic driving takeover performance through a subjective weighting method and an objective weighting method. Calculating the optimal weight of each related characteristic index in the multi-modal data, constructing an automatic driving takeover performance comprehensive evaluation model to obtain the comprehensive score, and constructing a generalized linear hybrid model based on the comprehensive score and the influence factor data to obtain the comprehensive evaluation result of the automatic driving takeover performance. Evaluating the influence importance of different influence factors in the influence factor data on the comprehensive score of the takeover performance; according to the method, the multi-dimensional takeover performance comprehensive evaluation model based on the DEA model is constructed by acquiring the multi-modal data and extracting the characteristic indexes of the multi-modal data, so that the problem of single-dimensional limitation of an existing evaluation system is solved, and the actual score of the automatic driving takeover performance is accurately reflected.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving tests, and in particular, to a method and system for evaluating the takeover performance of autonomous driving considering multiple influencing factors. Background Art

[0002] With the rapid development of autonomous driving technology, the human-machine co-driving mode has become a key part of future intelligent transportation systems; in the human-machine co-driving mode, the driver and the autonomous driving system jointly participate in driving tasks. When the system encounters complex situations that it cannot handle, the driver needs to take over the vehicle control to ensure safety; however, the current evaluation methods for autonomous driving takeover performance have many defects and are difficult to comprehensively and accurately measure the performance of the autonomous driving system during the takeover process.

[0003] The takeover performance of the driver is a complex concept covering multiple levels, involving multiple dimensions such as the safety, comfort, and smoothness of the takeover. Therefore, it is crucial to evaluate the takeover performance reasonably and comprehensively; most of the existing technologies focus on the assessment of vehicle risks before the driver takes over, while there is little research on the evaluation of takeover performance during the takeover process. In the existing autonomous driving takeover performance evaluation methods, they often only focus on a single original takeover performance evaluation index, with the limitation of insufficient evaluation dimensions. For example, they only judge based on driving parameters such as vehicle speed, acceleration, and steering angle recorded by vehicle sensors. However, these data can only reflect the vehicle operation level and cannot deeply understand the actual state and subjective experience of the driver during the takeover, ignoring the importance of the driver as a key subject in human-machine co-driving and unable to comprehensively measure the overall quality of the takeover process.

[0004] At the same time, there is currently a lack of effective means to deeply analyze the many influencing factors of autonomous driving takeover performance, and it is difficult to quantify the influence degree of each influencing factor on the takeover performance score. Therefore, although it is known that these factors will affect the takeover performance, it is not clear to what extent each factor affects the final effect of the takeover and how different factors interact with each other. This makes it difficult for autonomous driving system developers to optimize the system functions targeted. Summary of the Invention

[0005] In view of the above deficiencies of the prior art, the present invention provides a method and system for evaluating the takeover performance of autonomous driving considering multiple influencing factors to solve the problems of single evaluation dimension of takeover performance, insufficient analysis of influencing factors of takeover performance, and less attention to the takeover process in the prior art.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a method for evaluating the takeover performance of autonomous driving considering multiple influencing factors, including the following steps:

[0008] S1: Construct a driving data acquisition component and set up various traffic takeover scenarios;

[0009] S2: Based on the driving data acquisition component, obtain multi-modal data and influencing factor data during the whole process of the automatic driving system switching working modes under different traffic takeover scenarios;

[0010] The multi-modal data includes vehicle driving data, driver physiological data, and subjective evaluation data;

[0011] The influencing factor data includes driver inherent influencing factor data, environmental influencing factor data, and situational influencing factor data;

[0012] S3: Preprocess the multi-modal data and extract multiple relevant feature indicators from the preprocessed multi-modal data;

[0013] S4: Configure weights for each relevant feature indicator through subjective weighting method and objective weighting method, and calculate the optimal weights of each relevant feature indicator;

[0014] S5: Construct a comprehensive evaluation model for automatic driving takeover performance based on the DEA model, and obtain the comprehensive score of automatic driving takeover performance through each relevant feature indicator and its optimal weight;

[0015] S6: Based on the comprehensive takeover performance score and influencing factor data, construct a generalized linear mixed model to evaluate the influence importance of different influencing factors in the influencing factor data on the comprehensive takeover performance score.

[0016] Optionally, it further includes S7: Feed back the relevant feature indicators of the multi-modal data, the comprehensive score of automatic driving takeover performance, and the influence importance to the entity related to automatic driving optimization.

[0017] Optionally, the S1 specifically includes the following steps:

[0018] S101: Construct a driving data acquisition component;

[0019] S102: By importing OpenStreetMap data into the driving simulation software in the driving data acquisition component, use the actual map information to set up various traffic takeover scenarios;

[0020] S103: Through the driving simulation software, adjust the environmental parameters and traffic flow conditions in the traffic takeover scenarios to make the traffic takeover scenarios more realistic;

[0021] S104: Set different trigger point positions in various traffic takeover scenario maps.

[0022] Optionally, the driving data acquisition component includes a physiological data acquisition unit, a subjective evaluation acquisition unit, a driver information acquisition unit, a simulated cockpit, and multiple electronic display devices;

[0023] The simulated cockpit is provided with a cockpit controller for simulating driving and is deployed with driving simulation software; the driving simulation software is used to collect vehicle driving data and environmental impact factor data; the physiological data acquisition unit is used to collect driver physiological data and situational impact factor data; the subjective evaluation acquisition unit is used to collect the subjective evaluation data of the driver in different traffic takeover scenarios; the driver information acquisition unit is used to collect driver inherent impact factor data.

[0024] Optionally, S2 specifically includes the following steps:

[0025] S201: Run the data acquisition component to obtain driver inherent impact factor data;

[0026] The driver inherent impact factor data includes driver age, driver gender, and driving experience;

[0027] S202: When the vehicle's autonomous driving reaches the takeover trigger point position of the traffic takeover scenario, the autonomous driving system issues a manual driving takeover request and issues a warning to prompt the driver to take over, so as to realize the switching of the autonomous driving system from the autonomous driving mode to the manual driving mode;

[0028] S203: When the driver determines that the current road environment meets the autonomous driving conditions, trigger the autonomous driving takeover to realize the switching of the autonomous driving system from the manual driving mode to the autonomous driving mode;

[0029] S204: Starting from the moment when the manual driving takeover request is triggered, synchronously obtain the vehicle driving data, driver physiological data, environmental impact factor data, and situational impact factor data during the whole process of the vehicle's autonomous driving system switching working modes;

[0030] The environmental impact factor data includes weather conditions and road types; the situational impact factor data includes the driver's distraction situation;

[0031] S205: After each traffic takeover scenario simulation ends, obtain the driver's subjective evaluation data.

[0032] Optionally, S3 specifically includes the following steps:

[0033] S301: Perform data preprocessing on the vehicle driving data and driver physiological data;

[0034] The data preprocessing includes denoising processing, outlier removal processing, normalization processing, and filling in missing values;

[0035] S302: Align and associate the subjective evaluation data, preprocessed vehicle driving data, and preprocessed driver physiological data according to the time stamps to construct a standardized data set;

[0036] S303: Extract relevant characteristic indicators of vehicle driving data, driver physiological data, and subjective evaluation data.

[0037] Optionally, step S5 specifically includes the following steps:

[0038] S501: Based on the DEA method, construct a comprehensive evaluation model for autonomous driving takeover performance, and use autonomous driving takeover events under different takeover scenarios as different decision-making units in the comprehensive evaluation model;

[0039] S502: Use each relevant characteristic indicator obtained in each takeover scenario and the optimal weight of the relevant characteristic indicator as the input variables of the corresponding decision-making unit;

[0040] S503: Based on the input-oriented CCR model with constant returns to scale, calculate the relative efficiency value of each decision-making unit to obtain the comprehensive score of the takeover performance of different takeover events;

[0041] The CCR model includes an objective function and constraint conditions;

[0042] The expression of the objective function is:

[0043]

[0044] In the formula, P is the comprehensive score of autonomous driving takeover performance; m is the total number of characteristic indicators; n is the serial number of the characteristic indicator; The optimal weight of the nth characteristic indicator; is the performance value of the nth characteristic indicator;

[0045] The expression of the constraint condition is:

[0046]

[0047] In the formula, s.t. is the constraint condition; W is the optimal weight vector; T is the transpose symbol.

[0048] Optionally, for the generalized linear mixed model in S6, its functional expression is:

[0049] G1 + G2 + G3

[0050] +

[0051]

[0052]

[0053] Wherein, is the comprehensive performance evaluation score; i is the takeover times serial number; j is the driver serial number; k is the driving scenario serial number; is the fixed effect; is the driver random effect; is the scenario random effect; is the random error term; is the fixed effect intercept; M is the total number of influencing factors; p and q are different influencing factors, where p, q = 1, 2, …, M; is the fixed effect coefficient of the p-th influencing factor; is the interaction effect coefficient between the p-th influencing factor and the q-th influencing factor; is the value of the p-th influencing factor at the i-th takeover for the j-th driver in the k-th driving scenario; is the random intercept of the j-th driver; is the random slope of the j-th driver for the p-th influencing factor; is the random intercept of the k-th scenario; is the random slope for the p-th influencing factor in the k-th scenario.

[0054] In a second aspect, the present invention provides an automatic driving takeover performance evaluation system considering multiple influencing factors, which is characterized by including:

[0055] A scenario construction module for constructing a driving data acquisition component and building multiple traffic takeover scenarios;

[0056] A data acquisition module for collecting multi-modal data and influencing factor data during the whole process of the automatic driving system switching working modes in different traffic takeover scenarios based on the driving data acquisition component;

[0057] The multi-modal data includes vehicle driving data, driver physiological data and subjective evaluation data;

[0058] The influencing factor data includes driver inherent influencing factor data, environmental influencing factor data and situational influencing factor data;

[0059] A feature extraction module for preprocessing the multi-modal data and extracting multiple relevant feature indicators from the preprocessed multi-modal data;

[0060] A weight calculation module, which is used to configure weights for each relevant feature index through subjective and objective weighting methods, and calculate the optimal weights of each relevant feature index;

[0061] A comprehensive scoring module, which is used to construct a comprehensive evaluation model for autonomous driving takeover performance based on the DEA model, and obtain the comprehensive score of autonomous driving takeover performance through each relevant feature index and its optimal weight;

[0062] An influence importance module, which is used to construct a generalized linear mixed model based on the comprehensive takeover performance score and influence factor data to evaluate the influence importance of different influence factors in the influence factor data on the comprehensive takeover performance score.

[0063] Optionally, it further includes an information feedback module, which is used to feedback the relevant feature indexes of the multimodal data, the comprehensive score of autonomous driving takeover performance, and the influence importance to the entity related to autonomous driving optimization.

[0064] The beneficial effects brought by the embodiments provided by the present invention include:

[0065] The present invention obtains multimodal data through a driving data acquisition component, and constructs a multi-dimensional takeover performance comprehensive evaluation model based on the DEA model based on the feature indexes of the multimodal data, solves the limitation of the single dimension of the existing evaluation system, and realizes the actual score that accurately reflects the autonomous driving takeover performance;

[0066] The present invention quantifies the influence factors of the comprehensive takeover performance score through a generalized linear mixed model, analyzes the interaction effects of driver distraction, driver's individual characteristics and environmental factors, breaks through the limitation of traditional qualitative analysis, and realizes the dynamic analysis and accurate attribution of the influence of multiple influence factors on the takeover performance, providing strong support for accurately studying the action mechanism of each influence factor on the takeover performance in different takeover scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can also obtain other embodiments based on these drawings.

[0068] Figure 1 The flowchart showing the method for evaluating autonomous driving takeover performance considering multiple influence factors according to the embodiments of this specification;

[0069] Figure 2 The structural flowchart showing the comprehensive evaluation model for autonomous driving takeover performance according to the embodiments of this specification;

[0070] Figure 3 Shows a structural flowchart of a generalized linear mixed model according to an embodiment of the present specification;

[0071] Figure 4 Shows a working schematic diagram of an autonomous driving takeover performance evaluation system considering multiple influencing factors according to an embodiment of the present specification. Detailed implementation manners

[0072] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In the following detailed description, many specific details are set forth in order to provide a comprehensive understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without some of these specific details. The following description of the embodiments is only intended to provide a better understanding of the present invention by showing examples of the present invention.

[0073] However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known technologies are omitted to avoid unnecessarily confusing the concepts of the present disclosure.

[0074] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The term "including" used herein indicates the presence of features, steps, operations, but does not exclude the presence or addition of one or more other features. It should be noted that all terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of the present specification and should not be interpreted in an idealized or overly rigid manner.

[0075] Embodiment 1

[0076] As Figures 1-3 shown, this embodiment provides a method for evaluating the performance of autonomous driving takeover considering multiple influencing factors, including the following steps:

[0077] S1: Construct a driving data acquisition component and set up multiple traffic takeover scenarios;

[0078] Exemplarily, S1 includes the following specific steps:

[0079] S101: Construct a driving data acquisition component;

[0080] Specifically, the driving data acquisition component includes a physiological data acquisition unit, a subjective evaluation acquisition unit, a driver information acquisition unit, a virtual cockpit, and multiple electronic display devices;

[0081] The virtual cockpit is provided with a cockpit controller for simulating driving and is deployed with driving simulation software; the driving simulation software is used to collect vehicle driving data and environmental impact factor data; the physiological data acquisition unit is used to collect the physiological data of the driver and situational impact factor data; the subjective evaluation acquisition unit is used to collect the subjective evaluation data of the driver under different traffic takeover scenarios; the driver information acquisition unit is used to collect the inherent impact factor data of the driver;

[0082] In this embodiment, when the driver starts the driving simulation software in the virtual cockpit to start simulating driving, the driving simulation software collects vehicle driving data by real-time reading various state information of the vehicle to accurately reflect the actual situation during the driving process; at the same time, the driving simulation software obtains environmental impact factor data according to the pre-set simulation scenario parameters in advance.

[0083] In some embodiments, the physiological data acquisition unit is a wearable physiological monitoring device; when the driver enters the virtual cockpit, the physiological data acquisition unit continuously monitors and records the physiological data of the driver in a direct contact or non-contact manner until the simulation driving ends, so as to capture the physiological state changes of the driver during the whole driving process;

[0084] Among them, the wearable physiological monitoring device includes an electroencephalograph, an electrocardiograph, a skin conductance measurement device, and an eye tracker.

[0085] In this embodiment, during the simulation driving process, the driver is required to perform different distraction tasks (such as cognitive distraction, visual distraction, operation distraction, etc.), and the eye movement trajectory when the driver's attention is transferred is captured by the eye tracker, the heart rate variability fluctuation caused by the distraction task is monitored in real time by the electrocardiograph, or the emotional stress level is reflected by the change of skin resistance through the skin conductance measurement device to detect the driver's performance under different distraction behaviors, so as to obtain the driver's distraction situation; it is also possible to synchronize the physiological data of the driver obtained by the physiological data acquisition unit and the vehicle driving data obtained by the driving simulation software through millisecond-level timestamps to construct a distraction state recognition model to obtain the driver's distraction situation.

[0086] In some embodiments, the subjective evaluation collection unit is the pre-designed NASA-TLX subjective evaluation questionnaire. The evaluation dimensions of the NASA-TLX subjective evaluation questionnaire include mental workload level (ML), physical workload level (PL), time demand level (TR), task performance level (TP), effort level (EL), and frustration level (FL), so as to reflect the driver's evaluations in multiple aspects such as driving experience, takeover difficulty, and self-feelings in this scenario through multiple evaluation dimensions; the driver information collection unit is the pre-designed driver information questionnaire, and the information to be filled in the driver information questionnaire includes the driver's age, gender, and driving experience.

[0087] Among them, both the NASA-TLX subjective evaluation questionnaire and the driver information questionnaire have been pre-configured in the interactive terminal of the cockpit controller. By having the driver fill in the NASA-TLX subjective evaluation questionnaire and the driver information questionnaire on the interactive terminal, the corresponding subjective evaluation data and driver inherent influencing factor data can be obtained.

[0088] S102: Import the OpenStreetMap data into the driving simulation software in the driving data collection component, and use the actual map information to build various traffic takeover scenarios;

[0089] S103: Through the driving simulation software, adjust the environmental parameters and traffic flow conditions in the traffic takeover scenarios to make the traffic takeover scenarios more realistic;

[0090] S104: Set different trigger point positions in various traffic takeover scenario maps.

[0091] In this embodiment, the trigger point refers to a specific condition or event that causes the automatic driving system to prompt the driver to take over the vehicle; for example, when the vehicle approaches a road construction area, encounters an unrecognizable obstacle, or the sensor performance is limited in bad weather, etc., the system may trigger a takeover prompt when the vehicle enters the range affected by the event.

[0092] S2: Based on the driving data collection component, obtain multi-modal data and influencing factor data in the whole process of the automatic driving system switching working modes under different traffic takeover scenarios;

[0093] In some embodiments, the multi-modal data includes vehicle driving data, driver physiological data, and subjective evaluation data; the influencing factor data includes driver inherent influencing factor data, environmental influencing factor data, and situational influencing factor data;

[0094] Exemplarily, S2 specifically includes the following steps:

[0095] S201: Run the data collection component to obtain driver inherent influencing factor data;

[0096] Among them, the driver's inherent influencing factor data includes the driver's age, gender, and driving experience;

[0097] S202: When the vehicle's autonomous driving reaches the takeover trigger point position of the traffic takeover scenario, the autonomous driving system issues a manual driving takeover request and issues a warning to prompt the driver to take over, so as to realize the switching of the autonomous driving system from the autonomous driving mode to the manual driving mode;

[0098] Among them, when the autonomous driving system issues a manual driving takeover request and issues a warning to prompt the driver to take over, after the driver confirms the switch to the manual driving mode, the autonomous driving system switches from the autonomous driving mode to the manual driving mode;

[0099] S203: When the driver determines that the current road environment meets the autonomous driving conditions, trigger the autonomous driving takeover to realize the switching of the autonomous driving system from the manual driving mode to the autonomous driving mode;

[0100] Among them, when the driver determines that the current road environment meets the autonomous driving conditions, the driver manually presses the autonomous driving mode trigger button to trigger the autonomous driving to take over the vehicle, realizing the switching of the autonomous driving system from the manual driving mode to the autonomous driving mode;

[0101] S204: Starting from the moment when the manual driving takeover request is triggered, synchronously collect and obtain the vehicle driving data, driver physiological data, environmental influencing factor data, and situational influencing factor data during the whole process of the vehicle's autonomous driving system switching working modes;

[0102] Among them, the whole process of the vehicle's autonomous driving system switching to the working mode is the whole process starting from switching from the autonomous driving mode to the manual driving mode until the vehicle switches back to the autonomous driving mode;

[0103] Specifically, the driver physiological data includes electroencephalogram data, electrocardiogram data, galvanic skin response data, and eye movement data; the environmental influencing factor data includes weather conditions, road types, and traffic flow; the situational influencing factor data includes the driver's distraction situation;

[0104] It should be noted that in some embodiments, the comprehensive performance scores in aspects such as driving behavior, driving safety, human-machine interaction, and driving environment adaptability can also be studied by obtaining multi-modal data and influencing factor data during the whole process of simulated driving.

[0105] S205: After each traffic takeover scenario simulation ends, obtain the driver's subjective evaluation data;

[0106] S3: Preprocess the multi-modal data and extract multiple relevant feature indicators from the preprocessed multi-modal data;

[0107] Exemplarily, S3 specifically includes the following steps:

[0108] S301: Perform data preprocessing on vehicle driving data and driver physiological data;

[0109] Specifically, the data preprocessing includes denoising processing, outlier removal processing, normalization processing, and filling in missing values;

[0110] In this embodiment, the noise signals in the vehicle driving data are removed through a digital filtering algorithm, and the abnormal data points in the obvious vehicle driving data are identified and removed through a logical judgment rule; at the same time, the driver physiological data collected by different physiological monitoring devices are normalized, and the electroencephalogram data, electrocardiogram data, galvanic skin response data, and eye movement data are uniformly converted into the standard numerical range for subsequent comprehensive analysis; finally, for the possible data missing situations in the vehicle driving data and driver physiological data, a data interpolation algorithm is used to fill in the missing values to ensure data continuity and integrity.

[0111] S302: Align and associate the subjective evaluation data, preprocessed vehicle driving data, and preprocessed driver physiological data according to the time stamp to construct a standardized data set;

[0112] S303: Extract the relevant characteristic indicators of the vehicle driving data, driver physiological data, and subjective evaluation data.

[0113] In some embodiments, the relevant characteristic indicators of the vehicle driving data include vehicle driving speed, vehicle acceleration, pedal performance, steering wheel angle, collision time, and lane centerline offset distance;

[0114] Among them, the time to collision (TTC) is a dynamic safety indicator that comprehensively considers the vehicle speed and the headway distance, and is used to measure the estimated time for the vehicle to collide with the obstacle in front.

[0115] In some embodiments, the relevant characteristic indicators of the driver physiological data include the frequency ratio of θ waves to β waves in the electroencephalogram, the heart rate growth rate of the driver, the heart rate variability of the driver, the galvanic skin response growth rate of the driver, the proportion of the fixation duration in the key area, and the pupil area change rate.

[0116] In some embodiments, the relevant characteristic indicator of the subjective evaluation data includes the overall load of the driver, where the overall load refers to the mean value of the sum of the scores of each evaluation sub-dimension in the evaluation form by the driver multiplied by the corresponding sub-dimension weight, so as to quantify the overall cognitive load of the driver during the takeover process.

[0117] S4: Configure weights for each relevant characteristic indicator through the subjective weighting method and the objective weighting method, and calculate the optimal weights of each relevant characteristic indicator;

[0118] Exemplarily, S4 specifically includes the following steps:

[0119] By using the subjective weighting method and the objective weighting method, subjective weights and objective weights are respectively assigned to each relevant feature index in the multi-modal data;

[0120] By using the weighted average method, the subjective weight and the objective weight of each relevant feature index are combined to obtain the optimal weight of each relevant feature index;

[0121] Specifically, the functional expression of the weighted average method is:

[0122] In the formula, The optimal weight of the nth feature index; Is the combination coefficient, and 0 ≤ ≤ 1; Is the subjective weight of the nth feature index; Is the objective weight of the nth feature index;

[0123] Among them, The value of reflects the degree of emphasis on subjective judgment and objective data; if Is close to 1, it is more inclined to the subjective weight; if Is close to 0, it is more dependent on the objective weight; The specific value is determined according to the actual situation to balance the influence of subjective and objective factors on the weight.

[0124] Among them, the optimal weight of each relevant feature index can reflect the relative importance of different relevant feature indexes in evaluating the takeover performance.

[0125] S5: Construct a comprehensive evaluation model for the automatic driving takeover performance based on the DEA model, and obtain the comprehensive score of the automatic driving takeover performance through each relevant feature index and its optimal weight;

[0126] Exemplarily, S5 specifically includes the following steps:

[0127] S501: Based on the DEA method, construct a comprehensive evaluation model for the automatic driving takeover performance, and use the automatic driving takeover events in different takeover scenarios as different decision-making units in the comprehensive evaluation model;

[0128] S502: Use each relevant feature index and the optimal weight of the relevant feature index obtained in each takeover scenario as the input variables of the corresponding decision-making unit;

[0129] S503: Based on the input-oriented CCR model with constant returns to scale, calculate the relative efficiency value of each decision-making unit to obtain the comprehensive score of the takeover performance of different takeover events;

[0130] Specifically, the CCR model aims to maximize the comprehensive evaluation score of the autonomous driving takeover performance, constructs an objective function, and establishes constraint conditions to ensure that the comprehensive evaluation score of the takeover performance remains within a reasonable range and the optimal weights of all relevant characteristic indicators are non-negative.

[0131] In some embodiments, the expression of the objective function is:

[0132]

[0133] In the formula, P is the comprehensive evaluation score of the autonomous driving takeover performance; m is the total number of characteristic indicators; n is the serial number of the characteristic indicator; The optimal weight of the nth characteristic indicator; is the performance value of the nth characteristic indicator;

[0134] In some embodiments, the expression of the constraint condition is:

[0135]

[0136] In the formula, s.t. is the constraint condition; W is the optimal weight vector; T is the transpose symbol;

[0137] It should be noted that by setting the range reference value of the comprehensive evaluation score of the autonomous driving takeover performance in different takeover scenarios to 1, the comprehensive evaluation scores of different autonomous driving takeover events are comparable. When the comprehensive evaluation score is 1, it means that the takeover performance of the driver in the corresponding takeover scenario reaches the optimal; at the same time, the optimal weights of the relevant characteristic indicators represent the importance of the relevant characteristic indicators in the comprehensive evaluation, that is, if the optimal weight of the relevant characteristic indicator is negative, the optimal weight of the relevant characteristic indicator itself has no practical significance.

[0138] Among them, the takeover performance score output by the autonomous driving takeover performance comprehensive evaluation model is in the range of 0 to 1. The larger the value, the better the performance level of the driver during the takeover process, intuitively reflecting the overall quality level of each autonomous driving takeover operation, and dividing the comprehensive evaluation threshold according to the comprehensive evaluation score output by the model. The range of 0.9 - 1 is determined as excellent, 0.8 - 0.9 is determined as good, and below 0.8 is determined as poor takeover performance.

[0139] The DEA method (Data Envelopment Analysis) in this embodiment is a linear programming method for evaluating the relative efficiency of decision-making units (DMUs) with multiple inputs and multiple outputs. It evaluates the efficiency of decision-making units by comparing multiple input and output indicators and is applicable to complex systems with multiple inputs and multiple outputs. Among them, the CCR model is a classic model of the Data Envelopment Analysis (DEA) method. Its core idea is to evaluate the production efficiency of multiple decision-making units through linear programming based on relative efficiency. The comprehensive evaluation model of autonomous driving takeover performance constructed in this embodiment is also based on the theory of data envelopment analysis. Taking a single takeover event as a decision-making unit, the comprehensive takeover performance score of different takeover events is obtained by calculating the relative efficiency value of each decision-making unit.

[0140] S6: Based on the comprehensive takeover performance score and the influencing factor data, construct a generalized linear mixed model to evaluate the importance of different influencing factors in the influencing factor data on the comprehensive takeover performance score.

[0141] As Figure 3 shown, exemplarily, S6 specifically includes the following steps:

[0142] S601: Construct a generalized linear mixed model with the comprehensive autonomous driving takeover performance score as the dependent variable; S602: Based on the influencing factor data, construct the fixed effect term in the generalized linear mixed model to quantify the average influence of the main effect coefficients of different influencing factors and the interaction effect coefficients between different influencing factors on the comprehensive takeover performance score.

[0143] Among them, the main effect coefficient of an influencing factor refers to the average degree of influence on the comprehensive takeover performance score when this influencing factor changes by one unit while other influencing factors are fixed; the interaction effect coefficient between different influencing factors refers to the combined influence of the interaction between two different influencing factors on the takeover performance.

[0144] In this embodiment, by considering the interaction effects of different influencing factors, the complex influence mechanism of each influencing factor on the takeover performance under different combinations can be explored more deeply;

[0145] In some embodiments, the complex influence of the interaction effects of multiple influencing factors on the takeover performance score under different combinations can be understood by setting and considering higher-order interaction effect coefficients of the third order and above.

[0146] S603: Based on the influencing factor data, construct the driver random effect term in the mixed model to quantify the difference in the basic level of different drivers in terms of takeover performance;

[0147] In this embodiment, considering the inherent differences among different drivers that cannot be explained by influencing factors in the basic level of takeover performance, such as differences in the natural reaction speed and attention concentration ability of different drivers, even under the same external conditions, the basic levels of takeover performance of different drivers are not the same. Therefore, by introducing a driver random effect term, the differences in the degree of influence of different driver individuals on the takeover performance score when facing the same change in influencing factors are reflected, that is, the sensitivity of different drivers to changes in different influencing factors.

[0148] S604: Based on the influencing factor data, construct the scenario random effect term in the hybrid model to quantify the influence of different driving scenarios on the takeover performance;

[0149] In this embodiment, considering that the scenario random effect term has complexity and diversity by considering different driving scenarios, even for the same driver and the same values of influencing factors, the basic levels of takeover performance may be different in different scenarios. Therefore, by introducing a scenario random effect term, the moderating effect of the relationship between each influencing factor and the takeover performance score in different driving scenarios is reflected, that is, the degree of influence and the way of action of the same influencing factor on the takeover performance score will vary in different scenarios.

[0150] S605: Introduce a random error term into the hybrid model to quantify other unexplained random errors;

[0151] Among them, the random error is a sudden and tiny interference factor that is difficult to quantify in advance and incorporated into the model;

[0152] S606: Based on the hybrid model, evaluate the importance of the influence of different influencing factors in the influencing factor data on the comprehensive takeover performance score.

[0153] Among them, the functional expression of the generalized linear mixed model is:

[0154] G1 + G2 + G3

[0155] +

[0156]

[0157]

[0158] In the formula, is the comprehensive score of the autonomous driving performance; i is the numbering of the takeover times; j is the driver number; k is the driving scenario number; is the fixed effect; is the driver random effect; is the scene random effect; is the random error term; is the fixed effect intercept, which is used to reflect the comprehensive takeover performance score under the ideal benchmark state; M is the total number of influencing factors; p and q are different influencing factors, where p, q = 1, 2, …, M; is the fixed effect coefficient of the p-th influencing factor; is the interaction effect coefficient between the p-th influencing factor and the q-th influencing factor; is the value of the p-th influencing factor at the i-th takeover of the j-th driver and the k-th driving scene; is the random intercept of the j-th driver, which reflects the deviation between the comprehensive takeover performance score of this driver and the average comprehensive takeover performance score under the benchmark state; is the random slope of the j-th driver for the p-th influencing factor, which is used to reflect the difference in the influence degree of different driver individuals on the comprehensive takeover performance score when facing the change of the p-th influencing factor; is the random intercept of the k-th scene, which is used to reflect the deviation between the comprehensive takeover performance score of this scene and the average comprehensive takeover performance score under the benchmark state; is the random slope for the p-th influencing factor in the k-th scene, which is used to reflect the difference in the influence degree of the change of the p-th influencing factor in different driving scenes on the comprehensive takeover performance score.

[0159] In some embodiments, it further includes S7: feeding back the relevant characteristic indexes of the multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the influence importance to the entities related to autonomous driving optimization.

[0160] Among them, the entities related to autonomous driving optimization include various entities such as autonomous driving system developers, traffic management departments, drivers, enterprises, and experts;

[0161] In this embodiment, when the relevant characteristic indexes of the multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the influence importance are fed back to the autonomous driving system developers, the developers can use the feedback information to guide them to formulate personalized system optimization strategies, accurately adjust the way, intensity, and timing of the takeover prompt, optimize the layout of vehicle sensors, and improve the design of the human-machine interaction interface, so that the autonomous driving system can better adapt to diverse actual situations and improve the coordination and fluency in the process of human-machine co-driving.

[0162] After the relevant feature indicators of multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the importance of the influence are fed back to the driver, the driver can intuitively understand his own performance advantages and disadvantages in terms of the safety, timeliness, accuracy of vehicle control, and his own state adjustment through the data, so as to understand the weak links of his own driving level under the influence of different factors, and then targeted improve driving skills and improve driving habits.

[0163] After the relevant feature indicators of multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the importance of the influence are fed back to the traffic management department, it is possible to formulate a more practical and targeted driver training syllabus and traffic management strategy according to the importance of different influencing factors among different driver groups.

[0164] In this embodiment, after the relevant feature indicators of multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the importance of the influence are fed back to the entities related to autonomous driving optimization for application, and then the model parameters in the autonomous driving takeover performance evaluation method are optimized through the actual effects after application, so as to achieve a virtuous cycle from evaluation to application, and then from application feedback to evaluation system optimization.

[0165] And then fed back to the evaluation system, so that the autonomous driving evaluation system is optimized.

[0166] Embodiment 2

[0167] As Figure 4 shown, this embodiment provides an autonomous driving takeover performance evaluation system considering multiple influencing factors, which is characterized in that it includes:

[0168] A scenario construction module for constructing a driving data acquisition component and building a variety of traffic takeover scenarios;

[0169] A data acquisition module for obtaining multi-modal data and influencing factor data in the whole process of the autonomous driving system switching working modes under different traffic takeover scenarios based on the driving data acquisition component;

[0170] In some embodiments, the multi-modal data includes vehicle driving data, driver physiological data, and subjective evaluation data;

[0171] In some embodiments, the influencing factor data includes driver inherent influencing factor data, environmental influencing factor data, and situational influencing factor data;

[0172] A feature extraction module for preprocessing the multi-modal data and extracting multiple relevant feature indicators from the preprocessed multi-modal data;

[0173] A weight calculation module is used to configure weights for each relevant feature index through subjective and objective weighting methods, and calculate the optimal weights of each relevant feature index.

[0174] A comprehensive scoring module is used to construct a comprehensive evaluation model for autonomous driving takeover performance based on the DEA model, and obtain the comprehensive score of autonomous driving takeover performance through each relevant feature index and its optimal weight.

[0175] An influence importance module is used to construct a generalized linear mixed model based on the comprehensive takeover performance score and influence factor data to evaluate the influence importance of different influence factors in the influence factor data on the comprehensive takeover performance score.

[0176] In some embodiments, an information feedback module is further included, which is used to feedback the relevant feature indexes of the multimodal data, the comprehensive score of autonomous driving takeover performance, and the influence importance to the entities related to autonomous driving optimization.

[0177] Among them, the entities related to autonomous driving optimization include various entities such as autonomous driving system developers, traffic management departments, drivers, enterprises, and experts.

[0178] To sum up, in this embodiment, the driving data acquisition component is used to obtain multimodal data, extract the feature indexes of the multimodal data, and construct a multi-dimensional comprehensive evaluation model for takeover performance based on the DEA model, so as to comprehensively evaluate the takeover performance from multiple dimensions such as safety, timeliness, accuracy, and driver state, solve the limitation of the single dimension of the existing evaluation system, make the content of the comprehensive evaluation of takeover performance more rich and comprehensive, and be able to truly reflect the actual situation of autonomous driving takeover.

[0179] This embodiment is different from the prior art in generally studying the influence of common influence factors on the autonomous driving takeover performance and lacking systematic and in-depth quantitative analysis. This embodiment constructs a generalized linear mixed model to quantify the influence factors of the comprehensive takeover performance score. By reasonably setting the dependent variable (i.e., the comprehensive evaluation score of takeover performance), independent variables (including various influence factors), random effects (considering variation factors such as driver individual differences and driving scenario characteristics that are difficult to directly explain by independent variables), and fixed effects (main effects and interaction effects of independent variables), it accurately analyzes the interaction effects of driver distraction degree, driver individual characteristics, and environmental factors on the comprehensive takeover performance score, breaks through the limitation of traditional qualitative analysis, and realizes the dynamic analysis and accurate attribution of the influence of multiple influence factors on takeover performance, providing strong support for accurately studying the action mechanism of each influence factor on takeover performance in different takeover scenarios.

[0180] In this embodiment, the relevant feature indicators of multi-modal data, the comprehensive score of the autonomous driving takeover performance, and the impact importance are fed back to the entities related to autonomous driving optimization to construct a feedback closed-loop mechanism. Through a feedback mechanism covering multiple entities and forming a closed loop, a virtuous cycle from evaluation to application and then from application back to the optimization of the evaluation system is realized, promoting the optimization of the autonomous driving system, the improvement of the driver's ability, and the improvement of traffic management strategies, facilitating the coordinated development of the human-machine co-driving mode, and giving full play to the practical value of the comprehensive score of the takeover performance.

[0181] The above are only the preferred embodiments of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes falling within the meaning and scope of the equivalent elements of the claims within the present invention.

Claims

1. A method for evaluating the performance of automatic driving takeover considering multiple influencing factors, characterized in that: The steps include: S1: Build driving data collection components and set up various traffic takeover scenarios; S2: Based on the driving data collection component, obtain the multimodal data and influencing factor data of the entire process of the autonomous driving system switching working mode in different traffic takeover scenarios; The multimodal data includes vehicle driving data, driver physiological data and subjective evaluation data; The influencing factor data includes driver's inherent influencing factor data, environmental influencing factor data and situational influencing factor data; S3: preprocessing the multimodal data and extracting multiple relevant feature indicators from the preprocessed multimodal data; S4: Through subjective weighting method and objective weighting method, weights are assigned to each relevant feature indicator, and the optimal weights of each relevant feature indicator are calculated; S5: Construct a comprehensive evaluation model of autonomous driving takeover performance based on the DEA model, and obtain the comprehensive score of autonomous driving takeover performance through various relevant characteristic indicators and their optimal weights; S6: Based on the comprehensive score of takeover performance and the influencing factor data, a generalized linear mixed model is constructed to evaluate the importance of different influencing factors in the influencing factor data on the comprehensive score of takeover performance.

2. The method according to claim 1, characterized in that It also includes S7: feeding back the relevant characteristic indicators of the multimodal data, the comprehensive score of the autonomous driving takeover performance and the impact importance to the subject related to the autonomous driving optimization.

3. The method according to claim 1, characterized in that The S1 specifically includes the following steps: S101: Build driving data collection components; S102: By importing OpenStreetMap data into the driving simulation software in the driving data collection component, various traffic takeover scenarios are built using actual map information; S103: adjusting environmental parameters and traffic flow conditions in the traffic takeover scenario through driving simulation software to make the traffic takeover scenario more realistic; S104: Setting different takeover trigger point locations in various traffic takeover scenario maps.

4. The method according to claim 1 or 3, characterized in that: The driving data collection component includes a physiological data collection unit, a subjective evaluation collection unit, a driver information collection unit, a simulated cockpit and a plurality of electronic display devices; The simulated cockpit is provided with a cockpit controller for simulating driving and is deployed with driving simulation software; The driving simulation software is used to collect vehicle driving data and environmental influencing factor data; The physiological data collection unit is used to collect the driver's physiological data and situational influencing factor data; The subjective evaluation collection unit is used to collect subjective evaluation data of drivers in different traffic takeover scenarios; The driver information collection unit is used to collect the driver's inherent influencing factor data.

5. The method according to claim 1, characterized in that The S2 specifically includes the following steps: S201: running the data acquisition component to obtain the driver's inherent influencing factor data; The driver's inherent influencing factor data includes the driver's age, driver's gender and driving experience; S202: When the vehicle reaches the takeover trigger point of the traffic takeover scenario, the automatic driving system issues a manual driving takeover request and issues an early warning to prompt the driver to take over, so that the automatic driving system switches from the automatic driving mode to the manual driving mode; S203: When the driver determines that the current road environment meets the conditions for automatic driving, triggering automatic driving takeover to enable the automatic driving system to switch from the manual driving mode to the automatic driving mode; S204: From the moment the manual driving takeover request is triggered, synchronously acquiring vehicle driving data, driver physiological data, environmental influencing factor data, and situational influencing factor data during the entire process of the vehicle's automatic driving system switching working mode; The environmental influencing factor data includes weather conditions and road types; the situational influencing factor data includes driver distraction; S205: After each traffic takeover scenario simulation is completed, the driver's subjective evaluation data is obtained.

6. The method according to claim 1, characterized in that The S3 specifically includes the following steps: S301: Preprocessing the vehicle driving data and the driver's physiological data; The data preprocessing includes denoising, outlier removal, normalization and missing value completion; S302: aligning and associating the subjective evaluation data, the preprocessed vehicle driving data, and the preprocessed driver physiological data according to timestamps to construct a standard data set; S303: Extract relevant characteristic indicators of vehicle driving data, driver physiological data and subjective evaluation data.

7. The method according to claim 1, characterized in that The S5 specifically includes the following steps: S501: Based on the DEA method, a comprehensive evaluation model of autonomous driving takeover performance is constructed, and autonomous driving takeover events under different takeover scenarios are used as different decision units in the comprehensive evaluation model; S502: taking each relevant characteristic index and the optimal weight of the relevant characteristic index obtained in each takeover scenario as input variables of the corresponding decision unit; S503: Based on the input-oriented scale return constant CCR model, the relative efficiency value of each decision-making unit is calculated to obtain the comprehensive score of takeover performance of different takeover events; The CCR model includes an objective function and constraints; The expression of the objective function is: ； Where, P is the comprehensive score of the autonomous driving takeover performance; m is the total number of characteristic indicators; n is the sequence number of the characteristic indicator; The optimal weight of the nth feature index; is the performance value of the nth characteristic indicator; The constraint condition is expressed as: ； Where st is the constraint condition; W is the optimal weight vector; T is the transposed symbol.

8. The method according to claim 1, characterized in that The generalized linear mixed model in S6 has a functional expression as follows: G1+G2+G3 ; + ； ； ； In the formula, is the comprehensive performance evaluation score; i is the number of times the takeover occurred; j is the driver number; k is the driving scene number; is a fixed effect; is the driver random effect; is the random effect of scene; is the random error term; is the fixed effect intercept; M is the total number of influencing factors; p and q are different influencing factors, where p, q = 1, 2, …, M; is the fixed effect coefficient of the pth influencing factor; is the interaction effect coefficient between the pth influencing factor and the qth influencing factor; is the value of the pth influencing factor at the i-th takeover under the j-th driver and the k-th driving scenario; is the random intercept of the jth driver; is the random slope of the jth driver on the pth influencing factor; is the random intercept of the kth scenario; is the random slope of the pth influencing factor in the kth scenario.

9. An automatic driving takeover performance evaluation system considering multiple influencing factors, characterized in that: include: Scenario building module, used to build driving data collection components and build various traffic takeover scenarios; The data collection module is used to collect multimodal data and influencing factor data during the whole process of the automatic driving system switching working mode in different traffic takeover scenarios based on the driving data collection component; The multimodal data includes vehicle driving data, driver physiological data and subjective evaluation data; The influencing factor data includes driver's inherent influencing factor data, environmental influencing factor data and situational influencing factor data; A feature extraction module is used to preprocess the multimodal data and extract multiple relevant feature indicators from the preprocessed multimodal data; The weight calculation module is used to configure weights for each relevant feature indicator through subjective weighting method and objective weighting method, and calculate the optimal weight of each relevant feature indicator; The comprehensive scoring module is used to construct a comprehensive evaluation model of autonomous driving takeover performance based on the DEA model, and obtain the comprehensive score of autonomous driving takeover performance through various relevant characteristic indicators and their optimal weights; The influence importance module is used to construct a generalized linear mixed model based on the comprehensive score of takeover performance and the influencing factor data to evaluate the influence importance of different influencing factors in the influencing factor data on the comprehensive score of takeover performance.

10. The system according to claim 9, characterized in that It also includes an information feedback module, which is used to feed back the relevant characteristic indicators of the multimodal data, the comprehensive score of the autonomous driving takeover performance and the impact importance to the subject related to the autonomous driving optimization.

Images