Automatic driving traffic regulation compliance evaluation method and electronic device

Abstract

This application discloses a method and electronic device for evaluating the traffic compliance of autonomous driving systems. The method includes: acquiring human driving datasets and autonomous driving datasets and extracting data from lane-changing scenarios; extracting traffic rule text from lane-changing scenarios and parsing the text based on a large language model; determining fuzzy traffic rule evaluation indicators based on the temporal logic expression corresponding to the fuzzy traffic rule text, and determining the limit value corresponding to the fuzzy traffic rule evaluation indicator for the vehicle under test; determining the fuzzy traffic rule score of the vehicle under test based on the limit value and the actual value corresponding to the fuzzy traffic rule evaluation indicator; determining the explicit traffic rule score of the vehicle under test; determining the behavioral normativity score of the vehicle under test; and determining the traffic compliance evaluation result based on the fuzzy traffic rule score, explicit traffic rule score, and behavioral normativity score. This method achieves an objective, comprehensive, and interpretable evaluation of the traffic compliance of autonomous driving systems in lane-changing scenarios.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, and in particular to an autonomous driving traffic regulation compliance evaluation method and electronic device. Background Technology

[0002] With the development of autonomous driving, autonomous vehicles will share roads with human-driven vehicles for a long time to come, creating a mixed traffic environment. Since there are no specific traffic rules designed for autonomous driving, autonomous vehicles must adhere to the same traffic rules as human drivers to ensure traffic safety in this mixed environment. For example, current autonomous vehicles rely on hard rules (such as speed limits of 60 km / h and prohibitions against running red lights) to understand and execute traffic rules expressed in natural language. Furthermore, existing road traffic rules contain a large number of qualitative descriptions and vague expressions, such as "maintain a safe distance," "slow down," and "do not impede traffic."

[0003] However, these vague traffic regulations are based on the cognitive habits and comprehension abilities of human drivers, making it difficult to directly translate them into instructions that can be executed by autonomous driving systems. This results in autonomous driving systems being unable to accurately determine whether their behavior complies with traffic regulations during actual road tests.

[0004] Furthermore, existing methods for evaluating the traffic compliance of autonomous driving systems exhibit two main tendencies. First, they over-rely on explicit traffic regulations (such as speed limits, prohibitions against running red lights, and prohibitions against crossing solid lines), neglecting the evaluation of ambiguous regulations (such as safe following distance and avoiding obstruction). Second, they remain at the qualitative evaluation level, with evaluation indicators typically set based on expert experience, lacking scientific data-driven determination methods. Neither method can truly reflect the degree of complete compliance of autonomous driving systems with traffic regulations. Ultimately, this results in the absence of a comprehensive evaluation method for autonomous driving traffic compliance, making it difficult to comprehensively, objectively, and interpretably evaluate the degree of traffic rule compliance by autonomous driving systems in complex scenarios. Summary of the Invention

[0005] This application addresses some of the shortcomings mentioned in the background technology by providing a method and electronic device for evaluating the traffic compliance of autonomous driving. This method scientifically determines the limits of fuzzy traffic regulation evaluation indicators through comparative analysis of human driving data and autonomous driving data, thus solving the evaluation bias problem caused by the reliance on expert experience in traditional methods. Simultaneously, by fusing fuzzy traffic regulation scores, explicit traffic regulation scores, and behavioral normativity scores, it achieves a comprehensive and objective evaluation of the degree of traffic regulation compliance of autonomous driving, providing an interpretable scientific basis for system verification.

[0006] Firstly, a method for evaluating the compliance of autonomous driving with traffic regulations is provided, comprising: acquiring a human driving dataset and an autonomous driving dataset and extracting data under lane-changing scenarios, wherein the human driving dataset includes driving data of multiple first vehicles driven by humans under lane-changing scenarios, and the autonomous driving dataset includes driving data of multiple second vehicles driven by autonomous vehicles under lane-changing scenarios; extracting traffic rule text under lane-changing scenarios, and parsing the traffic rule text based on a large language model to obtain the temporal logical expression corresponding to the traffic rule text, wherein the traffic rule text includes fuzzy traffic rule text and explicit traffic rule text; and based on the fuzzy... The temporal logic expression corresponding to the traffic regulations text is used to determine the fuzzy traffic regulations evaluation index and the limit value corresponding to the fuzzy traffic regulations evaluation index for the vehicle under test, wherein the vehicle under test is an autonomous vehicle. Based on the limit value and the actual value corresponding to the vehicle under test under the fuzzy traffic regulations evaluation index, the fuzzy traffic regulations score of the vehicle under test is determined. Based on the temporal logic expression corresponding to the explicit traffic regulations text, the explicit traffic regulations evaluation index is determined and the explicit traffic regulations score of the vehicle under test is determined. The behavioral compliance score of the vehicle under test is determined. Based on the fuzzy traffic regulations score, the explicit traffic regulations score, and the behavioral compliance score, the traffic regulations compliance evaluation result of the vehicle under test is determined.

[0007] This scheme solves the problem of objectively determining the limits of fuzzy traffic regulation evaluation indicators by comparing and analyzing driving data based on human driving data and autonomous driving data. Simultaneously, through a fusion evaluation mechanism of three types of evaluation indicators, it achieves a scientific measurement of the completeness and compliance of traffic regulation clauses, avoiding the one-sidedness of evaluations based on a single indicator and improving the completeness, objectivity, and interpretability of the evaluation results.

[0008] In conjunction with the first aspect, in a possible implementation of the first aspect, the traffic rule text is parsed based on a large language model to obtain the temporal logic expression corresponding to the traffic rule text. This includes: decomposing the traffic rule text into multiple atomic propositions, which are used to indicate the behavioral themes, constraints, triggering conditions, and behavioral requirements in the traffic rule text; guiding the large language model to determine the logical relationships and temporal constraints in the traffic rule text based on templated prompts; and determining the temporal logic expression corresponding to the traffic rule text based on the multiple atomic propositions, the logical relationships, and the temporal constraints.

[0009] This scheme leverages the powerful semantic understanding capabilities of large language models to accurately transform natural language traffic regulations into computable temporal logic expressions, providing a foundation for subsequent evaluation.

[0010] In conjunction with the first aspect, in a possible implementation of the first aspect, the fuzzy traffic regulation evaluation index includes the safe distance between the vehicle and the following vehicle in the adjacent lane, and the actual value corresponding to the test vehicle under the fuzzy traffic regulation evaluation index is the actual distance between the test vehicle and the following vehicle in the adjacent lane.

[0011] This scheme establishes evaluation indicators for vague traffic regulations, addressing ambiguous clauses such as "must not affect the normal driving of other motor vehicles."

[0012] In conjunction with the first aspect, in a possible implementation of the first aspect, determining the limit value corresponding to the fuzzy traffic regulation evaluation index of the vehicle under test includes: determining first driving behavior information of the plurality of first vehicles and their adjacent lane following vehicles based on the human driving dataset, and determining second driving behavior information of the plurality of second vehicles and their adjacent lane following vehicles based on the autonomous driving dataset, wherein the first driving behavior information includes the speed information of the adjacent lane following vehicles of the first vehicle, and the second driving behavior information includes the speed information of the adjacent lane following vehicles of the second vehicle; grouping the plurality of first vehicles into multiple speed ranges based on the first driving behavior information to obtain multiple first vehicle speed groups, and grouping the plurality of second vehicles into multiple speed ranges based on the second driving behavior information to obtain multiple second vehicle speed groups. The system comprises: a plurality of first vehicle speed groups and a plurality of second vehicle speed groups, wherein the speed range of each first vehicle speed distribution in the plurality of first vehicle speed groups is the same as the speed range of the corresponding second vehicle speed group; determining a plurality of first limits of fuzzy traffic regulations corresponding to the plurality of first vehicle speed groups and a plurality of second limits of fuzzy traffic regulations corresponding to the plurality of second vehicle speed groups, wherein the plurality of first limits and the plurality of second limits correspond one-to-one with the plurality of speed ranges; and weightedly fusing each of the plurality of first limits with each of the corresponding second limits to obtain a plurality of fused limits of fuzzy traffic regulations corresponding to the plurality of speed ranges, wherein each of the plurality of fused limits is used to constrain the driving behavior of vehicles in the speed range corresponding to each fused limit.

[0013] This scheme employs a speed range grouping strategy to address the issue of dynamically adjusting evaluation limits under different operating conditions. Furthermore, by weighted fusion of human driving and autonomous driving data, the scientific validity and scenario adaptability of the limit settings are ensured.

[0014] In conjunction with the first aspect, in possible implementations of the first aspect, the first driving behavior information also includes the speed, minimum acceleration, average speed, minimum acceleration of the following vehicle in the adjacent lane, average speed of the following vehicle in the adjacent lane, longitudinal distance, lane change time, time to collision (TTC), and time headway (THW) of the plurality of first vehicles. The second driving behavior information also includes the speed, minimum acceleration, average speed, minimum acceleration of the following vehicle in the adjacent lane, average speed of the following vehicle in the adjacent lane, longitudinal distance, lane change time, TTC, and THW of the plurality of second vehicles.

[0015] This scheme extracts multi-dimensional driving behavior features, covering the entire lifecycle from decision-making and execution to stabilization, providing a comprehensive data foundation for subsequent calculations and enhancing the completeness of the evaluation system.

[0016] In conjunction with the first aspect, in a possible implementation of the first aspect, based on the first driving behavior information, the multiple first vehicles are grouped into multiple speed ranges to obtain multiple first vehicle speed groups, including: based on the first driving behavior information and kernel density estimation (KDE), the multiple first vehicles are grouped into multiple speed ranges to obtain multiple first vehicle speed groups, and the constraint index statistical significance of the multiple first vehicle speed groups is verified by at least one testing method; based on the second driving behavior information, the multiple second vehicles are grouped into multiple speed ranges to obtain multiple second vehicle speed groups, including: based on the second driving behavior information and KDE, the multiple second vehicles are grouped into multiple speed ranges to obtain multiple second vehicle speed groups, and the constraint index statistical significance of the multiple second vehicle speed groups is verified by at least one testing method.

[0017] This scheme utilizes kernel density estimation to nonparametrically model the speeds of vehicles following in adjacent lanes. It determines speed grouping based on the speed distribution of vehicles following in adjacent lanes, rather than using existing fixed grouping methods. Furthermore, the inclusion of statistical significance tests ensures the scientific reliability of the grouping results.

[0018] In conjunction with the first aspect, in a possible implementation of the first aspect, determining multiple first limits of the fuzzy traffic regulations corresponding to the multiple first vehicle speed groups and multiple second limits of the fuzzy traffic regulations corresponding to the multiple second vehicle speed groups includes: modeling within the multiple first vehicle speed groups and the multiple second vehicle speed groups using a Gaussian mixture model (GMM), wherein the parameters of the GMM are determined by the maximum likelihood estimation method, and the optimal number of clusters of the GMM is determined by the Bayesian information criterion (BIC); determining constraint index thresholds corresponding to the multiple first vehicle speed groups and the multiple second vehicle speed groups based on the GMM and the boundary points based on posterior probability; determining multiple candidate limits for each first vehicle speed group and each second vehicle speed group; and selecting the multiple first limits and the multiple second limits from the multiple candidate limits based on the recall rate and false positive rate (FPR) corresponding to the multiple candidate limits and the constraint index thresholds.

[0019] This approach enables GMM modeling to accurately characterize the distribution features of driving behavior, and the optimal cluster number selection based on BIC avoids subjective intervention. Furthermore, the use of recall rate and false positive rate as dual constraints ensures that the final limit is determined safely and stably.

[0020] In conjunction with the first aspect, in a possible implementation of the first aspect, each of the plurality of first limits is weighted and fused with each of the corresponding second limits to obtain a plurality of fused limits for the fuzzy traffic regulations corresponding to the plurality of speed ranges, including: determining a driving mode consistency index (DMCI) evaluation value jointly corresponding to each of the plurality of first limits and each of the corresponding second limits based on each of the plurality of first limits and each of the corresponding second limits; determining a first weight for each of the plurality of first limits and a second weight for each of the corresponding second limits based on the DMCI evaluation value jointly corresponding to each of the plurality of first limits and each of the corresponding second limits; and determining the plurality of fused limits based on the plurality of first limits and the first weight, the plurality of second limits and the second weight.

[0021] This scheme, based on the dynamic weight allocation mechanism of DMCI evaluation values, integrates human driving experience with autonomous driving characteristics, enabling the evaluation limits to meet both the rationality of traffic habits and the safety requirements of autonomous driving.

[0022] In conjunction with the first aspect, in a possible implementation of the first aspect, the fuzzy traffic rule score of the vehicle under test is determined based on the limit and the actual value corresponding to the vehicle under test under the fuzzy traffic rule evaluation index, including: when the actual value is greater than the limit, the fuzzy traffic rule score is determined to be 1; when the actual value is less than the limit, the fuzzy traffic rule score is determined to be the ratio of the actual value to the limit.

[0023] In conjunction with the first aspect, in a possible implementation of the first aspect, determining the explicit traffic rule score of the vehicle under test includes: obtaining the score corresponding to each explicit traffic rule among one or more explicit traffic rules of the vehicle under test, wherein when the explicit traffic rule is satisfied, the corresponding score is 1, and when the explicit traffic rule is violated, the corresponding score is 0; and determining the explicit traffic rule score of the vehicle under test based on the score and weight corresponding to each explicit traffic rule among the one or more explicit traffic rules.

[0024] This scheme eliminates the ambiguity of subjective judgment by mapping compliance with each explicit traffic rule to a binary score (1 for compliance, 0 for violation). Furthermore, it reduces computational complexity by employing a linear weighted summation method.

[0025] In conjunction with the first aspect, in possible implementations of the first aspect, determining the behavioral normativity score of the vehicle under test includes: obtaining the score corresponding to each of the multiple normativity evaluation indicators used to characterize behavioral normativity, wherein, for normativity evaluation indicators where smaller values ​​are better, the score is determined based on a linear decay function of an upper threshold; for normativity evaluation indicators where larger values ​​are better, the score is determined based on a linear proportional function of a threshold; and for normativity interval-type evaluation indicators, the score is determined based on a piecewise function; and determining the behavioral normativity score based on the score and weight corresponding to each of the multiple normativity evaluation indicators.

[0026] This scheme designs linear decay functions, linear proportional functions, and piecewise functions to match the scoring function with the physical meaning of the evaluation index, thus avoiding evaluation distortion caused by a unified function.

[0027] In conjunction with the first aspect, in one of the possible implementations of the first aspect, the traffic compliance evaluation result of the vehicle under test is determined based on the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score, including: determining the weights corresponding to the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score respectively according to the entropy weight method; and determining the traffic compliance evaluation result based on the fuzzy traffic regulation score, the explicit traffic regulation score, the behavioral normativity score, and their respective weights.

[0028] This scheme uses the entropy weight method to automatically calculate the weights based on the data dispersion of each score, avoiding the bias or misjudgment that may be caused by human weighting, and making the comprehensive results more data-driven.

[0029] In a second aspect, an electronic device is provided, including one or more processors; one or more memories; the one or more memories storing one or more computer programs, the one or more computer programs including instructions that, when executed by the one or more processors, cause a method as described in any of the possible implementations of the first aspect to be performed.

[0030] Thirdly, a computer-readable storage medium is provided that stores computer instructions that, when executed on a computer, cause a method as described in any of the first aspects to possible implementations of the first aspect to be performed.

[0031] Fourthly, a computer program product is provided that, when run on a computer, causes the computer to perform a method as described in the first aspect and any possible implementation thereof. Attached Figure Description

[0032] Figure 1 A schematic flowchart of the traffic regulation compliance evaluation method for autonomous driving provided in an embodiment of this application is shown;

[0033] Figure 2 A flowchart of the traffic regulation compliance evaluation process for autonomous driving provided in an embodiment of this application is shown;

[0034] Figure 3 A schematic diagram of a lane-changing scenario provided in an embodiment of this application is shown;

[0035] Figure 4 This is a structural schematic diagram of a device provided in an embodiment of this application;

[0036] Figure 5 This is a structural schematic diagram of a system on a chip (SoC) provided in an embodiment of this application. Detailed Implementation

[0037] The technical solution of this application is described below with reference to the accompanying drawings.

[0038] With the development of autonomous driving technology, autonomous vehicles will share roads with human-driven vehicles for a considerable period, creating a mixed traffic environment. Against this backdrop, the evaluation of autonomous driving systems is becoming increasingly important, especially the evaluation of their traffic compliance. Traffic compliance evaluation of autonomous driving systems is a core technology for verifying and assessing the degree to which autonomous driving systems adhere to traffic regulations and their safety capabilities in real-world road operations.

[0039] However, as mentioned in the background section, current autonomous driving traffic compliance evaluation methods suffer from the following technical shortcomings: Existing evaluations primarily focus on explicit traffic rule indicators (such as speed limits and traffic lights), lacking effective means to evaluate ambiguous traffic rule clauses (such as maintaining a safe following distance), making it difficult to comprehensively reflect the degree of complete compliance of autonomous driving systems with traffic rules. For quantifiable evaluation indicators, limits are often based on expert experience or simple statistics, lacking scientific methods for determining limits through comparative analysis of human driving data and autonomous driving data, leading to discrepancies between evaluation limits and real-world traffic scenarios. Furthermore, as mentioned in the background section, existing evaluation methods lack a complete set of methods for evaluating autonomous driving traffic compliance, making it difficult to comprehensively, objectively, and interpretably evaluate the degree of traffic rule compliance by autonomous driving systems in complex scenarios.

[0040] For example, existing autonomous driving traffic compliance evaluation technologies mainly include comprehensive evaluation methods based on traffic violation evaluation index grading, traffic compliance evaluation methods based on scene recognition and automatic determination of traffic rules, and autonomous driving traffic compliance evaluation methods based on rule engines and digital models.

[0041] For example, the main idea of ​​a comprehensive evaluation method based on traffic violation evaluation indicators is to first obtain vehicle driving data, construct a traffic regulation compliance evaluation system, and divide the evaluation indicators into multiple levels. Then, based on the severity of different violations, demerit points, and preset weights, various violation evaluation indicators are scored separately and weighted to obtain a final traffic regulation compliance evaluation score. This type of method is mainly based on existing lists of violations, demerit point rules, and preset weights. The evaluation object is more about the degree of violation of explicit traffic regulations, and its essence is still to grade and score explicit violations. For vague traffic regulations such as "maintain a safe following distance," "do not affect normal driving," and "slow down," this type of method usually lacks the process of determining the limit value of the evaluation indicator itself, making it difficult to answer questions such as "what exactly is a safe following distance?" and "how to quantify the impact on normal driving," and it does not introduce the evaluation of such vague traffic regulations. Therefore, although this type of method can evaluate known violations, it is difficult to solve the problem of objectively determining the limit value of the evaluation indicator in vague traffic regulations, and it is also difficult to comprehensively evaluate the degree of compliance with traffic rules.

[0042] For example, the main idea of ​​traffic rule compliance evaluation methods based on scene recognition and automatic traffic rule determination is to first identify the traffic scene in which the vehicle is located by acquiring information such as the scene map, traffic targets, external environment, and location trajectory of the autonomous driving test object, and then construct semantic regions such as lane line areas, stop line areas, and pedestrian projection areas within the scene. Further, by combining the vehicle's direction of travel, speed, relative positional relationships, and target traffic elements, the method automatically determines whether the tested object violates traffic rules. However, this type of method is better suited to handling rules with clear boundaries and conditions. For clauses such as "must not affect the normal driving of other vehicles," which have semantic ambiguity and strong scene dependence, corresponding judgment boundaries still need to be given in advance; otherwise, it is difficult to directly execute automatic judgment. Therefore, although this type of method solves the problem of automatic judgment efficiency, it fails to fundamentally solve the problem of determining fuzzy traffic rule evaluation indicators and evaluation limits.

[0043] For example, the main idea behind autonomous driving traffic compliance assessment methods based on rule engines and digital models is to first acquire the target vehicle's driving data, parse the rule configuration file of a pre-set rule engine system, break down the rules into rules, conditions, and mathematical models, and then use mathematical functions or models to process and evaluate the current driving data based on the parsing results, thereby generating autonomous driving traffic compliance assessment results. However, this type of method focuses on the rule parsing and mathematical evaluation process, and essentially still relies on pre-set rule expressions and model parameters. For fuzzy traffic regulations, although they can be expressed to some extent through models, there is still a lack of targeted solutions for how to obtain the key parameters, boundary conditions, and evaluation limits in the model. These methods easily rely on manual setting, expert experience, or pre-configuration, lacking an objective determination mechanism based on real driving data, scenario differences, and safety stability constraints. Therefore, when dealing with fuzzy traffic regulations, this type of method still suffers from subjective evaluation criteria and insufficient interpretability.

[0044] In summary, while existing technologies have studied the traffic compliance of autonomous driving systems from aspects such as violation scoring, automatic traffic rule determination, and rule engine mathematical evaluation, they still have the following shortcomings: First, most existing methods focus on the determination or scoring of explicit traffic rule clauses, lacking the introduction of evaluation indicators for vague traffic rules such as "maintaining a safe following distance" and "not affecting normal driving." Second, even when existing methods introduce evaluation indicators related to safety, stability, or behavioral characteristics, they are mostly used as evaluation inputs or rule parameters, but lack a method to objectively determine the limits of evaluation indicators for vague traffic rules based on real driving data. Third, existing methods usually treat explicit traffic rule evaluation, vague traffic rule evaluation, and behavioral normativity analysis separately, making it difficult to conduct a complete, unified, and interpretable comprehensive evaluation of the traffic compliance of autonomous driving systems in specific scenarios.

[0045] This application proposes a method 100 for evaluating the traffic compliance of autonomous driving. Figure 1 A schematic flowchart of an autonomous driving traffic compliance evaluation method 100 provided in an embodiment of this application is shown. Figure 1 As shown, method 100 includes steps S110 to S170. In method 100, by comparing and analyzing human driving data and autonomous driving data, the limit values ​​of fuzzy traffic regulation evaluation indicators are scientifically determined, solving the evaluation bias problem caused by the reliance on expert experience in traditional methods. Simultaneously, by integrating fuzzy traffic regulation scores, explicit traffic regulation scores, and behavioral normativity scores, a comprehensive and objective evaluation of the degree of compliance with traffic regulations by autonomous driving is achieved, providing an interpretable scientific basis for system verification.

[0046] Step S110: Obtain the human driving dataset and the autonomous driving dataset, and extract data from lane change scenarios.

[0047] Specifically, the human driving dataset includes driving data of multiple first vehicles driven by humans in lane-changing scenarios, and the autonomous driving dataset includes driving data of multiple second vehicles driven autonomously in lane-changing scenarios. Therefore, embodiments of this application can extract driving data in lane-changing scenarios based on the human driving dataset and the autonomous driving dataset. A lane-changing scenario can refer to the operation process of a vehicle changing lanes in a specific road environment, such as merging into an urban expressway ramp or overtaking on a highway. For example, the human driving dataset can be collected from vehicle driving data recorded by a real road monitoring system, and the autonomous driving dataset can originate from vehicle driving data of autonomous vehicles in closed areas or real open roads. In embodiments of this application, driving data can include dynamic information such as vehicle trajectory, speed, acceleration, and position coordinates, as well as static information such as road type and lane line position.

[0048] Step S120: Extract the traffic rule text for the lane change scenario, and parse the traffic rule text based on the large language model to obtain the temporal logic expression corresponding to the traffic rule text.

[0049] The traffic rule text includes both fuzzy and explicit texts. Specifically, fuzzy texts can refer to rule clauses containing subjective judgment requirements, such as maintaining a safe following distance as mentioned earlier. Explicit texts refer to rules with specific numerical requirements, such as speed limits. In the embodiments of this application, temporal logic expression refers to converting natural language rules into machine-processable mathematical expressions. For example, the rule that lane changes must not affect the normal driving of other vehicles can be converted into a temporal logic expression of the lane change time t∈[t_start,t_end] and the safe distance between the lane change time t and the vehicles in adjacent lanes.

[0050]

[0051] Optionally, in step S120, parsing the traffic rule text based on the large language model includes: decomposing the traffic rule text into multiple atomic propositions, which are used to indicate the behavioral themes, constraints, triggering conditions, and behavioral requirements in the traffic rule text; guiding the large language model to determine the logical relationships and temporal constraints in the traffic rule text based on templated prompts; and determining the temporal logical expression corresponding to the traffic rule text based on the multiple atomic propositions, the logical relationships, and the temporal constraints.

[0052] For example, the large language model decomposes the statement "lane changes must not affect the normal driving of other vehicles" into atomic propositions. The behavioral subject is the vehicle itself, the constraint object is vehicles in adjacent lanes, the trigger condition is the start of the lane change operation, and the behavioral requirement is that the longitudinal distance to vehicles in adjacent lanes is greater than a safety threshold. This ultimately generates a temporal logic expression such as the lane change time t∈[t_start,t_end] and the safe distance to vehicles in adjacent lanes:

[0053]

[0054] This scheme leverages the powerful semantic understanding capabilities of large language models to accurately transform natural language traffic regulations into computable temporal logic expressions, achieving precise conversion from natural language traffic regulations to computable models and laying the foundation for evaluation.

[0055] Step S130: Based on the temporal logic expression corresponding to the fuzzy traffic regulation text, determine the fuzzy traffic regulation evaluation index, and determine the limit value corresponding to the fuzzy traffic regulation evaluation index for the vehicle under test.

[0056] Specifically, the limit value refers to the threshold standard used to determine whether autonomous driving behavior complies with fuzzy traffic regulations. It should be understood that, in the embodiments of this application, the fuzzy traffic regulation evaluation index refers to converting subjective rules into measurable physical quantities, such as evaluating "changing lanes must not affect vehicles behind in adjacent lanes" as the minimum longitudinal distance between the vehicle and vehicles behind in adjacent lanes.

[0057] Optionally, in step S130, determining the limit value corresponding to the fuzzy traffic regulation evaluation index of the vehicle under test includes: determining the first driving behavior information of the plurality of first vehicles and their adjacent lane following vehicles based on the human driving dataset, and determining the second driving behavior information of the plurality of second vehicles and their adjacent lane following vehicles based on the autonomous driving dataset, wherein the first driving behavior information includes the speed information of the adjacent lane following vehicles of the first vehicle, and the second driving behavior information includes the speed information of the adjacent lane following vehicles of the second vehicle; grouping the plurality of first vehicles into multiple speed ranges based on the first driving behavior information to obtain multiple first vehicle speed groups, and grouping the plurality of second vehicles into multiple speed ranges based on the second driving behavior information to obtain multiple second vehicle speed groups, wherein the... Multiple first vehicle speed groups and multiple second vehicle speed groups are one-to-one. The speed interval of each first vehicle speed distribution in the multiple first vehicle speed groups is the same as the speed interval of the corresponding second vehicle speed group. Multiple first limits of fuzzy traffic rules corresponding to the multiple first vehicle speed groups are determined, and multiple second limits of fuzzy traffic rules corresponding to the multiple second vehicle speed groups are determined. The multiple first limits and the multiple second limits are one-to-one corresponding to the multiple speed intervals. Each of the multiple first limits and each of the corresponding second limits are weighted and fused to obtain multiple fused limits of fuzzy traffic rules corresponding to the multiple speed intervals. Each of the multiple fused limits is used to constrain the driving behavior of vehicles in the speed interval corresponding to each fused limit.

[0058] Specifically, the multiple first vehicle speed groups and the multiple second vehicle speed groups correspond one-to-one, and the speed range of each first vehicle speed distribution in the multiple first vehicle speed groups is the same as the speed range of the corresponding second vehicle speed group. For example, the speed range can be divided into multiple groups such as low speed (e.g., 0-30km / h), medium speed (e.g., 30-60km / h), and high speed (e.g., 60-90km / h), which can correspond to scenarios such as urban congestion, ordinary roads, and highways, respectively. The multiple first limits and the multiple second limits correspond one-to-one with the multiple speed ranges. It should be understood that the limit refers to the evaluation boundary such as the minimum safe distance or the maximum permissible speed difference to meet traffic rule compliance. For example, for the 60-90km / h speed group, the minimum safe distance from the following vehicle when changing lanes can be determined to be 50 meters.

[0059] Specifically, the first driving behavior information includes the speed information of vehicles following in adjacent lanes of the plurality of first vehicles, and the second driving behavior information includes the speed information of vehicles following in adjacent lanes of the plurality of second vehicles. The vehicle following in an adjacent lane refers to the closest vehicle behind the target vehicle in the target lane when the target vehicle changes lanes. In embodiments of this application, the speed information includes parameters such as instantaneous speed, average speed, and rate of change of speed. For example, when a vehicle intends to change lanes to the left, the vehicle closest to the vehicle in the left rear lane is the vehicle following in the adjacent lane, and its speed information will be recorded in real time. For example, the first driving behavior information also includes the speed, minimum acceleration, average speed of the plurality of first vehicles, the minimum acceleration of vehicles following in adjacent lanes, the average speed of vehicles following in adjacent lanes, longitudinal distance, lane change time, TTC, and THW of the plurality of second vehicles. Here, TTC can refer to the estimated collision time while both vehicles are in their current state of motion, and THW can refer to the time required for the vehicle to reach the position of the preceding vehicle while maintaining the current speed difference. Furthermore, this can cover vehicle dynamics and interaction timing features, providing multi-dimensional input features for subsequent Gaussian mixture models.

[0060] Optionally, in step S130, based on the first driving behavior information and KDE, the multiple first vehicles can be grouped into multiple speed ranges to obtain multiple first vehicle speed groups, and the statistical significance of the constraint index of the multiple first vehicle speed groups can be verified by at least one testing method. Furthermore, based on the second driving behavior information and KDE, the multiple second vehicles can be grouped into multiple speed ranges to obtain multiple second vehicle speed groups, and the statistical significance of the constraint index of the multiple second vehicle speed groups can be verified by at least one testing method. This scheme utilizes KDE to accurately capture speed distribution characteristics and ensures the scientific validity of the grouping through statistical testing. The detailed process of speed grouping will be described in the following embodiments, and will not be repeated here. This scheme uses kernel density estimation to perform nonparametric modeling of the speed of vehicles following in adjacent lanes, and can determine how to group speeds based on the speed distribution of vehicles following in adjacent lanes, rather than using existing fixed grouping methods. Furthermore, the combination of statistical significance testing ensures the scientific reliability of the grouping results.

[0061] Optionally, in step S130, a Generative Model (GMM) is used for modeling within the plurality of first vehicle speed groups and the plurality of second vehicle speed groups, respectively. The parameters of the GMM are determined by the maximum likelihood estimation method, and the optimal number of clusters in the GMM is determined by the Biological Identifier (BIC). Then, constraint index thresholds corresponding to the plurality of first vehicle speed groups and the plurality of second vehicle speed groups can be determined based on the GMM and the boundary points based on posterior probability. Next, multiple candidate limits are determined for each first vehicle speed group and each second vehicle speed group. Then, the recall rate and free-float probability (FPR) corresponding to the multiple candidate limits are determined, and the plurality of first limits and the plurality of second limits are selected based on the recall rate and FPR corresponding to the multiple candidate limits and the constraint index thresholds. This scheme controls the false positive rate while ensuring safety, achieving a balance between safety and behavioral norms. It should be understood that recall rate refers to the proportion of correctly identified dangerous scenarios, and FPR refers to the proportion of false positives that classify safe scenarios as dangerous. The following embodiments will describe this process in detail, and will not be repeated here.

[0062] Optionally, in step S130, specifically, each of the multiple fusion limits is used to constrain the driving behavior of vehicles within the speed range corresponding to each fusion limit. It should be understood that the fusion limit will serve as the decision threshold for the autonomous driving system regarding fuzzy traffic rules. For example, for the 30-60 km / h group, the limit of 40 meters generated from human driving data and the limit of 35 meters generated from autonomous driving data are weighted and fused, ultimately outputting a fusion limit of 37 meters.

[0063] Optionally, in step S130, a DMCI evaluation value can be determined based on each of the plurality of first limits and each corresponding second limit. Then, based on the DMCI evaluation value, a first weight for each of the plurality of first limits and a second weight for each corresponding second limit are determined. Finally, the plurality of fusion limits are determined based on the plurality of first limits and their first weights, and the plurality of second limits and their second weights. This scheme utilizes DMCI to dynamically adjust weights, ensuring the robustness and universality of the fusion limits. It should be understood that DMCI refers to Distribution Matching Consistency Index, used to evaluate the degree of difference in behavioral distributions between two datasets. For example, when the distribution of autonomous driving behavior differs significantly from that of human driving within a certain speed group, the DMCI evaluation value decreases. In this case, the weight of the autonomous driving limit is reduced to 0.3, and the weight of the human driving limit is increased to 0.7, making the fusion result more suitable for autonomous driving characteristics.

[0064] Step S140: Determine the fuzzy traffic rule score of the vehicle under test based on the limit value and the actual value corresponding to the vehicle under test under the fuzzy traffic rule evaluation index.

[0065] Specifically, the fuzzy traffic rule score refers to the evaluation value of the degree to which the autonomous driving behavior conforms to fuzzy traffic rules, such as calculating the score based on the ratio of the actual safe distance to the limit. For example, the fuzzy traffic rule evaluation index includes the safe distance between the vehicle and the vehicle behind in the adjacent lane, and the actual value corresponding to the test vehicle under the fuzzy traffic rule evaluation index is the actual distance between the test vehicle and the vehicle behind in the adjacent lane. Furthermore, step S140 can determine the fuzzy traffic rule score as 1 (actual value is greater than or equal to the limit) based on the magnitude of the actual value and the limit, or determine the fuzzy traffic rule score as the ratio of the actual value of the evaluation index to the limit (actual value is less than the limit).

[0066] Step S150: Based on the temporal logic expression corresponding to the explicit traffic regulation text, determine the explicit traffic regulation evaluation index and determine the explicit traffic regulation score of the vehicle under test.

[0067] Specifically, the traffic regulation evaluation indicators are defined as rule-based evaluation indicators with clear legal boundaries that can directly determine whether a violation has occurred. These include, but are not limited to, traffic marking prohibition evaluation indicators, speed limit evaluation indicators, and lane change prohibition evaluation indicators for specific road sections.

[0068] Optionally, in step S150, this embodiment of the application may use a binary scoring method to determine the score of a single explicit traffic rule evaluation indicator. For example, when a single evaluation indicator meets the rule, the score is 1; when a single evaluation indicator violates the rule, the score is 0. Further, this embodiment of the application may assign corresponding weights according to the importance of different explicit traffic rule evaluation indicators, and perform a weighted summation of the scores of each explicit traffic rule evaluation indicator to obtain the explicit traffic rule score in the lane-changing scenario.

[0069] Step S160: Determine the behavioral normativity score of the vehicle under test.

[0070] Specifically, behavioral normative evaluation indicators are used to characterize the degree of standardized implementation of lane-changing behavior by autonomous vehicles. These indicators reflect the constraints imposed by road traffic regulations on driving operations and can comprehensively reflect the smoothness, timing rationality, execution stability, and degree of disturbance to surrounding traffic participants in lane-changing behavior. These behavioral normative evaluation indicators include, but are not limited to, lane-changing duration, longitudinal acceleration, lateral acceleration, longitudinal jerk, lateral jerk, acceptable clearance in the target lane, and speed change of following vehicles in the target lane.

[0071] Optionally, in embodiments of this application, the behavioral normativity score can be obtained by normalizing and weighting the behavioral normativity evaluation indicators. For example, for evaluation indicators such as longitudinal acceleration, lateral acceleration, longitudinal jerk, and lateral jerk, where "smaller values ​​are better," a linear decay function based on an upper threshold is used to calculate the score of each individual evaluation indicator. As another example, for evaluation indicators such as acceptable gap, where "larger values ​​are better," a linear proportional function based on a threshold is used to calculate the score of each individual evaluation indicator. Furthermore, for evaluation indicators such as lane change duration, which have normative intervals, a piecewise interval function is used for scoring. Further, weights are assigned according to the importance of each behavioral normativity evaluation indicator, and the scores of each individual evaluation indicator are weighted and summed to obtain the behavioral normativity score.

[0072] Step S170: Determine the traffic regulation compliance evaluation result of the vehicle under test based on the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score.

[0073] Specifically, the evaluation result is a comprehensive evaluation index formed by weighted fusion of three types of scores, for example, calculating the total score with weights of 40% for fuzzy traffic regulations, 30% for explicit traffic regulations, and 30% for standardization. In the embodiments of this application, the weight corresponding to each part can be determined based on the entropy weight method or the expert experience method.

[0074] This application addresses the challenge of objectively determining the evaluation index limits for fuzzy traffic regulations through comparative analysis of driving data based on human driving data and autonomous driving data. Simultaneously, by employing a fusion evaluation mechanism of three types of evaluation indicators, it achieves a scientific measurement of the complete compliance with traffic regulations, avoiding the bias inherent in evaluations based on a single indicator and improving the completeness, objectivity, and interpretability of the evaluation results. Furthermore, this application utilizes techniques such as temporal logic transformation, dynamic grouping limits, and probabilistic model optimization to achieve precise quantification of complex traffic rules, thereby enhancing the efficiency and reliability of autonomous driving system verification.

[0075] The following is an introduction Figure 1Detailed embodiments of each step are shown below. Specifically, in the embodiments of this application, lane-changing scenario data are first extracted from human driving datasets and autonomous driving datasets respectively to obtain lane-changing behavior characteristics under different driving modes. Next, all traffic rules are extracted from the scenarios, and the traffic rules expressed in natural language are used to perform temporal logic modeling using a large language model to form a rule representation that can be computed by the autonomous driving system. Then, based on the temporal logic expression, fuzzy traffic rule evaluation indicators for lane-changing scenarios are determined, and the limits of the fuzzy traffic rule evaluation indicators are determined by combining data analysis and modeling methods, and the fuzzy traffic rule score is determined. Then, based on the temporal logic expression, explicit traffic rule evaluation indicators for lane-changing scenarios are determined, and multiple explicit traffic rule evaluation indicators are weighted and summed to determine the explicit traffic rule score. Next, behavioral compliance evaluation indicators for lane-changing scenarios are determined, and the behavioral compliance score is determined through analysis and weighted calculation among the evaluation indicators. Finally, the fuzzy traffic rule score, explicit traffic rule score, and behavioral compliance score are weighted and summed to obtain the comprehensive evaluation result of the traffic rule compliance of the lane-changing behavior of the vehicle under test.

[0076] Figure 2 A flowchart illustrating the traffic compliance evaluation process for autonomous vehicles provided in this application embodiment is shown.

[0077] refer to Figure 2 ,like Figure 2 As shown, firstly, corresponding to step S110, the original autonomous driving and human driving datasets are acquired, and data in lane-changing scenarios are extracted. Furthermore, based on the temporal behavior modeling of lane-changing intentions, full-cycle scenario segments covering decision-making, execution, and stabilization are extracted, and structured data containing multi-dimensional interaction information between the vehicle and surrounding vehicles is obtained.

[0078] Figure 3 A schematic diagram of a lane-changing scenario provided by an embodiment of this application is shown. Figure 3 As shown, the three lanes traveling in the same direction are separated by white dashed lines, allowing vehicles to change lanes only when it is safe to do so. The vehicle (green) is located in the bottom lane and plans to change lanes from the right lane to the left (middle) lane to overtake. The lane change path is as follows: from its current position, it travels along an arc to the upper right, crosses the dashed lane line, enters the middle lane, and finally stops in front of the vehicle in the adjacent blue lane of the original middle lane. Before changing lanes, the vehicle is positioned slightly to the side and slightly in front of the vehicle in the adjacent lane; the lane change will cut into the normal driving path of the vehicle behind. During this lane change, it is necessary to ensure sufficient lateral and longitudinal safety distance from the vehicle behind, guaranteeing the following vehicle has ample reaction and braking space, and must not impede vehicles traveling normally in the adjacent lane.

[0079] For example, embodiments of this application can perform temporal behavior modeling in both autonomous driving datasets and human driving datasets to identify vehicle lane-changing intentions, shifting the extraction starting point to the early stages of driver or system decision-making. Subsequently, the extracted data... Figure 3 The lane-change scenario segment shown is defined as a time window covering the entire process of intent formation, lateral execution, and behavior stabilization. The extracted structured lane-change scenario can include multi-dimensional labeled information such as vehicle trajectory data (own vehicle and surrounding vehicles), intent origin, road topology, lane-change time, own vehicle speed and acceleration, speed and acceleration of following vehicles in adjacent lanes, and longitudinal distance during lane change.

[0080] For example, 10-dimensional feature vectors are extracted from the lane-changing vehicle and the following vehicle in the adjacent lane within the lane-changing driving segment, namely [vehicle speed, minimum acceleration of the lane-changing vehicle, average speed of the lane-changing vehicle, speed of the following vehicle in the adjacent lane, minimum acceleration of the following vehicle in the adjacent lane, average speed of the following vehicle in the adjacent lane, longitudinal distance, lane-changing time, TTC, THW]. This temporal modeling method can cover the entire cycle of lane changing, from the decision-making stage to the stable behavior chain, avoiding reliance solely on data from the execution stage, and comprehensively reflecting driving logic and behavioral consequences. This segment extraction method not only extracts the vehicle's trajectory but also continuously records the interaction with surrounding vehicles before and after the lane change.

[0081] For example, human driving and autonomous driving data can be constructed into trajectory segments in the form of time windows. Multidimensional features are extracted from each frame to form a time series, including information such as vehicle position, speed, and acceleration. Considering the extraction of scene segments throughout the entire cycle, a Long Short-Term Memory (LSTM) network is introduced for temporal behavior modeling, with the following formula:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088] In the formula, Forgotten Gate; For input gates; Candidate memory units; The state of the memory unit; For output gate; Output in hidden state; For the first The observation time series of the frames; This is the hidden state from the previous moment; The state of the memory unit from the previous moment; This is the Sigmoid function, with an output range of [0,1]. It is a hyperbolic tangent function with an output range of [-1, 1]. The weight matrix is ​​a learnable matrix; is a learnable bias matrix.

[0089] Furthermore, the LSTM temporal modeling used in the above methods can be replaced by Hidden Markov Model (HMM) temporal modeling. HMM modeling constructs the lane-changing behavior states and their state transition probabilities to perform temporal modeling of different behavioral stages during the lane-changing process. However, compared to LSTM, HMM has limited expressive power in capturing multi-factor interactions and nonlinear dynamic evolution in complex driving scenarios, especially its relatively weak ability to model long temporal dependencies and high-dimensional feature spaces.

[0090] Subsequently, corresponding to step S120, traffic rule text related to the lane-changing scenario can be extracted, and a large language model can be used to perform semantic parsing and temporal logic modeling on the traffic rules expressed in natural language. Furthermore, the traffic rules can be converted into a rule expression that can be computed by the autonomous driving system, providing a basis for the subsequent extraction of fuzzy and explicit traffic rule evaluation indicators.

[0091] For example, using a large language model to perform temporal logic modeling on traffic rules: first, the traffic rules are decomposed into several atomic propositions to represent the behavioral themes, constraints, triggering conditions, and behavioral requirements in the rules; a templated Prompt is designed to guide the large language model to identify the logical relationships and temporal constraints of the traffic rules; Boolean operators and temporal operators are used to connect the atomic propositions to construct the corresponding temporal logic expressions.

[0092] The expression for measuring sequential logic can be shown below:

[0093]

[0094] In the formula, All of these are atomic propositions; The phrase "defined as" can be understood as "equal to" or "equivalent to". To indicate "or", the right side of the equal sign in the above equation is indicated by... Separated expressions can all be considered equivalent to the expression on the left side of the equals sign; , , For the Boolean operators "NOT", "AND", and "OR"; , , , For the timing operators "global", "future", "past", and "once"; For a specific time interval, if If not explicitly specified, the default time is from the current time to the end time.

[0095] Next, corresponding to step S130, based on the temporal logic expression generated by the large language model, the fuzzy traffic rule evaluation index for lane-changing scenarios is determined. For example, based on the requirement that "on roads with two or more lanes in the same direction, a vehicle changing lanes shall not affect the normal driving of vehicles traveling in the relevant lanes," the requirement of "not affecting the normal driving of other vehicles" can be specifically evaluated as the safe distance from the vehicle behind in the adjacent lane. Based on this, the embodiments of this application can define the evaluation index (i.e., the fuzzy traffic rule evaluation index) corresponding to the clauses of fuzzy traffic rules as the safe distance from the vehicle behind in the adjacent lane, which is used to evaluate the degree of impact of the autonomous vehicle on the vehicle behind in the adjacent lane during the lane-changing process.

[0096] Furthermore, in extracting lane-change scene segments, the speed data of vehicles following in adjacent lanes can be processed. The KDE method can be used to perform nonparametric modeling of the speed distribution of vehicles following in adjacent lanes, obtaining smooth probability density curves to more clearly reveal the overall characteristics and multi-peak structure of the speed distribution. Based on this, to accurately characterize driving behavior patterns within different speed ranges, a speed grouping method based on distribution characteristics is proposed, dividing the samples into four regions: low speed, medium / high speed, and speeding.

[0097] For example, the median of the overall speed samples of vehicles following in adjacent lanes is used as the dividing point of the main density region, dividing the samples into left and right parts. For the left section, to isolate the influence of congestion / low-speed conditions on the behavior distribution, the left-tail speed samples are grouped separately; for the right section, considering that the speed limit has been exceeded, this part of the samples is grouped separately to distinguish the high-speed violation conditions. This step aims to solve the problem of high false alarm rate in low-speed scenarios and high false alarm rate in high-speed scenarios caused by the difference in speed in traditional single constraint index threshold models. The theoretical basis is that as the vehicle speed increases, the safety margin should increase accordingly, and the required safe distance between vehicles also increases. If speed grouping is not performed, the uniform constraint index threshold may be too lenient at low speeds and too strict at high speeds, failing to reasonably reflect driving behavior and safety requirements in different speed ranges.

[0098] For example, in this embodiment of the application, the speeds of adjacent vehicles in the above feature vectors can be used as samples to group speeds, employing KDE, as shown in the following formula:

[0099]

[0100] In the formula, For the velocity to be estimated, a value is assigned. For sample speed, For bandwidth.

[0101] Continue to refer to Figure 2 Furthermore, embodiments of this application can also construct constraint indicators based on the dual standards of "safety" and "stability," and verify the statistical significance of speed groupings using the Kruskal-Wallis H test and the Mann-Whitney U test to ensure that the differences in constraint indicators are statistically significant. "Safety" ensures that no collision occurs during lane changes, guaranteeing the traffic safety of the vehicle and surrounding vehicles; "stability" ensures that lane changes do not interfere with the normal driving of vehicles in adjacent lanes, maintaining their stability and thus avoiding impact on surrounding traffic. For example, the Kruskal-Wallis H test and the Mann-Whitney U test are used to verify the statistical significance of the constraint indicators (minimum acceleration of the following vehicle in the adjacent lane and THW) in the speed groupings, with the formula shown below:

[0102]

[0103]

[0104] In the formula, This represents the total number of samples; Number of groups; For the first Group sample size; For the first The rank sum of the groups; The number of samples in any two groups (groups 1 and 2 or groups 1 and 3) are respectively. Let be the rank sum of sample 1.

[0105] Furthermore, to assess whether there are significant differences in the dual-standard constraint indices across different speed groups, this embodiment employs a phased nonparametric testing strategy. First, a Kruskal-Wallis H global test is performed on the constraint index distributions of each speed group to compare whether there are significant differences in the constraint indices between different speed groups. If the test result reaches a significant level (α=0.05), the null hypothesis is rejected, indicating a global difference in the constraint indices among the speed groups. Based on the significance of the global test, a Mann-Whitney U test is further used to perform pairwise comparisons of all speed groups to identify specific differences between groups. To effectively control the inflation of Type I errors caused by multiple comparisons, the Bonferroni correction method is used to adjust the significance level to ensure the reliability of the results. This step further demonstrates the rationality of the speed groupings, ensuring that the differences in the constraint indices evaluated across different speed ranges are statistically significant.

[0106] After grouping by speed, continue to refer to Figure 2This application embodiment can use Gaussian Model (GMM) combined with posterior probability analysis to adaptively determine the probability thresholds of dynamic constraint indicators for each speed group. For example, to determine the constraint indicator thresholds for different speed groups, a traffic safety assessment method combining GMM and posterior probability analysis is used. The formula for the GMM model is shown below:

[0107]

[0108] In the formula Number of clusters (candidates) ) For the first Component weights; For the first The mean and variance of a Gaussian distribution; x is the number of samples for the selected constraint index.

[0109] In embodiments of this application, the parameters of the GMM model can be learned based on the expectation-maximization (EM) algorithm. The EM algorithm optimizes the model parameters (including mean, covariance, and weights) iteratively to maximize the log-likelihood function of the data. For example, the parameters in the GMM are obtained through maximum likelihood estimation, i.e., the maximum likelihood function is:

[0110]

[0111] In the formula, .

[0112] Since the objective function is a hybrid form that can be solved directly. This is quite difficult, therefore the EM algorithm is used for iterative estimation. The posterior probability in the E-step is:

[0113]

[0114] The parameters updated in step M are:

[0115]

[0116] In the formula For the first The data points were observed. The EM was iterated alternately until convergence, and the GMM parameters were obtained.

[0117] Then, the optimal number of GMM clusters is automatically selected using BIC to ensure that the model achieves the best fitness and accuracy when fitting the data, avoiding overfitting or underfitting problems caused by inappropriate cluster selection. After fitting, the posterior probability of each sample is calculated, and an impact probability threshold (P=0.5) is set based on this to determine whether a sample belongs to the potentially risky "impact class". The formula is as follows:

[0118]

[0119] In the formula, For the maximum likelihood of the K-component model, The number of model parameters is given. The K that minimizes BIC(K) is ultimately selected as the optimal mixing score for that velocity group.

[0120] Finally, the threshold of the constraint index for each speed group is determined based on the posterior probability cutoff point (p=0.5), and the formula is as follows:

[0121]

[0122] Continue to refer to Figure 2 This application also introduces a dynamic optimization method for distance limits based on dual constraints of target recall rate and false positive rate, adjusting the limits with the goal of minimizing the false positive rate. Specifically, this application employs a dynamic limit adjustment method based on dual constraint performance. By adaptively adjusting the distance limit under different speed groups, while meeting the preset recall rate requirement, the FPR is controlled below a predetermined threshold. Under these dual constraints, the minimum feasible limit that meets the conditions is selected, thereby avoiding the limitations of fixed limit settings in traditional static methods. Compared with traditional methods, this method can flexibly adjust the limit according to changes in actual data, improving the accuracy and applicability of traffic safety assessment, while ensuring effective reduction of false positives under the premise of high recall rate, enhancing the practicality and accuracy of the model.

[0123] For example, to determine the limits of the fuzzy traffic rule evaluation index under each speed group, the fuzzy traffic rule quantity under the lane change scenario is the limit of distance, and all distances in each speed group are extracted. Set the candidate limit set as a deduplicated sort of these distances: For a given candidate limit value thr, the prediction rule is as follows:

[0124]

[0125] In the formula This is an indicator function: 1 if the condition is true, 0 otherwise.

[0126] For each candidate limit value thr, Perform binary classification and calculate recall:

[0127]

[0128] In the formula, TP represents the actual impact and the predicted impact; FN represents the actual impact but the predicted non-impact (missed detection).

[0129] Calculate FPR:

[0130]

[0131] In the formula, FP represents cases where the actual data is unaffected but the prediction is affected (false positives); TN represents cases where neither the actual data nor the prediction is affected. Scan each value from smallest to largest. Calculate recall and FPR, and find the first condition that satisfies... And simultaneously satisfy To determine the limits of fuzzy traffic regulation evaluation indicators for different speed ranges.

[0132] Continue to refer to Figure 2 By comparing the DMCI with the safety margin dynamic adjustment factor (SMDAF) and fusing the limits from autonomous driving and human driving datasets, recommended safety standard limits with strong generalization ability are generated.

[0133] Specifically, a multi-index fusion comparison framework is adopted to systematically evaluate the distance limits obtained from the above steps for autonomous driving and human driving datasets. First, the DMCI (Driving Distance Index) is calculated for each speed group. An exponential decay function is used to evaluate the degree of agreement between the safe distance limits derived from the two driving modes, under the premise of achieving the same safety recall rate. Simultaneously, the SMDAF (Safety Distance Decision Adversarial Factor) is calculated, directly reflecting the conservative / aggressive tendency of the autonomous driving strategy relative to human driving. Based on the dynamic weights of the DMCI, the two types of limits (autonomous driving and human driving) are weighted and fused to generate data-driven recommended fused limits.

[0134] For example, DMCI is used to evaluate at the same safety level ( The formula for the similarity between the safety boundaries (distance limits) set by autonomous driving and human driving under these conditions is as follows:

[0135]

[0136] In the formula, For the first Automated driving safety distance limits under the speed group; For the first Human driving safety distance limits under the speed group; The standardization factor is usually set to 1 to adjust for sensitivity; DMCI≈1 indicates that the limits are highly consistent.

[0137] AMDAF directly measures the degree of conservatism of autonomous driving relative to human driving, as shown in the formula below:

[0138]

[0139] If SMDAF < 1, it indicates that autonomous driving is more conservative; otherwise, it indicates that autonomous driving is more lenient; if SMDAF = 1, the margin settings are completely consistent.

[0140] Furthermore, based on the limits for autonomous driving and human driving under different speed groups, the dynamic weights of DMCI are used for weighted fusion to generate the data-driven fusion limits shown below:

[0141]

[0142] In the formula, the weights It is not a fixed value, but is dynamically determined by DMCI, such as:

[0143]

[0144] In the formula, k is a weight-sensitive adjustment factor, which is usually greater than 1. At a certain speed, if the DMCI is very high (close to 1), it means that the two strategies are consistent and the limit for autonomous driving can be adopted; if the DMCI is very low (close to 0), it means that the difference is huge, and at this time, we should rely more on the experience of the human driving model and adopt the human driving limit with stronger interpretability.

[0145] Finally, continue to refer to Figure 2 Corresponding to step S140, a linear mapping method can be used to calculate the fuzzy traffic regulation score. Specifically, the actual value of the fuzzy traffic regulation evaluation index is compared with the corresponding limit value of the fuzzy traffic regulation evaluation index: when the actual value is greater than or equal to the limit value, the score is 1, indicating that it fully complies with the provisions of the corresponding fuzzy traffic regulation; when the actual value is lower than the limit value, the score is linearly mapped to a range of 0 to 1 according to the ratio of the actual value to the limit value, that is, the score is equal to the actual value divided by the limit value, thereby quantitatively reflecting the degree to which the actual behavior deviates from the safety limit value.

[0146] For example, the fuzzy traffic rule score is calculated using a linear mapping method, and the calculation process is shown in the following formula:

[0147]

[0148] In the formula The actual values ​​of the fuzzy traffic regulation evaluation indicators; Limits for fuzzy traffic regulation evaluation indicators; To assign points to ambiguous traffic regulations.

[0149] In other embodiments of this application, the linear mapping method used in the above-described fuzzy traffic rule scoring can be replaced by a piecewise function for calculation, as shown in the following formula:

[0150]

[0151] The piecewise function method pre-divides several scoring intervals based on the relative relationship between the actual value and the limit value of the fuzzy traffic regulation evaluation index, and assigns a fixed score to each interval. Compared with the linear mapping method, the piecewise function method uses a discrete interval scoring approach, which is prone to abrupt score changes near the interval boundaries, making it difficult to continuously reflect subtle changes in the actual value of the fuzzy traffic regulation evaluation index.

[0152] Next, corresponding to step S150, a binary scoring method is used to determine the score of each explicit traffic rule evaluation indicator: when an individual evaluation indicator meets the rule, the score is 1; when an individual evaluation indicator violates the rule, the score is 0. Further, corresponding weights are assigned according to the importance of different explicit traffic rule evaluation indicators, and the scores of each explicit traffic rule evaluation indicator are weighted and summed to obtain the explicit traffic rule score in the lane-changing scenario.

[0153] For example, a single explicit traffic regulation evaluation indicator can be directly evaluated using a binary scoring method, as shown in the following formula:

[0154]

[0155] For the overall clear traffic rule score, it can be used Calculate and satisfy: . The weights of each specific traffic regulation are determined using the entropy weight method based on the importance of the evaluation indicators for different specific traffic regulations.

[0156] Corresponding to step S160, in the lane-changing scenario, behavioral normativity evaluation indicators are used to characterize the degree of standardized implementation of lane-changing behavior by autonomous vehicles. These indicators reflect the constraints imposed by road traffic regulations on driving operations and can comprehensively reflect the smoothness, timing rationality, execution stability, and degree of disturbance to surrounding traffic participants in the vehicle's lane-changing behavior. The behavioral normativity evaluation indicators include, but are not limited to, lane-changing duration, longitudinal acceleration, lateral acceleration, longitudinal jerk, lateral jerk, acceptable clearance in the target lane, and the change in speed of following vehicles in the target lane.

[0157] For example, the score of a single normative evaluation indicator is calculated by dividing it into three categories according to the type of evaluation indicator, where let the first category be... The actual value of each behavioral normative evaluation indicator is The corresponding score is .

[0158] The normative evaluation index that "the smaller the value, the better" can be represented by a linear decay function based on an upper threshold, as shown in the following formula:

[0159]

[0160] In the formula This is the upper limit threshold of the evaluation indicator;

[0161] For evaluation indicators where "the larger the value, the better," a threshold-based linear proportional function can be used, as shown in the following formula:

[0162]

[0163] In the formula This is the lowest reasonable threshold for this evaluation indicator;

[0164] For standardized interval-type evaluation indicators, piecewise functions can be used, as shown in the following formula:

[0165]

[0166] The normed interval is: The overall normative evaluation index score can be used... Calculate and satisfy . The weights of each normative evaluation indicator can be determined using the entropy weight method.

[0167] Behavioral conduct scores can be obtained by normalizing and weighting the behavioral conduct evaluation indicators. For example, for evaluation indicators such as longitudinal acceleration, lateral acceleration, longitudinal jerk, and lateral jerk, where "smaller values ​​are better," a linear decay function based on an upper threshold is used to calculate the score for each indicator. For evaluation indicators such as acceptable gaps, where "larger values ​​are better," a linear proportional function based on a threshold is used to calculate the score for each indicator. For evaluation indicators such as lane change duration, which have a standardized range, a piecewise interval function is used for scoring. Furthermore, weights are assigned according to the importance of each behavioral conduct evaluation indicator, and the scores of each indicator are weighted and summed to obtain the behavioral conduct score.

[0168] In other embodiments of this application, the method described above for calculating scores based on a single normative evaluation index, categorized into three types according to the evaluation index, can be replaced by a uniform distance-based scoring method, with the calculation formula as follows:

[0169]

[0170] In the formula The actual value of a single evaluation indicator; This serves as a reference value for the evaluation indicator; The allowable deviation range; This is a single evaluation indicator score. Compared to calculating single evaluation indicator scores by category, the unified distance model scoring cannot fully reflect the inherent attributes of each evaluation indicator.

[0171] Finally, corresponding to step S170, the fuzzy traffic rule score, explicit traffic rule score, and behavioral normativity score are weighted and fused to obtain the comprehensive traffic rule compliance score of the autonomous vehicle under test in the lane changing scenario.

[0172] For example, the fuzzy traffic regulation compliance score, the explicit traffic regulation compliance score, and the behavioral normativity score are weighted and fused to obtain the comprehensive traffic regulation compliance score of the autonomous vehicle in the lane-changing scenario. The formula for calculating the comprehensive score is as follows:

[0173]

[0174] In the formula The overall score is based on compliance with traffic regulations. To score for ambiguous traffic regulations; To clarify the points for traffic regulations; Scoring based on behavioral norms; , and These are the weights corresponding to the scores, and satisfy... .

[0175] The weights mentioned above are all determined using the entropy weight method. Specifically, let there be a total of A lane change sample, constructing a... The original evaluation matrix is ​​composed of Then, the original evaluation matrix is ​​normalized to obtain a normalized matrix. Furthermore, calculate the first... The evaluation results of the class are in the first The proportion in each sample And calculate the first Information entropy of class evaluation results Furthermore, based on: Calculate the first The coefficient of difference of the evaluation results is ultimately based on get The corresponding weights.

[0176] In the aforementioned methods for determining weights, the entropy weight method can be replaced by the expert experience method. The expert experience method involves experts in fields such as traffic safety and autonomous driving testing setting weights based on the importance of each evaluation indicator. Compared to the entropy weight method, the expert experience method relies heavily on the subjective judgment of experts, and differences in evaluation may exist between different experts, thus easily affecting the consistency and objectivity of weight allocation.

[0177] Ultimately, in this embodiment, the scenario data extraction method based on full-cycle lane-change time-series modeling can cover the entire process of lane-change behavior, from initial decision-making to final stable lane-change completion, comprehensively capturing all key stages in the lane-change process. Furthermore, this not only ensures that the extracted data covers all time periods of lane changes but also provides complete and continuous behavioral information, providing sufficient basis for subsequent analysis and decision-making. This comprehensive and high-quality scenario data extraction method can effectively reflect the dynamic evolution of lane-change behavior, enhancing the model's adaptability to actual traffic environments and its predictive accuracy.

[0178] Furthermore, a large language model is used to perform semantic parsing and temporal logic modeling of traffic rules expressed in natural language. Then, leveraging the powerful semantic understanding capabilities of the large language model, the natural language traffic regulations can be accurately converted into computable temporal logic expressions, providing a foundation for subsequent evaluation.

[0179] Subsequently, based on the significance tests of the Kruskal-Wallis H and Mann-Whitney U methods, not only is the speed grouping reasonable, but it also provides a statistically significant scientific basis for the analysis of the differences in constraint indicators under different speed ranges. Combined with the Gaussian mixture model, the threshold values ​​of constraint indicators under different speed groups can be realized, thus eliminating the limitations of a single fixed value.

[0180] Then, by using a distance limit determination method based on dual constraints of target recall rate and FPR, the limit can be flexibly adjusted according to changes in actual data, solving the problem that fixed limits become invalid due to changes in scenarios. This enables the system to adapt to different road conditions and traffic flow states, greatly improving the accuracy and applicability of traffic safety assessment.

[0181] Then, the method of cross-dataset limit comparison analysis and fusion based on DMCI and SMDAF solves the narrowness of fuzzy traffic regulations no longer being limited to modeling a single dataset, but rather a limit calculation method that integrates multiple datasets. It draws on the commonalities of driving behavior in a wider range of scenarios, greatly improving the practicality of the technology.

[0182] Finally, a comprehensive evaluation model was constructed that integrates explicit traffic rule evaluation indicators, fuzzy traffic rule evaluation indicators, and behavioral normative evaluation indicators. By modeling, determining limits, and calculating scores for fuzzy traffic rule clauses, and then weighting and fusing the resulting fuzzy traffic rule evaluation results, explicit traffic rule judgment results, and behavioral normative evaluation results, a comprehensive evaluation of the degree of traffic rule compliance of autonomous vehicles in lane-changing scenarios was conducted, improving the completeness, objectivity, and interpretability of the evaluation results.

[0183] This application also provides a computer program product that, when run on an electronic device, causes the electronic device to execute the technical solutions described in the above embodiments. Its implementation principle and technical effects are similar to those of the related embodiments described above, and will not be repeated here.

[0184] This application also provides a readable storage medium containing instructions that, when executed by an electronic device, cause the electronic device to perform the technical solutions described in the above embodiments. The implementation principle and technical effects are similar and will not be repeated here.

[0185] This application also provides a chip for executing instructions. When the chip is running, it executes the technical solutions described in the above embodiments. Its implementation principle and technical effects are similar and will not be repeated here.

[0186] The hardware module of this application is described below, which can be used to implement the aforementioned method 100.

[0187] Now for reference Figure 4 The diagram shows a block diagram of a device 400 according to one embodiment of this application. Device 400 may include one or more processors 401 coupled to a controller hub 403. In at least one embodiment, the controller hub 403 communicates with the processor 401 via a multi-branch bus such as a front side bus (FSB), a point-to-point interface such as a quickpath interconnect (QPI), or a similar connection 410. The processor 401 executes instructions controlling general types of data processing operations. In one embodiment, the controller hub 403 includes, but is not limited to, a graphics memory controller hub (GMCH) (not shown) and an input / output hub (IOH) (which may be on a separate chip) (not shown), wherein the GMCH includes memory and a graphics controller and is coupled to the IOH.

[0188] Device 400 may also include a coprocessor 402 and a memory 404 coupled to a controller hub 403. Alternatively, one or both of the memory and the GMCH may be integrated within the processor, with memory 404 and coprocessor 402 directly coupled to processor 401 and controller hub 403, which resides on a single chip with the IOH. Memory 404 may be dynamic random access memory (DRAM), phase change memory (PCM), or a combination of both. In one embodiment, coprocessor 402 is a dedicated processor, such as, for example, a high-throughput MIC processor (many integrated core, MIC), a network or communication processor, a compression engine, a graphics processor, a general-purpose computing on GPU (GPGPU), or an embedded processor, etc. Optional properties of coprocessor 402 are indicated by dashed lines. Figure 4 middle.

[0189] Memory 404, as a computer-readable storage medium, may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. For example, memory 404 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device such as one or more hard-disk drives (HDD(s)), one or more compact disc (CD) drives, and / or one or more digital versatile disc (DVD) drives.

[0190] In one embodiment, device 400 may further include a network interface controller (NIC) 406. NIC 406 may include a transceiver for providing a radio interface to device 400, thereby enabling communication with any other suitable device, such as a front-end module, antenna, etc. In various embodiments, NIC 406 may be integrated with other components of device 400. NIC 406 can implement the functionality of the communication unit in the above embodiments.

[0191] Device 400 may further include input / output (I / O) device 405. I / O 405 may include: a user interface designed to enable a user to interact with device 400; a peripheral component interface designed to enable peripheral components to also interact with device 400; and / or sensors designed to determine environmental conditions and / or location information related to device 400.

[0192] It is worth noting that, Figure 4 This is merely an example. That is, although... Figure 4 The diagram shows that device 400 includes multiple devices such as processor 401, controller hub 403, and memory 404. However, in actual applications, devices using the methods of this application may include only a portion of the devices in device 400. For example, it may include only processor 401 and NIC 406. Figure 4 The properties of the optional devices are shown in dashed lines. According to some embodiments of this application, the memory 404, which is a computer-readable storage medium, stores instructions that, when executed on a computer, cause the device 400 to perform the methods according to the above embodiments. Specific details can be found in the methods of the above embodiments, and will not be repeated here.

[0193] Now for reference Figure 5 The diagram shown is a block diagram of a SoC 500 according to an embodiment of this application. Figure 5 In the diagram, similar components share the same reference numerals. Additionally, dashed boxes are an optional feature for more advanced SoCs. Figure 5 In this SoC 500, the following are included: an interconnect unit 550 coupled to an application processor 510; a system proxy unit 580; a bus controller unit 590; an integrated memory controller unit 540; a group or one or more coprocessors 520, which may include integrated graphics logic, an image processor, an audio processor, and a video processor; a static random access memory (SRAM) unit 530; and a direct memory access (DMA) unit 560. In one embodiment, the coprocessor 520 includes a dedicated processor, such as, for example, a network or communication processor, a compression engine, a GPGPU, a high-throughput MIC processor, or an embedded processor.

[0194] The static random-access memory (SRAM) unit 530 may include one or more computer-readable media for storing data and / or instructions. The computer-readable storage medium may store instructions, specifically, temporary and permanent copies of those instructions. These instructions may include, when executed by at least one unit in the processor, causing the SoC 500 to perform the autonomous driving traffic compliance evaluation method according to the above embodiments, as detailed in the methods described above, which will not be repeated here.

[0195] Various embodiments of the mechanisms disclosed in this application can be implemented in hardware, software, firmware, or combinations of these implementation methods. Embodiments of this application can be implemented as computer programs or program code executable on a programmable system, the programmable system including at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0196] Program code can be applied to input instructions to execute the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, the processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application-specific integrated circuit (ASIC), or a microprocessor.

[0197] The program code can be implemented using a high-level procedural language or an object-oriented programming language to communicate with the processing system. Assembly language or machine language can also be used when needed. In fact, the mechanisms described in this application are not limited to any particular programming language. In either case, the language can be a compiled language or an interpreted language.

[0198] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored thereon on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, the instructions may be distributed via a network or through other computer-readable media. Therefore, machine-readable media may include any mechanism for storing or transmitting information in a machine-readable (e.g., computer-readable) form, including but not limited to floppy disks, optical disks, CD-ROMs, compact disc read-only memory (CD-ROMs), magneto-optical disks, read-only memory (ROM), random-access memory (RAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic cards or optical cards, flash memory, or tangible machine-readable storage for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) using the Internet in the form of electrical, optical, acoustic, or other forms of propagated signals. Therefore, machine-readable media includes any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a machine-readable (e.g., computer-readable) form.

[0199] In the accompanying drawings, some structural or methodological features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. Rather, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the accompanying drawings. Furthermore, including structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0200] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0201] It should be noted that in the examples and description of this patent, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0202] Although this application has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A method for evaluating the traffic compliance of automated driving systems, characterized in that, include: Acquire human driving datasets and autonomous driving datasets and extract data in lane change scenarios, wherein the human driving dataset includes driving data of multiple first vehicles driven by humans in the lane change scenario, and the autonomous driving dataset includes driving data of multiple second vehicles driven by autonomous vehicles in the lane change scenario. The traffic rule text in the lane-changing scenario is extracted, and the traffic rule text is parsed based on a large language model to obtain the temporal logic expression corresponding to the traffic rule text. The traffic rule text includes fuzzy traffic rule text and explicit traffic rule text. Based on the temporal logic expression corresponding to the fuzzy traffic regulation text, a fuzzy traffic regulation evaluation index is determined, and the limit value corresponding to the fuzzy traffic regulation evaluation index of the vehicle under test is determined, wherein the vehicle under test is an autonomous vehicle. Based on the limit value and the actual value corresponding to the vehicle under the fuzzy traffic rule evaluation index, the fuzzy traffic rule score of the vehicle under test is determined. Based on the temporal logic expression corresponding to the explicit traffic regulations text, the explicit traffic regulations evaluation indicators are determined, and the explicit traffic regulations score of the vehicle under test is determined. Determine the behavioral normativity score of the vehicle under test; The traffic regulation compliance evaluation result of the vehicle under test is determined based on the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score.

2. The method according to claim 1, characterized in that, The process of parsing the traffic rule text based on a large language model to obtain the temporal logic expression corresponding to the traffic rule text includes: The traffic rule text is decomposed into multiple atomic propositions, which are used to indicate the behavioral themes, constraints, triggering conditions, and behavioral requirements in the traffic rule text. Based on templated prompts, the large language model is guided to determine the logical relationships and temporal constraints in the traffic rule text. Based on the multiple atomic propositions, the logical relationships, and the timing constraints, the timing logic expression corresponding to the traffic rule text is determined.

3. The method according to claim 1 or 2, characterized in that, The fuzzy traffic regulation evaluation index includes the safe distance between a vehicle and a vehicle following in an adjacent lane. The actual value corresponding to the vehicle under the fuzzy traffic regulation evaluation index is the actual distance between the vehicle under the test and the vehicle following in an adjacent lane.

4. The method according to claim 1 or 2, characterized in that, The step of determining the limit value corresponding to the fuzzy traffic regulation evaluation index of the vehicle under test includes: Based on the human driving dataset, first driving behavior information of the plurality of first vehicles and their adjacent lane following vehicles is determined, and second driving behavior information of the plurality of second vehicles and their adjacent lane following vehicles is determined based on the autonomous driving dataset, wherein the first driving behavior information includes the speed information of the adjacent lane following vehicles of the first vehicle, and the second driving behavior information includes the speed information of the adjacent lane following vehicles of the second vehicle. Based on the first driving behavior information, the plurality of first vehicles are grouped into multiple speed ranges to obtain multiple first vehicle speed groups; and based on the second driving behavior information, the plurality of second vehicles are grouped into multiple speed ranges to obtain multiple second vehicle speed groups. The plurality of first vehicle speed groups and the plurality of second vehicle speed groups correspond one-to-one, and the speed range in which the speed distribution of each first vehicle in the plurality of first vehicle speed groups is located is the same as the speed range in which the corresponding second vehicle speed group is located. Determine multiple first limits of fuzzy traffic regulations corresponding to the multiple first vehicle speed groups, and determine multiple second limits of fuzzy traffic regulations corresponding to the multiple second vehicle speed groups, wherein the multiple first limits and the multiple second limits correspond one-to-one with the multiple speed intervals; Each of the plurality of first limits is weighted and fused with each corresponding second limit to obtain a plurality of fused limits for the fuzzy traffic regulations corresponding to the plurality of speed ranges. Each of the plurality of fused limits is used to constrain the driving behavior of vehicles in the speed range corresponding to each fused limit.

5. The method according to claim 4, characterized in that, The first driving behavior information also includes the speed, minimum acceleration, average speed, minimum acceleration of the following vehicle in the adjacent lane, average speed of the following vehicle in the adjacent lane, longitudinal distance, lane change time, time to collision (TTC), and headway (THW) of the multiple first vehicles. The second driving behavior information also includes the speed, minimum acceleration, average speed, minimum acceleration of the following vehicle in the adjacent lane, average speed of the following vehicle in the adjacent lane, longitudinal distance, lane change time, TTC, and THW of the multiple second vehicles.

6. The method according to claim 4, characterized in that, The step of grouping the plurality of first vehicles into multiple speed ranges based on the first driving behavior information to obtain multiple first vehicle speed groups includes: Based on the first driving behavior information and kernel density estimation analysis (KDE), the multiple first vehicles are grouped into multiple speed ranges to obtain multiple first vehicle speed groups, and the statistical significance of the constraint index of the multiple first vehicle speed groups is verified by at least one testing method. The step of grouping the plurality of second vehicles into multiple speed ranges based on the second driving behavior information to obtain multiple second vehicle speed groups includes: Based on the second driving behavior information and KDE, the multiple second vehicles are grouped into multiple speed ranges to obtain multiple second vehicle speed groups, and the constraint index statistical significance of the multiple second vehicle speed groups is verified by at least one testing method.

7. The method according to claim 4, characterized in that, The step of determining multiple first limits of fuzzy traffic regulations corresponding to the multiple first vehicle speed groups and determining multiple second limits of fuzzy traffic regulations corresponding to the multiple second vehicle speed groups includes: Gaussian Mixture Model (GMM) is used to model the multiple first vehicle speed groups and the multiple second vehicle speed groups respectively. The parameters of the GMM are determined by the maximum likelihood estimation method, and the optimal number of clusters of the GMM is determined by the Bayesian Information Criterion (BIC). Based on the GMM and the boundary point based on posterior probability, the constraint index thresholds corresponding to the plurality of first vehicle speed groups and the plurality of second vehicle speed groups are determined. Determine multiple candidate limits for each first vehicle speed group and each second vehicle speed group; Based on the recall rate and false positive rate corresponding to the multiple candidate limits and the constraint index threshold, the multiple first limits and the multiple second limits are obtained from the multiple candidate limits.

8. The method according to claim 4, characterized in that, The step of weightedly fusing each of the plurality of first limits with each corresponding second limit to obtain a plurality of fused limits for the fuzzy traffic regulations corresponding to the plurality of speed ranges includes: Based on each of the plurality of first limits and each corresponding second limit, determine the DMCI evaluation value that is jointly corresponding to each of the plurality of first limits and each corresponding second limit; Based on the DMCI evaluation value corresponding to each of the plurality of first limits and its corresponding second limit, determine the first weight of each of the plurality of first limits and the second weight of each of the second limits corresponding to each of the plurality of first limits; The plurality of fusion limits are determined based on the plurality of first limits and the first weight, the plurality of second limits and the second weight.

9. The method according to claim 1 or 2, characterized in that, The step of determining the fuzzy traffic rule score of the vehicle under test based on the limit value and the actual value corresponding to the vehicle under test under the fuzzy traffic rule evaluation index includes: If the actual value is greater than the limit, the fuzzy traffic rule score is determined to be 1. If the actual value is less than the limit, the fuzzy traffic rule score is determined to be the ratio of the actual value to the limit. The determination of the specific traffic regulations score of the vehicle under test includes: Obtain the score corresponding to each of one or more explicit traffic regulations for the vehicle under test, wherein when the explicit traffic regulation is met, the corresponding score is 1, and when the explicit traffic regulation is violated, the corresponding score is 0; The score of the vehicle under test is determined based on the score and weight of each explicit traffic regulation in the one or more explicit traffic regulations. The determination of the behavioral normativity score of the vehicle under test includes: Obtain the score corresponding to each normative evaluation index among multiple normative evaluation indices used to characterize behavioral normativity. For normative evaluation indices where smaller values ​​are better, the score is determined based on a linear decay function of the upper limit threshold. For normative evaluation indices where larger values ​​are better, the score is determined based on a linear proportional function of the threshold. For normative interval-type evaluation indices, the score is determined based on a piecewise function. The behavioral normativity score is determined based on the score and weight of each normative evaluation indicator among the multiple normative evaluation indicators.

10. The method according to claim 1 or 2, characterized in that, The step of determining the traffic regulation compliance evaluation result of the vehicle under test based on the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score includes: Based on the entropy weight method, determine the weights corresponding to the fuzzy traffic rule score, the explicit traffic rule score, and the behavioral normativity score, respectively. The traffic regulation compliance evaluation result is determined based on the fuzzy traffic regulation score, the explicit traffic regulation score, and the behavioral normativity score, as well as their respective weights.

Images