Automatic driving simulation test method and device, electronic equipment and storage medium

Abstract

This invention relates to an autonomous driving simulation testing method, apparatus, electronic device, and storage medium. The autonomous driving simulation testing method includes: obtaining the safety constraint boundary of an autonomous driving algorithm; determining the optimal test scenario parameters from multiple first test scenario parameters within the safety constraint boundary range; executing the optimal test scenario parameters in a digital twin scenario corresponding to the simulation testing environment to obtain first health state data within the digital twin corresponding to the autonomous driving algorithm; if the first health state data determines that the autonomous driving algorithm response is abnormal or fluctuates abnormally, the optimal test scenario parameters are determined as the failure boundary of the autonomous driving algorithm. The embodiments of this application can ensure the continuity of the testing process, avoid test interruptions caused by collisions or algorithm exits, and improve the efficiency of boundary exploration. Simultaneously, it can achieve deep perception of the actual operating conditions within the autonomous driving algorithm, enabling accurate calibration of the algorithm's failure boundary.

Description

Technical Field

[0001] This application relates to the field of autonomous driving simulation testing, and in particular to an autonomous driving simulation testing method, apparatus, electronic device and storage medium. Background Technology

[0002] With the rapid development and widespread application of autonomous driving technology, the safe deployment of autonomous driving systems relies on massive amounts of testing and verification. Simulation testing technology based on virtual scenarios has become an important means to improve testing efficiency and reduce testing costs. Existing simulation testing technologies mainly focus on two directions: first, scenario-based test case expansion, which constructs comprehensive test scenarios by combining elements such as traffic participant behavior, weather conditions, and road structure; and second, game-theoretic testing based on intelligent agent interaction, which uses methods such as reinforcement learning to induce adversarial behavior in background vehicles to expose the defects of the algorithm under test.

[0003] However, existing adversarial testing methods often only pursue the exposure of high-risk scenarios, making it difficult to achieve a balance between boundary exploration and test continuity. This results in frequent interruptions of simulation testing during the boundary exploration process due to collisions or abnormal exits of the autonomous driving algorithm, making the boundary exploration efficiency of the autonomous driving algorithm extremely low. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this application provides an autonomous driving simulation test method, apparatus, electronic device and storage medium.

[0005] Firstly, this application provides an autonomous driving simulation testing method, including: Obtain the safety constraint boundary of the autonomous driving algorithm, wherein the safety constraint boundary includes multiple first test scenario parameters that can ensure that the simulation test of the autonomous driving algorithm is not interrupted; Among the multiple first test scenario parameters within the scope of the safety constraint boundary, the optimal test scenario parameters are determined; The optimal test scenario parameters are executed in the digital twin scenario corresponding to the simulation test environment to obtain the first health status data inside the digital twin corresponding to the autonomous driving algorithm. If the first health status data is greater than the first critical threshold, the optimal test scenario parameters will be determined as the failure boundary that causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally.

[0006] Optionally, the safety constraint boundaries of the autonomous driving algorithm are obtained, including: Obtain the test scenario parameter set corresponding to the current test scenario, wherein the test scenario parameter set includes multiple second test scenario parameters; Obtain the health status sequence within the digital twin based on historical records; Based on the health status data of the previous time period in the health status sequence, a first test scenario parameter is determined from multiple second test scenario parameters to cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally, and is used as the safety constraint boundary.

[0007] Optionally, obtain the set of test scenario parameters corresponding to the current test scenario, including: Obtain the health status sequence, test environment data, and vehicle status data within the digital twin based on historical records; Based on the health status sequence, the test environment parameters and the vehicle status data, environmental prediction is performed to determine the high-risk environment types and their probability of occurrence that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. Obtain the health status sequence and traffic participant behavioral characteristic parameters within the historical records of the digital twin; Based on the health status sequence and the behavioral feature parameters, behavioral prediction is performed to determine the critical behavioral parameters that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. The test scenario parameter set is determined based on the high-risk environment type, the probability of occurrence, and the critical behavior parameters.

[0008] Optionally, a set of test scenario parameters is determined based on the high-risk environment type, the probability of occurrence, and the critical behavior parameters, including: The test environment data corresponding to high-risk environment types with an occurrence probability higher than a preset occurrence probability threshold are adjusted multiple times for parameter types and / or parameter values ​​to obtain an environment data set. The critical behavior parameters are adjusted multiple times based on parameter type and / or parameter value to obtain a set of behavior parameters; The test scenario parameter set is determined based on the environmental data set and the behavioral parameter set.

[0009] Optionally, among a plurality of first test scenario parameters within the scope of the safety constraint boundary, the optimal test scenario parameters are determined, including: Obtain the second health status data of the digital twin from the previous time period; The test pressure is determined based on the second health status data; Among the multiple first test scenario parameters within the scope of the safety constraint boundary, the optimal test scenario parameter is determined based on the test pressure.

[0010] Optionally, the test stress is determined based on the second health status data, including: Obtain the baseline pressure value, high-state threshold, low-state threshold, and adjustment coefficient; The test pressure is determined based on the second health status data, the baseline pressure value, the high status threshold, the low status threshold, and the adjustment coefficient.

[0011] Optionally, among multiple first test scenario parameters within the safety constraint boundary range, the optimal test scenario parameter is determined based on the test pressure, including: The step size control factor is determined based on the test pressure. Based on the multiple first test scenario parameters within the safety constraint boundary range and the test pressure, determine the first fitness value corresponding to each first test scenario parameter and the optimal solution among the multiple first fitness values. Obtain a first search direction, and generate multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction, the step size control factor, and the test pressure, wherein the multiple third test scenario parameters are located within the safety constraint boundary range; A second fitness value corresponding to each of the third test scenario parameters is determined based on the multiple third test scenario parameters; Determine the latest optimal solution from among multiple second fitness values ​​and the optimal solution; Adjust the first search direction and / or adjust the step size control factor according to the test pressure, and re-execute the step of generating multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction and the step size control factor until a preset number of repetitions is reached. Then, determine the first test scenario parameters or third test scenario parameters corresponding to the latest optimal solution as the optimal test scenario parameters.

[0012] Optionally, determining a first fitness value corresponding to each first test scenario parameter based on multiple first test scenario parameters within the safety constraint boundary range and the test pressure includes: For each parameter of the first test scenario, obtain the failure probability, scenario coverage, and boundary exploration degree corresponding to the parameter of the first test scenario. The failure probability is the probability that the first test scenario parameter causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally. The scenario coverage is the diversity of the first test scenario parameter. The boundary exploration degree is the degree to which the first test scenario parameter is close to the safety constraint boundary. Based on the test pressure, determine the first weighting coefficient corresponding to the failure probability, the second weighting coefficient corresponding to the scene coverage, and the third weighting coefficient corresponding to the boundary exploration degree; The first fitness value is determined based on the failure probability, the first weighting coefficient, the scene coverage, the second weighting coefficient, the boundary exploration degree, and the third weighting coefficient.

[0013] Optionally, determining the optimal solution among a plurality of first fitness values ​​includes: Determine the maximum fitness value among the plurality of first fitness values; The maximum fitness value is determined as the optimal solution.

[0014] Optionally, adjusting the first search direction includes: Obtain the maximum value among the second fitness values ​​corresponding to multiple third test scenario parameters; The compensation control quantity is determined based on the fitness value corresponding to the latest optimal solution and the maximum value among the second fitness values ​​corresponding to the multiple third test scenario parameters. The adjusted first search direction is determined based on the compensation control amount.

[0015] Optionally, determining the adjusted first search direction based on the compensation control amount includes: If the compensation control amount increases, the direction of the optimal solution region where the latest optimal solution is located will be determined as the adjusted first search direction; If the compensation control amount decreases, the region containing the maximum value among the second fitness values ​​corresponding to the multiple third test scenario parameters is taken as the optimal solution region, and the direction of the optimal solution region is determined as the adjusted first search direction.

[0016] Optionally, determining the step size control factor based on the test pressure includes: The step size adjustment coefficient is determined based on the test pressure, and the step size adjustment coefficient is inversely proportional to the test pressure. Get the base step size; The step size control factor is determined based on the base step size and the step size adjustment coefficient.

[0017] Secondly, this application provides an autonomous driving simulation testing device, comprising: The first acquisition module is used to acquire the safety constraint boundary of the autonomous driving algorithm. The safety constraint boundary includes multiple first test scenario parameters that can ensure that the simulation test of the autonomous driving algorithm is not interrupted. The first determining module is used to determine the optimal test scenario parameters among multiple first test scenario parameters within the range of the safety constraint boundary. The second acquisition module is used to execute the optimal test scenario parameters in the digital twin scenario corresponding to the simulation test environment to acquire the first health status data inside the digital twin corresponding to the autonomous driving algorithm. The second determining module is used to determine the optimal test scenario parameters as the failure boundary that causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally if the first health status data is greater than the first critical threshold.

[0018] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes a program stored in memory, it implements the autonomous driving simulation test method described in any of the first aspects.

[0019] Fourthly, this application provides a computer-readable storage medium storing a program for an autonomous driving simulation test method, wherein when the program for the autonomous driving simulation test method is executed by a processor, it implements the steps of the autonomous driving simulation test method described in any of the first aspects.

[0020] The beneficial effects of this invention are: This application's embodiments obtain a safety constraint boundary that ensures uninterrupted simulation testing, determine the optimal test scenario parameters within it, and then determine whether the algorithm has failed based on the first health state data after execution, identifying the parameters at the time of anomalies as the failure boundary. This ensures the continuity of the testing process, avoids test interruptions due to collisions or algorithm exits, improves the efficiency of boundary exploration, and ensures that all disturbances to the subsequent autonomous driving algorithm are within its tolerable range, significantly improving the automation level of simulation testing. Simultaneously, this application constructs a digital twin of the autonomous driving algorithm and a digital twin scenario of the simulation testing environment. By obtaining the first health state data within the digital twin, a two-way mapping and linkage mechanism is implemented between the first health state data within the autonomous driving algorithm and the optimal test scenario parameters of the simulation testing scenario. This enables deep perception of the actual operating status within the autonomous driving algorithm, facilitating the determination of failure boundaries based on the internal state of the autonomous driving algorithm during boundary exploration, achieving accurate calibration of algorithm failure boundaries, and providing a reproducible scenario knowledge base for subsequent testing. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating an autonomous driving simulation testing method provided in this application embodiment; Figure 2 for Figure 1 Flowchart of step S101; Figure 3 for Figure 2 Flowchart of step S201; Figure 4 for Figure 1 Flowchart of step S102; Figure 5 for Figure 4 Flowchart of step S403; Figure 6 A structural diagram of an autonomous driving simulation testing device provided in this application embodiment; Figure 7 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Existing simulation testing methods, particularly adversarial testing methods, often prioritize exposure to high-risk scenarios, making it difficult to balance boundary exploration with test continuity. This often leads to frequent interruptions during boundary exploration due to collisions or abnormal exits of the autonomous driving algorithm, resulting in extremely low boundary exploration efficiency for autonomous driving algorithms. Therefore, this application provides an autonomous driving simulation testing method, apparatus, electronic device, and storage medium. This application provides an autonomous driving simulation testing method, such as... Figure 1 As shown, it includes: Step S101: Obtain the safety constraint boundary of the autonomous driving algorithm. The safety constraint boundary includes multiple first test scenario parameters that can ensure the simulation test of the autonomous driving algorithm is not interrupted. In this embodiment, the autonomous driving algorithm refers to the algorithmic logic of the perception, decision-making, and control modules included in the autonomous driving system under test. The safety constraint boundary is an initial safety boundary calculated using a risk estimation function and a preset safety threshold based on a historical sequence of the autonomous driving algorithm's health status (i.e., a sequence including indicators such as perception uncertainty, decision hesitation, and control confidence). It represents a predicted value under ideal operating conditions. Only test scenario parameters (including environmental parameter adjustments and background vehicle behavior disturbances) falling within this safety boundary are allowed to execute, ensuring that all disturbances are within the algorithm's tolerance range and preventing test interruption. The first test scenario parameter refers to the test scenario parameters located within the safety constraint boundary. This first test scenario parameter ensures that the simulation test of the autonomous driving algorithm is not interrupted due to collisions or abnormal exits. The first test scenario parameter includes test environment parameters (such as road type, traffic flow, and weather parameters) and behavioral characteristic parameters of traffic participants (such as the lane-changing behavior of vehicles ahead).

[0026] The health status sequence includes multiple health status data arranged in chronological order. The health status data includes values ​​for perceived uncertainty, decision hesitation, and control confidence.

[0027] Perception uncertainty refers to the confidence level of the perception module in the autonomous driving algorithm in judging the category, location, and attributes of detected targets. It is used to quantify the confidence level of perception. Perception uncertainty is calculated by collecting the target detection confidence level output by the perception model in the autonomous driving algorithm. For each frame of n detected targets, the uncertainty of each individual target is calculated based on the information entropy of the perception results; where, This represents the predicted probability of the perceptual model for the category to which each target belongs. Finally, the average of the uncertainties of all targets is taken as the overall perceptual uncertainty of the current frame. :

[0028] Decision hesitation is the degree of indecisiveness or uncertainty experienced by the decision-making module of an autonomous driving algorithm when generating multiple candidate trajectories due to trajectory fluctuations and conflicting plans. Decision hesitation is determined by collecting multiple consecutive frames of candidate trajectories output by the decision-making module, calculating the combined value of fluctuation and difference frame by frame based on the degree of fluctuation in a single frame and the similarity to adjacent frames, and then averaging the results within a time window. :

[0029] in, This indicates the current frame's stacking volatility. In practical applications, it can be used to... Perform normalization (e.g., divide by the maximum permissible deviation). This indicates the similarity between the trajectory of the current frame and the previous frame. This indicates the degree of similarity; the greater the difference, the greater the hesitation in decision-making. This indicates that the average value is taken over a time window.

[0030] Control confidence, or the confidence level of planned control, refers to the degree of consistency between the control module commands issued and the actual execution of commands in an autonomous driving algorithm. It quantifies the reliability and stability of the precise execution of control commands. Planned control confidence is calculated by collecting feedback on the expected output of control commands and the actual execution; the control confidence level is then obtained by mapping the negative exponent of the deviation. The smaller the bias, the higher the confidence level.

[0031] in, To control the desired result of the command, For the actual control output results, As a normalization factor, This indicates the deviation between expectations and actual execution.

[0032] The above indicators are combined to form health status data. :

[0033] In this step, a pre-set safety constraint boundary can be obtained, which contains multiple first test scenario parameters.

[0034] Step S102: Determine the optimal test scenario parameters from among the multiple first test scenario parameters within the scope of the safety constraint boundary. In this embodiment of the application, the optimal test scenario parameters refer to the test scenario parameters that maximize the test benefits (maximizing test benefits means optimizing the test scenario parameters so that the test process can more effectively expose the problems of the algorithm under test, but without causing the test to stop due to collision or exceeding the test range).

[0035] In this step, the optimal test scenario parameters can be quickly searched among the multiple first test scenario parameters contained in the safety constraint boundary, with the objective function of maximizing test benefits. The final test scenario parameters obtained are the optimal test scenario parameters.

[0036] Step S103: Execute the optimal test scenario parameters in the digital twin scenario corresponding to the simulation test environment to obtain the first health status data inside the digital twin corresponding to the autonomous driving algorithm. In this embodiment of the application, the simulation test environment refers to the real test environment used for simulation testing of autonomous driving algorithms. The digital twin scene is generated based on real road scenes and maps the state of the real test environment in real time. It includes, but is not limited to, road states such as highways, urban expressways, and rural roads, traffic participants such as cars, trucks, pedestrians, bicycles, and motorcycles, and weather conditions such as rain, snow, and fog. It includes elements such as high-precision maps based on OpenDRIVE and dynamic traffic participants based on OpenSCENARIO. In other words, the digital twin scene includes static scene elements (such as road information) and dynamic traffic participants (such as other vehicles, pedestrians, etc.) and their behaviors (such as lane changing, acceleration, deceleration, etc., and pedestrian running behavior, etc.).

[0037] A digital twin is a lightweight simulation copy of the autonomous driving algorithm under test. It maps the internal state of the autonomous driving algorithm in real time, such as: the confidence level of perception (confidence distribution of target detection, sensor occlusion ratio, changes in visual blind spots), the uncertainty of the prediction module (trajectory planning set, trajectory score), and the error of the planned trajectory (tracking error, steering wheel fine adjustment). It includes simulation copies of the perception, decision-making, and control modules. Through the embedded state perception interface, it collects the intermediate outputs of the perception layer, decision-making layer, and control layer of the autonomous driving algorithm in real time to form health status data.

[0038] The first health status data refers to the health status data collected within the autonomous driving algorithm, which includes perception uncertainty, decision hesitation, and control confidence.

[0039] In this step, the optimal test scenario parameters can be sent to the digital twin scenario. The digital twin scenario adjusts the static environmental parameters or the dynamic traffic participant behavior accordingly and performs simulation. During the execution, the intermediate outputs of the algorithm's perception layer, decision layer, and control layer can be collected in real time through the state perception interface embedded in the digital twin. The first health state data representing the health level of the algorithm can be quantitatively extracted. The state perception interface can also have a certain storage function. After collecting the intermediate outputs of the algorithm's perception layer, decision layer, and control layer, it can be temporarily stored for a period of time. After reading the first health state data, the temporarily stored data in the state perception interface can be deleted.

[0040] Based on the obtained optimal trial action parameters, actions are executed in the digital twin scenario, such as adjusting the road friction coefficient and weather conditions, setting the behavior of background vehicles, and monitoring the changes in the internal state of the algorithm under test in real time during the execution process, as well as monitoring the real-time perception uncertainty, decision hesitation, and control confidence.

[0041] Step S104: If the first health status data determines that the autonomous driving algorithm response is abnormal or fluctuates abnormally, the optimal test scenario parameters are determined as the failure boundary of the autonomous driving algorithm.

[0042] In this embodiment of the application, failure refers to the unexpected behavior of the autonomous driving algorithm during the test, including a sudden increase in perception uncertainty, excessive decision hesitation, decreased control confidence, or failure situations such as trajectory deviation and emergency braking. The failure boundary refers to the set of test scenario parameters that cause the algorithm to fail, and this parameter combination is marked as the boundary sensitive area.

[0043] In this step, the algorithm can be judged as to whether it has failed based on the first health status data. The judgment method can be to compare the first health status data with the preset abnormal threshold. If the perceived uncertainty, decision hesitation or control confidence exceeds the corresponding threshold, or if events such as collision or exit are detected, the algorithm is judged to have failed. At this time, the optimal test scenario parameters currently being executed are determined as the failure boundary, and the optimal test scenario parameters, initial state, first health status data, and trial results (such as abnormal responses) are stored in the bidirectional mapping database.

[0044] The trial-and-error process is repeated. When a trial causes the algorithm's perceived uncertainty to exceed a threshold, the safety constraint boundary immediately takes effect, limiting the aggressiveness of subsequent trials (adjusting the test pressure mentioned above, i.e., adjusting environmental factors, weather conditions, the complexity of traffic participants, etc.). At the same time, the test pressure adaptive adjustment module will reduce the pressure and gradually increase it again after the algorithm's state recovers.

[0045] Through continuous testing, the system can expose the failure modes of the algorithm in various coupled scenarios, and the test was not interrupted throughout the process. All testing actions were performed within the safety constraint boundaries, and the test data is complete and reproducible.

[0046] This application's embodiments obtain a safety constraint boundary that ensures uninterrupted simulation testing, determine the optimal test scenario parameters within it, and then determine whether the algorithm has failed based on the first health state data after execution, identifying the parameters at the time of anomalies as the failure boundary. This ensures the continuity of the testing process, avoids test interruptions due to collisions or algorithm exits, improves the efficiency of boundary exploration, and ensures that all disturbances to the subsequent autonomous driving algorithm are within its tolerable range, significantly improving the automation level of simulation testing. Simultaneously, this application constructs a digital twin of the autonomous driving algorithm and a digital twin scenario of the simulation testing environment. By obtaining the first health state data within the digital twin, a two-way mapping and linkage mechanism is implemented between the first health state data within the autonomous driving algorithm and the optimal test scenario parameters of the simulation testing scenario. This enables deep perception of the actual operating status within the autonomous driving algorithm, facilitating the determination of failure boundaries based on the internal state of the autonomous driving algorithm during boundary exploration, achieving accurate calibration of algorithm failure boundaries, and providing a reproducible scenario knowledge base for subsequent testing.

[0047] In another embodiment of this application, step S101 obtains the safety constraint boundaries of the autonomous driving algorithm, such as... Figure 2 As shown, it includes: Step S201: Obtain the test scenario parameter set corresponding to the current test scenario, wherein the test scenario parameter set includes multiple second test scenario parameters; In this embodiment, the current test scenario can be a sub-test scenario (such as rainfall intensity, light intensity, lane line clarity, etc.) under a larger test scenario (such as urban roads on a rainy night). The test scenario parameter set refers to the complete set of all predefined adjustable test parameters, including test environment parameter adjustment actions (such as road type, traffic flow, and meteorological parameters) and behavioral characteristic parameters of traffic participants (such as lane-changing behavior and deceleration of the vehicle in front). It includes multiple test scenario parameters under this sub-test scenario, such as heavy rain, low light intensity, lane lines blurred due to water accumulation, and a vehicle suddenly crossing the road ahead. The second test scenario parameter refers to any combination of parameters in this set, and each second test scenario parameter represents a specific configuration that can be applied to the digital twin scenario.

[0048] In one embodiment of this application, step S201 obtains a set of test scenario parameters corresponding to the current test scenario, such as... Figure 3 As shown, it includes: Step S301: Obtain the health status sequence, test environment data, and vehicle status data within the digital twin from historical records; In this embodiment, the test environment data refers to the static environmental parameters corresponding to each health state data in the health state sequence, including road friction coefficient, weather conditions (rain, snow, fog), light intensity, road type, etc. The vehicle state data is the dynamic simulation result of the autonomous vehicle controlled by the autonomous driving algorithm running in the digital twin scenario, including: the vehicle's position coordinates in the simulation environment, vehicle speed: the vehicle's speed and acceleration, vehicle attitude: the vehicle's attitude angles (such as yaw angle, pitch angle, roll angle), control command execution results: the actual executed control commands and their effects, etc.

[0049] In this step, the historical health status sequence and the test environment data corresponding to each health status data in the health status sequence can be obtained.

[0050] Step S302: Based on the health status sequence, the test environment parameters and the vehicle status data, perform environmental prediction to determine the high-risk environment type and probability of occurrence that will cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. In this embodiment of the application, environmental prediction refers to using a time-series anomaly detection model to detect health state sequences, test environment parameters, and vehicle state data in order to predict the types of high-risk environments that may occur in the future and their probabilities. The time-series anomaly detection model includes, but is not limited to, statistical methods, machine learning methods, and ensemble learning methods such as isolated forests and random forests. Detection refers to detecting whether the output results of sensors (simulation test results of sensors and results identified after algorithm processing) are stable, whether the step size of the dynamic model oscillates, and whether the scene logic is normal.

[0051] High-risk environment types refer to categories of environmental conditions that may cause abnormal or fluctuating responses from autonomous driving algorithms, such as slippery roads, low visibility, and strong sunlight. The probability of occurrence refers to the numerical value indicating the likelihood of this high-risk environment type appearing within a future time window.

[0052] In this step, historical health status sequences, test environment data, and dynamic model results can be input into the time-series anomaly detection model. The model analyzes the correlation between changes in environmental parameters, dynamic response, and deterioration of the algorithm's health status, and outputs the types of high-risk environments that may occur in the future and their probability of occurrence.

[0053] Step S303: Obtain the health status sequence and traffic participant behavior characteristic parameters within the history of the digital twin; In this embodiment of the application, the behavioral characteristic parameters of traffic participants refer to the action parameters of traffic participants, such as the timing of the vehicle cutting in from the left lane, the magnitude of deceleration, the magnitude of acceleration, and the intention to change lanes.

[0054] In this step, the health status sequence of historical records and the behavioral characteristic parameters of traffic participants interacting with the autonomous driving algorithm in the digital twin scenario can be read.

[0055] Step S304: Based on the health status sequence and the behavioral feature parameters, predict the behavior and determine the critical behavioral parameters that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. In this embodiment, behavior prediction involves constructing a game matrix for a specific traffic participant, predicting the response pattern of the autonomous driving algorithm to different behaviors of that participant, and identifying critical behavior parameters that may lead to algorithm failure. Critical behavior parameters refer to the boundary values ​​at which the algorithm begins to fail when the traffic participant's behavior parameters reach a certain value.

[0056] Specifically, behavior prediction can monitor the following based on the decision logs, planned trajectories, and control commands of the autonomous driving algorithm: the confidence level of perception in recognizing specific obstacles, the judgment of the intentions of surrounding vehicles, and the continuity of trajectory planning (yes, this is actually the intermediate output of the algorithm under test in terms of perception, prediction, and planning). For example, for the vehicle in the left lane of the current interaction, a game matrix can be constructed to predict the response of the algorithm under test under different behaviors of the vehicle (constant speed, acceleration, deceleration, cutting in).

[0057] In this step, the response changes of the algorithm under different behavioral parameters can be analyzed based on the historical health status sequence and behavioral characteristic parameters to identify the critical behavioral parameters that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally.

[0058] Step S305: Determine the test scenario parameter set based on the high-risk environment type, the occurrence probability, and the critical behavior parameters.

[0059] In this step, the test environment data corresponding to high-risk environment types with an occurrence probability higher than a preset occurrence probability threshold can be adjusted multiple times for parameter types and / or parameter values ​​to obtain an environment data set; the critical behavior parameters can be adjusted multiple times for parameter types and / or parameter values ​​to obtain a behavior parameter set; and the test scenario parameter set can be determined based on the environment data set and the behavior parameter set.

[0060] Specifically, a probability threshold (e.g., 70%) can be set first. From the multiple high-risk environmental types and their probabilities output by the environmental prediction, high-risk environmental types with a probability higher than this threshold can be selected. Then, the test environment data corresponding to these high-risk environmental types can be adjusted multiple times. Each adjustment can change the parameter type (e.g., in a slippery road scenario, the friction coefficient, water depth, or tire adhesion coefficient can be adjusted) or change the parameter value (e.g., gradually reducing the friction coefficient from 0.5 to 0.3), thereby generating multiple combinations of environmental parameters. The set of these combinations constitutes the environmental data set.

[0061] Centered on the critical behavior parameters output by behavior prediction, the parameters are adjusted multiple times. During the adjustment, small step scans can be performed near the critical value, or the parameter type can be changed to cover more behavior patterns, thereby generating multiple combinations of behavior parameters. The set of these combinations is the behavior parameter set.

[0062] The environmental data set and the behavioral parameter set are combined (e.g., using a Cartesian product to pair all environmental parameter combinations with all behavioral parameter combinations) to generate a complete set of test scenario parameters. Each element in this set contains both specific environmental conditions (e.g., friction coefficient 0.4, rainfall 20 mm / h) and specific traffic participant behaviors (e.g., deceleration -3.0, cut-in time 2.0 seconds), thus covering various coupling situations between high-risk environments and critical behaviors.

[0063] By systematically adjusting the parameter types and values ​​of high-probability, high-risk environmental parameters and critical behavioral parameters, a rich and focused set of test scenario parameters was generated. This ensured the diversity of test scenarios and concentrated the search scope on the areas most likely to expose algorithm defects, avoiding blind searches in the entire parameter space and significantly improving the targeting and efficiency of the tests.

[0064] The set of test scenario parameters can be represented by the following formula:

[0065] For environmental data sets, This is a set of behavioral parameters.

[0066] This implementation identifies high-risk factors from three dimensions—static environment, vehicle dynamics response, and dynamic interaction—through environmental prediction and behavioral prediction, and defines a set of test scenario parameters accordingly. This makes the generation of test scenarios more predictive and can capture key failure modes such as dynamic instability, thereby improving the coverage and relevance of the test.

[0067] Step S202: Obtain the health status sequence within the digital twin based on historical records; Step S203: Based on the health status data of the previous time period in the health status sequence, determine the first test scenario parameter that causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally among multiple second test scenario parameters, and use it as the safety constraint boundary.

[0068] Safety constraint boundaries can be expressed by the following formula:

[0069] in, This refers to health status data from the previous time period. For the constructed risk estimation function, As a safety threshold, thus filtering out those that meet the requirements. A subset of test scenario parameters is used as a safety constraint boundary to ensure that the test scenario parameters do not cause abnormal or abnormal fluctuations in the response of the autonomous driving algorithm.

[0070] This application embodiment uses historical health status data to filter a subset of safety parameters, enabling the safety constraint boundary to have the adaptive capability of the internal health status of the autonomous driving algorithm, avoiding the blindness of manually setting the boundary, and ensuring that the parameters within the boundary will not cause the test to be interrupted.

[0071] In another embodiment of this application, step S102 determines the optimal test scenario parameters from among the multiple first test scenario parameters within the scope of the security constraint boundary, such as... Figure 4 As shown, it includes: Step S401: Obtain the second health status data of the digital twin in the previous time period; In this embodiment of the application, the previous cycle refers to the previous trial or the test process of the previous time step, and the second health status data refers to the health status data inside the digital twin collected at the end of the previous cycle. Like the first health status data, it consists of perceived uncertainty, decision hesitation, and control confidence.

[0072] Step S402: Determine the test pressure based on the second health status data; In this embodiment, the test pressure is used to adjust the aggressiveness of the trial of the optimal test scenario parameters.

[0073] The formula for calculating test pressure is:

[0074] in, The basic test pressure is given, and λ is the adjustment coefficient. , and These are the low-state threshold and the high-state threshold, respectively. The magnitude of the vector corresponding to the health status data.

[0075] when If the algorithm is considered to be in good condition, an aggressive testing strategy is adopted to increase the testing pressure (e.g., increasing the complexity of traffic flow, sudden movements of vehicles in the environment, etc., will all affect the testing pressure). At this time, a conservative probing strategy is adopted to reduce testing pressure or even suspend probing to ensure testing continuity. In this step, the baseline pressure value, high state threshold, low state threshold, and adjustment coefficient can be obtained; the test pressure is determined based on the second health state data, the baseline pressure value, the high state threshold, the low state threshold, and the adjustment coefficient.

[0076] Step S403: Among the multiple first test scenario parameters within the safety constraint boundary range, determine the optimal test scenario parameter based on the test pressure.

[0077] In this step, the optimization strategy can be adjusted based on the current testing pressure. Specifically, testing pressure affects the step size control factor or weight coefficient in the optimization algorithm (such as the Cuckoo Algorithm). Under high testing pressure, a search strategy that restricts the step size and leans towards fine-grained boundary exploration is adopted, while under low testing pressure, a search strategy that allows for a larger search step size and leans towards global exploration is used. When the testing pressure is high, a conservative trial-and-error strategy is adopted to protect the stable operation of the algorithm; when the testing pressure is low, an aggressive trial-and-error strategy is adopted to explore the performance boundary. Then, among multiple first test scenario parameters within the safety constraint boundary, the optimal test scenario parameters are searched to maximize the testing benefit.

[0078] Within the safety constraint boundary, the Cuckoo algorithm is used to directly search for the optimal trial action parameters, and the nest population is initialized. Each nest represents a set of trial action parameters. All initial nest positions are randomly generated within the safety constraint boundary to ensure that the initial solutions meet the system stability requirements and avoid invalid searches.

[0079] This application embodiment dynamically adjusts the test pressure based on the health status data of the previous cycle, and adaptively controls the search strategy of the optimization process according to the test pressure, realizing real-time linkage between test intensity and algorithm status. It increases the challenge to explore the boundary when the algorithm is healthy, and conservatively explores to ensure continuity when the algorithm deteriorates.

[0080] In another embodiment of this application, step S403 determines the optimal test scenario parameter based on the test pressure from among multiple first test scenario parameters within the safety constraint boundary range, such as... Figure 5 As shown, it includes: Step S501: Determine the step size control factor based on the test pressure; In this embodiment, the step size control factor refers to the step size control factor of the Lévy flight in the Cuckoo algorithm, denoted as α. The Lévy flight is a random walk model that performs a global search using the Lévy flight formula:

[0081] in, Let i be the position of the i-th individual in the (t+1)-th iteration. Let i be the position of the i-th individual in the t-th iteration. This is the step size control factor, used to adjust the size of the step. For random numbers that follow a Lévy distribution, the parameters are... Control the shape of the distribution.

[0082] because The test scenario parameter vector may contain parameters with different physical dimensions (e.g., the road friction coefficient is dimensionless, the deceleration of the vehicle next to it is in m / s², and the cutting-in time is in seconds). Therefore, in practical applications, all parameters can be normalized to the same order of magnitude (e.g., the [0,1] interval) before optimization calculations are performed.

[0083] In this step, a step size adjustment coefficient can be determined based on the test pressure, and the step size adjustment coefficient is inversely proportional to the test pressure; a base step size can be obtained, which can be the step size control factor of the previous cycle, or a default initial value if there is no previous cycle; and the step size control factor can be determined based on the base step size and the step size adjustment coefficient.

[0084] Test stress is a metric that reflects the internal operational stress and near-failure risk of the tested autonomous driving algorithm. It is calculated from the internal health status data perceived by the digital twin (such as perception uncertainty and decision hesitation). A higher test stress value indicates that the algorithm is more vulnerable in its current state and closer to its performance boundary or failure threshold; a lower test stress value indicates that the algorithm is more robust and has a larger robustness margin.

[0085] In other words, the current test pressure value can be mapped to the step size control factor α. The mapping relationship can be inversely proportional. That is, when the test pressure increases, the step size control factor α is decreased, so that the search can be finely and safely adjusted locally; when the test pressure decreases, the step size control factor α is increased, so that the search can perform large step size and global boundary exploration.

[0086] Step S502: Determine the first fitness value corresponding to each first test scenario parameter and the optimal solution among the multiple first fitness values ​​based on the multiple first test scenario parameters within the scope of the safety constraint boundary and the test pressure. In this embodiment, the first fitness value refers to the test benefit value of each first test scenario parameter. The higher the fitness value, the greater the test value of that test scenario parameter. The first fitness value is calculated by the fitness function, and the formula for calculating the fitness function is as follows:

[0087] in Failure probability represents the predicted probability that the autonomous driving algorithm will experience functional failure (such as trajectory deviation, emergency braking, misidentification, etc.) under the current test scenario parameters. For scene coverage gain, i.e. whether to introduce new combinations of traffic participants, weather conditions or road structure combinations, to improve test diversity; The degree of boundary exploration is determined by comparing it with existing test results to determine whether it approaches or exceeds the currently known safety boundary. This is used to evaluate the innovativeness and challenge of the test scenario parameters, with the first weighting coefficient being... Second weighting coefficient Third weighting coefficient It can be fixed or dynamically adjusted, depending on the application scenario and requirements. The weight coefficients are dynamically adjusted according to the actual situation during the testing process, which is suitable for scenarios that require flexible adjustment of testing strategies.

[0088] The optimal solution refers to the test scenario parameters corresponding to the maximum value among all current first fitness values.

[0089] In this step, for each first test scenario parameter, the failure probability, scenario coverage, and boundary exploration degree corresponding to the first test scenario parameter can be obtained. The failure probability is the probability that the first test scenario parameter causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally. The scenario coverage is the diversity of the first test scenario parameter. The boundary exploration degree is the degree to which the first test scenario parameter approaches the safety constraint boundary. A first weighting coefficient corresponding to the failure probability, a second weighting coefficient corresponding to the scenario coverage, and a third weighting coefficient corresponding to the boundary exploration degree are determined according to the test pressure. A first fitness value is determined according to the failure probability, the first weighting coefficient, the scenario coverage, the second weighting coefficient, the boundary exploration degree, and the third weighting coefficient.

[0090] Specifically, for each first test scenario parameter within the safety constraint boundary, its corresponding failure probability, scenario coverage, and boundary exploration degree can be calculated. The weight coefficients in the fitness function can be dynamically adjusted using test pressure (e.g., increasing the weight of boundary exploration degree under high pressure) to calculate the first fitness value of each parameter. Then, the maximum fitness value is determined among multiple first fitness values. The maximum fitness value is determined as the optimal solution.

[0091] Step S503: Obtain the first search direction, and generate multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction, the step size control factor, and the test pressure. The multiple third test scenario parameters are located within the safety constraint boundary range. In this embodiment of the application, the first search direction refers to the initial random direction of Levi's flight in the Cuckoo algorithm, which is usually determined by a random vector that follows the Levi distribution. The third test scenario parameter refers to a set of candidate test scenario parameters newly generated after the current bird's nest (first test scenario parameter) passes through the Levi flight.

[0092] In this step, the search direction of the current iteration can be obtained first (initially a random direction). Then, for each first test scenario parameter (as the current bird's nest), combined with the test pressure (the test pressure can be further adjusted by the step size control factor), a corresponding third test scenario parameter is generated using the Lévy flight formula. After generation, it is necessary to check whether the new parameter is within the safety constraint boundary. If it exceeds the boundary, it is corrected (e.g., mapped back to the boundary). Finally, multiple third test scenario parameters are obtained.

[0093] Step S504: Determine the second fitness value corresponding to each of the third test scenario parameters based on the multiple third test scenario parameters; In this embodiment of the application, the second fitness value refers to the test benefit value of each third test scenario parameter, which is calculated using the same fitness function as the first fitness value.

[0094] In this step, the second fitness value can be calculated using the fitness function described above for each third test scenario parameter.

[0095] Step S505: Determine the latest optimal solution among multiple second fitness values ​​and the optimal solution; In this embodiment of the application, the latest optimal solution refers to the solution with higher fitness after comparing the current optimal solution (the optimal solution of the previous iteration) with the optimal solution in the newly generated third test scenario parameters.

[0096] In this step, we can first find the maximum value of the second fitness value among all the parameters of the third test scenario, and then compare this maximum value with the maximum value among the first fitness values. If the maximum value is greater than the fitness of the original optimal solution, then we take it as the latest optimal solution; otherwise, we retain the original optimal solution.

[0097] Step S506: Adjust the first search direction and / or adjust the step size control factor according to the test pressure, and re-execute the step of generating multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction and the step size control factor until a preset number of repetitions is reached, and determine the first test scenario parameter or the third test scenario parameter corresponding to the latest optimal solution as the optimal test scenario parameter.

[0098] In this embodiment, the preset number of repetitions refers to the maximum number of iterations of the Cuckoo algorithm, such as 50 times or 10 times.

[0099] In this step, the search direction for the next round can be dynamically adjusted based on the fitness ratio of the current optimal solution to the newly generated nest (e.g., through compensating control variables). Adjustments are made, and the step size control factor is recalculated based on the current test pressure (which may remain unchanged or be updated). Then, the latest optimal solution and all third test scenario parameters (or the population after being filtered by the discard probability) are used as the first test scenario parameters for the next iteration. Steps S503 to S505 are repeated. When the number of iterations reaches the preset number of repetitions, the iteration is stopped, and the test scenario parameters corresponding to the latest optimal solution (which may be the initial first test scenario parameters or the third test scenario parameters generated in a certain round) are determined as the global optimal test scenario parameters.

[0100] In one embodiment of this application, adjusting the first search direction includes: obtaining the maximum value among the second fitness values ​​corresponding to multiple third test scenario parameters; determining a compensation control amount based on the fitness value corresponding to the latest optimal solution and the maximum value among the second fitness values ​​corresponding to multiple third test scenario parameters; and determining the adjusted first search direction based on the compensation control amount.

[0101] To avoid getting trapped in local optima and to balance exploration and development capabilities, a compensation control variable is introduced. The formula for calculating the compensation control variable is as follows:

[0102] in, To compensate for the control quantity, For adjustment coefficients, This is the fitness value (i.e., test reward) of the current globally optimal solution. This represents the fitness value of the current nest (candidate exploratory action). This is the adjustment coefficient.

[0103] When a significant high-quality solution is detected, h is increased to enhance the focused search of that region. (In the Cuckoo Algorithm, the focused search of high-quality solution regions is enhanced by adjusting the parameter h. Specifically, when a significant high-quality solution is detected, h is increased to enhance the search intensity of that region; when a high-quality solution is close to other solutions, h is decreased to avoid excessive perturbation that could lead to the loss of the current optimal solution. This maintains a high discovery probability in the early stages of iteration, promoting broad exploration, and gradually decreases as iteration progresses, shifting towards a local fine-grained search, effectively improving convergence stability. After convergence, the globally optimal trial action parameters are output.)

[0104] Determining the adjusted first search direction based on the compensation control amount includes: if the compensation control amount increases, determining the direction of the optimal solution region where the latest optimal solution is located as the adjusted first search direction; if the compensation control amount decreases, the region where the maximum value of the second fitness values ​​corresponding to the multiple third test scenario parameters is located is taken as the optimal solution region, and the direction of the optimal solution region is determined as the adjusted first search direction.

[0105] This application embodiment dynamically adjusts the search step size of the Cuckoo algorithm by testing pressure and synchronously adjusts the search direction during the iteration process, realizing the linkage between testing pressure and optimization behavior. Under high testing pressure, a small step size and a more refined approach are used for boundary exploration, while under low testing pressure, a large step size and a more global approach are used. This ensures the efficiency of boundary search and maintains the continuity of testing, significantly improving the search quality and adaptability of the optimal test scenario parameters.

[0106] In another embodiment of this application, an autonomous driving simulation testing device is also provided, such as... Figure 6 As shown, it includes: The first acquisition module 11 is used to acquire the safety constraint boundary of the autonomous driving algorithm. The safety constraint boundary includes multiple first test scenario parameters that can ensure that the simulation test of the autonomous driving algorithm is not interrupted. The first determining module 12 is used to determine the optimal test scenario parameters among multiple first test scenario parameters within the range of the safety constraint boundary. The second acquisition module 13 is used to execute the optimal test scenario parameters in the digital twin scenario corresponding to the simulation test environment and acquire the first health status data inside the digital twin corresponding to the autonomous driving algorithm. The second determining module 14 is used to determine the optimal test scenario parameters as the failure boundary that causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally if the first health status data is greater than the first critical threshold.

[0107] In another embodiment of this application, an electronic device is also provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the autonomous driving simulation test method described in any of the foregoing method embodiments.

[0108] The electronic device provided in this invention allows the processor to execute a program stored in memory to obtain a safety constraint boundary that ensures uninterrupted simulation testing. Within this boundary, optimal test scenario parameters are determined, and the algorithm's failure is assessed based on the first health status data after execution. Parameters at the time of anomalies are identified as failure boundaries. This ensures the continuity of the testing process, avoids test interruptions due to collisions or algorithm exits, improves the efficiency of boundary exploration, and ensures that all subsequent disturbances to the autonomous driving algorithm remain within its acceptable range, significantly enhancing the automation level of simulation testing. Furthermore, this application constructs a digital twin of the autonomous driving algorithm and a digital twin scenario of the simulation testing environment. By acquiring the first health status data within the digital twin, a bidirectional mapping and linkage mechanism is established between the first health status data within the autonomous driving algorithm and the optimal test scenario parameters of the simulation testing scenario. This enables deep perception of the actual operating status within the autonomous driving algorithm, facilitating the determination of failure boundaries based on the internal state of the autonomous driving algorithm during boundary exploration. This provides a reproducible scenario knowledge base for subsequent testing.

[0109] The communication bus 1140 mentioned in the above-mentioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0110] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0111] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0112] The processor 1110 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0113] In another embodiment of this application, a computer-readable storage medium is also provided, on which a program for an autonomous driving simulation test method is stored. When the program for the autonomous driving simulation test method is executed by a processor, it implements the steps of the autonomous driving simulation test method described in any of the foregoing method embodiments.

[0114] It should be noted that, in this document, 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.

[0115] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An autonomous driving simulation testing method, characterized in that, include: Obtain the safety constraint boundary of the autonomous driving algorithm, wherein the safety constraint boundary includes multiple first test scenario parameters that can ensure that the simulation test of the autonomous driving algorithm is not interrupted; Among the multiple first test scenario parameters within the scope of the safety constraint boundary, the optimal test scenario parameters are determined; The optimal test scenario parameters are executed in the digital twin scenario corresponding to the simulation test environment to obtain the first health status data inside the digital twin corresponding to the autonomous driving algorithm. If the first health status data determines that the autonomous driving algorithm response is abnormal or fluctuates abnormally, the optimal test scenario parameters are determined as the failure boundary of the autonomous driving algorithm.

2. The autonomous driving simulation testing method according to claim 1, characterized in that, Obtain the safety constraint boundaries of the autonomous driving algorithm, including: Obtain the test scenario parameter set corresponding to the current test scenario, wherein the test scenario parameter set includes multiple second test scenario parameters; Obtain the health status sequence within the digital twin based on historical records; Based on the health status data of the previous time period in the health status sequence, a first test scenario parameter is determined from multiple second test scenario parameters to cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally, and is used as the safety constraint boundary.

3. The autonomous driving simulation testing method according to claim 2, characterized in that, Retrieve the set of test scenario parameters corresponding to the current test scenario, including: Obtain the health status sequence, test environment data, and vehicle status data within the digital twin based on historical records; Based on the health status sequence, the test environment parameters and the vehicle status data, environmental prediction is performed to determine the high-risk environment types and their probability of occurrence that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. Obtain the health status sequence and traffic participant behavioral characteristic parameters within the historical records of the digital twin; Based on the health status sequence and the behavioral feature parameters, behavioral prediction is performed to determine the critical behavioral parameters that cause the autonomous driving algorithm to respond abnormally or fluctuate abnormally. The test scenario parameter set is determined based on the high-risk environment type, the probability of occurrence, and the critical behavior parameters.

4. The autonomous driving simulation testing method according to claim 3, characterized in that, The test scenario parameter set is determined based on the high-risk environment type, the probability of occurrence, and the critical behavior parameters, including: The test environment data corresponding to high-risk environment types with an occurrence probability higher than a preset occurrence probability threshold are adjusted multiple times for parameter types and / or parameter values ​​to obtain an environment data set. The critical behavior parameters are adjusted multiple times based on parameter type and / or parameter value to obtain a set of behavior parameters; The test scenario parameter set is determined based on the environmental data set and the behavioral parameter set.

5. The autonomous driving simulation testing method according to claim 1, characterized in that, Among multiple first test scenario parameters within the safety constraint boundary range, the optimal test scenario parameters are determined, including: Obtain the second health status data of the digital twin from the previous time period; The test pressure is determined based on the second health status data; Among the multiple first test scenario parameters within the scope of the safety constraint boundary, the optimal test scenario parameter is determined based on the test pressure.

6. The autonomous driving simulation testing method according to claim 5, characterized in that, The test stress is determined based on the second health status data, including: Obtain the baseline pressure value, high-state threshold, low-state threshold, and adjustment coefficient; The test pressure is determined based on the second health status data, the baseline pressure value, the high status threshold, the low status threshold, and the adjustment coefficient.

7. The autonomous driving simulation testing method according to claim 5, characterized in that, Among multiple first test scenario parameters within the safety constraint boundary range, the optimal test scenario parameters are determined based on the test pressure, including: The step size control factor is determined based on the test pressure. Based on the multiple first test scenario parameters within the safety constraint boundary range and the test pressure, determine the first fitness value corresponding to each first test scenario parameter and the optimal solution among the multiple first fitness values. Obtain a first search direction, and generate multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction, the step size control factor, and the test pressure, wherein the multiple third test scenario parameters are located within the safety constraint boundary range; A second fitness value corresponding to each of the third test scenario parameters is determined based on the multiple third test scenario parameters; Determine the latest optimal solution from among multiple second fitness values ​​and the optimal solution; Adjust the first search direction and / or adjust the step size control factor according to the test pressure, and re-execute the step of generating multiple third test scenario parameters based on multiple first test scenario parameters, the first search direction and the step size control factor until a preset number of repetitions is reached. Then, determine the first test scenario parameters or third test scenario parameters corresponding to the latest optimal solution as the optimal test scenario parameters.

8. The autonomous driving simulation testing method according to claim 7, characterized in that, Determine the first fitness value corresponding to each first test scenario parameter based on multiple first test scenario parameters within the safety constraint boundary range and the test pressure, including: For each parameter of the first test scenario, obtain the failure probability, scenario coverage, and boundary exploration degree corresponding to the parameter of the first test scenario. The failure probability is the probability that the first test scenario parameter causes the autonomous driving algorithm to respond abnormally or fluctuate abnormally. The scenario coverage is the diversity of the first test scenario parameter. The boundary exploration degree is the degree to which the first test scenario parameter is close to the safety constraint boundary. Based on the test pressure, determine the first weighting coefficient corresponding to the failure probability, the second weighting coefficient corresponding to the scene coverage, and the third weighting coefficient corresponding to the boundary exploration degree; The first fitness value is determined based on the failure probability, the first weighting coefficient, the scene coverage, the second weighting coefficient, the boundary exploration degree, and the third weighting coefficient.

9. The autonomous driving simulation testing method according to claim 7, characterized in that, Determining the optimal solution among multiple first fitness values ​​includes: Determine the maximum fitness value among the plurality of first fitness values; The maximum fitness value is determined as the optimal solution.

10. The autonomous driving simulation testing method according to claim 7, characterized in that, Adjusting the first search direction includes: Obtain the maximum value among the second fitness values ​​corresponding to multiple third test scenario parameters; The compensation control quantity is determined based on the fitness value corresponding to the latest optimal solution and the maximum value among the second fitness values ​​corresponding to the multiple third test scenario parameters. The adjusted first search direction is determined based on the compensation control amount.

11. The autonomous driving simulation testing method according to claim 10, characterized in that, Determining the adjusted first search direction based on the compensation control amount includes: If the compensation control amount increases, the direction of the optimal solution region where the latest optimal solution is located will be determined as the adjusted first search direction; If the compensation control amount decreases, the region containing the maximum value among the second fitness values ​​corresponding to the multiple third test scenario parameters is taken as the optimal solution region, and the direction of the optimal solution region is determined as the adjusted first search direction.

12. The autonomous driving simulation testing method according to claim 7, characterized in that, The step size control factor is determined based on the test pressure, including: The step size adjustment coefficient is determined based on the test pressure, and the step size adjustment coefficient is inversely proportional to the test pressure; Get the base step size; The step size control factor is determined based on the base step size and the step size adjustment coefficient.

13. An autonomous driving simulation testing device, characterized in that, include: The first acquisition module is used to acquire the safety constraint boundary of the autonomous driving algorithm. The safety constraint boundary includes multiple first test scenario parameters that can ensure that the simulation test of the autonomous driving algorithm is not interrupted. The first determining module is used to determine the optimal test scenario parameters among multiple first test scenario parameters within the range of the safety constraint boundary. The second acquisition module is used to execute the optimal test scenario parameters in the digital twin scenario corresponding to the simulation test environment to acquire the first health status data inside the digital twin corresponding to the autonomous driving algorithm. The second determining module is used to determine the optimal test scenario parameters as the failure boundary of the autonomous driving algorithm if the first health status data determines the autonomous driving algorithm.

14. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in a memory, it implements the autonomous driving simulation test method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for an autonomous driving simulation test method, which, when executed by a processor, implements the steps of the autonomous driving simulation test method according to any one of claims 1-12.

Images