L3-level automatic driving takeover performance decoupling test method based on co-simulation

Through the collaborative simulation platform, the key elements of L3 autonomous driving takeover performance are analyzed, and its migration ability is verified in unknown scenarios, which solves the problem of autonomous driving takeover performance regulation in the existing technology, improves testing efficiency and reduces costs.

CN119025403APending Publication Date: 2024-11-26TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410152053.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-02-02
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively regulate the performance of L3 autonomous driving takeover, especially under the complex interaction of 'man-machine-road-traffic' factors, it is difficult to achieve effective control of reaction time and risk aversion levels, and the cost of large-scale scenario testing High, low efficiency.

Method used

Using a collaborative simulation test method, the simulation platform built by SCANeR, ROS and Matlab&Simulink is used to generate test scenarios based on scene element clustering sampling, use case set, decoupling and analyzing the impact of key elements on docking performance, and verified by Kriging model. Migration in unknown scenarios.

Benefits of technology

Realize real-time takeover judgment and human-machine control switching of L3 autonomous driving vehicles, improves the efficiency of autonomous driving testing, can identify the impact of a single factor on docking performance from complex coupling relationships, reduces testing costs, and meets the growing automation Driving testing requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025403A_ABST
    Figure CN119025403A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of intelligent automobiles, and provides an L3-level automatic driving takeover performance decoupling test method based on co-simulation. A collaborative simulation platform is built through SCANeR driving simulation software, an ROS system and Simulink simulation software in Matlab, and L3-level automatic driving takeover judgment prompt and control right switching are achieved. According to the method, a traffic risk scene test case set generated by scene element clustering sampling is combined, and a takeover experiment is carried out based on a co-simulation platform, so that key causes influencing takeover performance are discriminated, the influence of key elements on the takeover performance is subjected to decoupling characterization, and the decoupling regulation and control effect of a single element on the performance is analyzed; and carrying out mobility verification, carrying out prediction by adopting a Kriging method, and further verifying the reliability and validity of the decoupling relationship in an unknown scene. The L3-level automatic driving takeover test system has the advantages of being convenient to operate and low in cost, can be widely applied to L3-level automatic driving takeover tests, improves the test efficiency, and meets the ever-increasing automatic driving test requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent vehicles, and specifically relates to a L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation. Background Art

[0002] Autonomous driving technology helps to improve the safety and efficiency of road traffic system operation and is an important trend in the future development of intelligent transportation. Level 3 autonomous driving technology is a necessary stage in the transition from assisted driving to fully autonomous driving. Users still need to take over vehicle control when exceeding the system's operational design domain (ODD).

[0003] Since there are many "man-machine-road-traffic" factors in the system that affect the driver's takeover performance, the hierarchical structure is complex, and there are interactions and complex coupling relationships, it is difficult to achieve effective regulation of takeover performance such as reaction time and risk aversion level; it is necessary to decouple the coupling factors, discretize the impact of single factors on takeover performance, and identify the key factors affecting takeover performance.

[0004] Driving simulation technology is widely used in the development and testing of autonomous driving technology due to its safety, controllability and repeatability. Co-simulation makes it possible to integrate the manual driving subsystem, autonomous driving subsystem and semi-autonomous driving subsystem, and can achieve real-time vehicle control switching, which is conducive to solving the takeover test problem that is difficult to study with traditional driving simulation and real vehicle experiments.

[0005] In addition, due to the diversity of road traffic environments, large-scale scenario testing is costly and inefficient. Therefore, it is of great significance to narrow the scope of scenarios to be tested and predict the test results in other unknown scenarios by analyzing existing information and existing experimental data. Summary of the invention

[0006] In order to overcome the problems existing in the above-mentioned prior art and utilize the advantages of the collaborative simulation platform in the field of autonomous driving testing, the purpose of the present invention is to provide a L3 autonomous driving takeover performance decoupling test method based on collaborative simulation, carry out driving simulation experiments in a targeted manner, conduct decoupling test analysis on the factors affecting the driver's takeover performance and verify the transferability of the conclusions, comprehensively consider internal and external validity while improving experimental efficiency.

[0007] To achieve the above object, the technical solution of the present invention is as follows: A L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation, comprising the steps of: S1: Combine scene element cluster sampling to generate a set of test scenarios for all elements of human-vehicle-environment; S2: Conduct decoupling tests based on the collaborative simulation platform and analyze the decoupling relationship; S3: Verify the transferability of the decoupling relationship in step S2 in unknown scenarios.

[0008] The collaborative simulation platform is a test system built with SCANeR, Robot Operating System (ROS) and Matlab&Simulink collaborative simulation, including manual driving subsystem, autonomous driving subsystem and L3 semi-autonomous driving subsystem; the collaborative simulation platform establishes a time synchronization mechanism in the traffic environment to achieve high-fidelity, low-latency human-vehicle collaborative control and control switching of L3 autonomous driving vehicles, as well as real-time interaction between driving simulation vehicles and environmental vehicles.

[0009] Furthermore, the platform constructs the traffic environment based on SCANeR, uses ROS communication to achieve integrated collaboration among various modules, and realizes the driver's simulated control through external driving control hardware equipment; by providing autonomous driving perception, decision-making and planning modules, the behavior of the vehicle in autonomous driving mode is simulated; and based on the autonomous driving ODD defined in the research, it determines whether the driver needs to take over vehicle control.

[0010] The manual driving subsystem includes a human-computer interaction device, which simulates the front windshield and the side windows during real car driving through a multi-screen display. A button for controlling the platform switch is set in the center of the steering wheel for the subject to decide the start and end of the experiment. The accelerator and brake pedals and the driving control components are equipped with sensors for detecting pedal movements, which can obtain feedback on vehicle speed and acceleration. The subsystem also obtains signals such as accelerator pedal percentage, brake pedal force, steering wheel angle data, etc. through the hardware signal receiving module of Matlab&Simulink, and transmits them back to the simulation platform system through ROS Toolbox to determine the operation of the vehicle in manual driving mode in SCANeR.

[0011] The autonomous driving subsystem performs risk state perception, decision planning and vehicle control based on the environmental vehicle data around the autonomous driving vehicle, implements the software functions of the autonomous driving algorithm in Matlab & Simulink, and realizes the operation of the vehicle in autonomous driving mode. According to the three stages of environmental perception, path decision, and vehicle control of the autonomous driving algorithm, the subsystem is divided into three parts for construction: autonomous driving subsystem perception module, decision planning module, and control module. The three modules receive and output the algorithm parameters of other modules through ROS communication, form an autonomous driving behavior control closed loop in a forward and backward order, and jointly control the operation of the autonomous driving vehicle during the simulation process.

[0012] Furthermore, the autonomous driving subsystem perception module receives the position, speed, acceleration, heading angle and other motion states of the environment vehicles in real time through SCANeR sensor simulation, and processes them in real time by the user-defined risk perception algorithm to dynamically and holistically perceive the surrounding safety risk situation; the perception module filters out irrelevant vehicles in the environment that do not pose a conflict risk according to user-defined standards, and then sends the key vehicle motion state parameters that pose a conflict risk with the autonomous driving main vehicle to the decision-making planning module through ROS communication.

[0013] Furthermore, the decision-making and planning module of the autonomous driving subsystem receives the key vehicle position, speed, acceleration, and heading angle output by the perception module through ROS communication, and then performs real-time trajectory planning through a user-defined trajectory planning algorithm. The user-defined trajectory planning algorithm only needs to meet specific input and output conditions to adapt to this collaborative simulation platform: the input is the position, speed, acceleration, and heading angle of the environment vehicle, and the output is the position, speed, heading angle, curvature, and acceleration of the trajectory point of the autonomous driving main vehicle in the planning time domain. The output parameters of the trajectory planning algorithm are sent to the control module of the autonomous driving subsystem through ROS communication.

[0014] Furthermore, the autonomous driving subsystem control module needs to calibrate the brakes and throttles of the vehicles in SCANeR before the module is built, so as to obtain the relationship between the braking force, throttle percentage and the vehicle speed and acceleration, i.e. the brake and throttle calibration table. After the control module receives the output parameters of the trajectory planning algorithm through ROS communication, it can use the PID algorithm and the linear secondary regulator control algorithm to output the throttle pedal percentage, brake pedal force, and steering wheel angle according to the brake and throttle calibration table, thereby realizing the control of the autonomous driving main vehicle. After the control command is updated, the vehicle in the autonomous driving mode in SCANeR responds immediately, and adjusts the trajectory and behavior of the vehicle in the driving simulation process in real time. Subsequently, the behavior of the autonomous driving will also be adjusted with the surrounding vehicles, forming a complete data transmission closed loop until the driving simulation process ends.

[0015] The L3 semi-automatic driving subsystem integrates the functions of the automatic driving subsystem and the manual driving subsystem, and adds a takeover judgment prompt module and a control right switching module. According to the vehicle ODD, a trigger event for the vehicle to send a prompt requesting the driver to take over is set or a trigger judgment index for the takeover prompt is calculated for risk assessment. When it is determined that the driver needs to take over the vehicle control, a prompt is triggered and the control is switched to the driver based on the driver's reaction.

[0016] Furthermore, the L3 semi-autonomous driving subsystem takeover determination prompt module, on the one hand, determines the need to remind the driver to take over according to the defined autonomous driving ODD when encountering (not limited to) road construction areas ahead, lack of lane markings, emergency cut-in of vehicles ahead, bad weather and other events that exceed the autonomous driving ODD; on the other hand, according to research needs, the position, speed, acceleration, heading angle and other kinematic data of each interactive vehicle in the ROS system are used to calculate the trigger determination index of the takeover prompt, and then determine whether the driver needs to be reminded to take over in the current scenario based on the calculation results. When it is determined that the driver needs to take over, a signal that the vehicle needs to take over is sent to the ROS system, triggering the takeover interface window and prompt sound.

[0017] Furthermore, the control right switching module of the L3 semi-automatic driving subsystem switches the control right of the L3 automatic driving vehicle according to the reaction of the test driver after the takeover is triggered. The driver's reaction includes but is not limited to stepping on the brake pedal, turning the steering wheel, pressing the function button, etc., which can be defined according to different test scenarios. The specific logic of the switch is: the platform starts timing from the moment the takeover prompt is triggered. If the signal corresponding to the driver's reaction is not received after the set time threshold, it is judged that the driver chooses not to take over the vehicle. At this time, the driving control right is not switched, and the vehicle still makes decisions and controls according to the automatic driving mode, and adopts corresponding emergency strategies; if the signal corresponding to the driver's reaction is received within the time threshold, it is judged that the driver chooses to take over the vehicle, and the vehicle control command is switched to be input by the driving simulation device; similarly, the takeover judgment prompt module can also trigger a prompt to ask the driver whether to switch back to the automatic driving mode. If the driver chooses to switch to the automatic driving mode, the control right is switched from the driving simulation modules to the automatic driving modules, otherwise the control right is not switched.

[0018] Preferably, a time synchronization mechanism is established in the traffic environment to achieve high-fidelity, low-latency human-vehicle collaborative control and control switching. Based on the ROS TimeSynchronizer time synchronization mechanism, the control data of different systems are uniformly received and a callback function of the synchronization result is generated. In the callback function, based on the control API of different subsystems, the control data is synchronously transmitted back to SCANeR, which not only ensures low-latency vehicle control switching, but also enables real-time interaction between autonomous driving vehicles and semi-autonomous driving vehicles in the same traffic environment.

[0019] The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation combines scene element cluster sampling to generate a full-element test scenario of human-vehicle-environment, and conducts the test based on the collaborative simulation platform. According to the test results, the impact of the single element of human-vehicle-environment on the takeover performance is analyzed, and the transferability of the analysis conclusions in other unknown scenarios is verified.

[0020] In step S1, based on the collaborative simulation platform and combined with the cluster sampling of scene elements, a full-element test scene of human-vehicle-environment is generated. The specific process is as follows: S11. Sort out n autonomous driving simulation test scenario elements according to needs (for example, natural conditions such as weather and lighting, road components such as road curvature and slope, traffic signals, traffic flow density and other traffic elements, vehicle status elements such as the speed of the main vehicle and the interactive vehicle, trigger event types and urgency, etc.) and different values, and discretize the continuous variables in the elements.

[0021] Optionally, the importance index of each scene element is analyzed based on methods such as the Analytic Hierarchy Process (AHP). A hierarchical model is constructed based on the scene elements, and a judgment matrix is ​​constructed based on the hierarchical affiliation using subjective judgment or the Delphi method, and a hierarchical single sorting and consistency test are performed, so as to finally obtain the importance index of each element and its value.

[0022] Optionally, since the importance indexes of different elements and their values ​​are not in the same reference system, a relative importance index is introduced to put all scene elements into the same reference system to obtain an intuitive comparison of the importance indexes of all factors or their values, and it is necessary to introduce element occurrence probability constraints so that specific elements cannot coexist in the same scene at the same time, so as to avoid the generation of simulation scenarios of non-realistic scenarios.

[0023] S12. After determining the relative importance index and avoiding conflicts among scene elements, scene element combinations of different complexities can be obtained. The N-wise combination test method is used to ensure that the interactions among n factors are fully covered, so as to ensure that the impact of these element interactions on the L3 autonomous driving takeover performance is detected during the simulation experiment.

[0024] S13. Through completely randomized experimental design, orthogonal Latin square design and other methods, we can find a test case set as small as possible to cover all possible scenario element pairs, thereby reducing the cost of testing and analysis. The test case set is represented by the following matrix: Test case set T m×n There are m test scenarios, each involving n traffic elements, where f ij Represents the value of the jth traffic element in the i-th test scene generated.

[0025] Preferably, a clustering method can be applied to calculate the similarity between m test cases using Euclidean distance, and finally select a certain number of categories of autonomous driving test scenarios to reduce the cost of testing and analysis. The new test case set is the following matrix (m'≤m): Test case set T' m’×n It includes m' test scenarios, each involving n traffic elements, where f ij Represents the value of the jth traffic element in the i-th test scene generated.

[0026] In step S2, based on the generated traffic risk scenario test case set T' m’×n , test on the collaborative simulation platform, collect takeover performance data, decouple the impact of key factors on takeover performance, and analyze the decoupling regulation effect of a single factor on takeover performance. The specific process is as follows: S21. Build the traffic environment subsystem, including static traffic environment and dynamic traffic environment settings. In the driving simulation software SCANeR, according to the user-defined test road alignment, set the basic parameters such as the number of lanes, lane types, and road surface undulations, import and add object elements such as signs, markings, and roadside buildings to achieve simulation modeling of the static traffic environment; import the basic simulation environment vehicle group with vehicle dynamics into the simulated traffic environment, set the trigger mode and path and trajectory planning rules of the interactive behavior according to the test case set, and achieve simulation modeling of the dynamic traffic environment.

[0027] Optionally, based on test requirements, the time series data such as the position, speed, acceleration, heading angle of each vehicle in the same traffic environment, as well as the driver's behavior data such as the steering wheel angle, accelerator pedal force, brake pedal force, etc. of the vehicle under control can be exported from SCANeR in the form of a csv file for subsequent data analysis.

[0028] S22. Calculate the driver takeover performance indicators based on the experimentally collected data, including but not limited to the braking reaction time and steering reaction time of takeover, the extreme value, average value, standard deviation of the steering wheel angle and brake pedal force within a given time window, the minimum pre-collision time before takeover, etc.

[0029] S23. Taking traffic factors as input independent variables and driver takeover performance as output dependent variable, analyze the coupling relationship of human-vehicle-environment interaction in the human-machine co-driving system, identify the influencing factors of takeover performance, decouple the impact of key factors on takeover performance, analyze the decoupling regulatory effect of a single traffic factor on takeover performance, and further focus on identifying the key factors that affect poor takeover performance.

[0030] In step S3, based on the spatial interpolation method of the Kriging model, the observed data in the existing test scenario is used to predict the response in the unknown experimental scenario to verify the transferability of the decoupling relationship analyzed in the second step to other scenarios. The specific process is as follows: S31. Construct a training sample library for the Kriging model based on the decoupling test. 1 ,x 2 ,…,x i ,…,x m’ They are respectively the test scenario use case set matrix T' m’×n The 1st, 2nd, ..., i, ..., m' rows correspond to m' test scenario use cases: T' 1,: ,T' 2,: ,…,T' i,: ,…,T' m’,: , construct the scene element independent variable matrix X = {x 1 ,x 2 ,…,x i ,…,x m’ Similarly, according to the test results, the corresponding sample response value matrix Y = {y 1 ,y 2 ,…,y i ,…,y m’}. The subscript i of x and y both represent the i-th test scenario use case.

[0031] S32. Estimate model hyperparameters based on the training sample library and construct a Kriging model. The Kriging model assumes that all observed data have a mean of μ and a variance of σ. 2 The m'-dimensional normal distribution of Kriging model is expressed as follows: Where wF(X) and wf i (X) represents the polynomial regression part, which provides a global approximation of the fitting model, and w is the weight of the polynomial regression term; the nonparametric random part Z(X) represents the Gaussian stationary random process model, which provides a local approximation of the fitting model.

[0032] Consider two points x k ,x q , the subscripts k and q represent the kth test scene and the qth test scene respectively. When the relative distance between two points is small, it is considered that y(x k ) and y(x q ) are highly correlated, and the function describing the correlation is as follows: The covariance is then defined as: From this we can see that the hyperparameters of the model are w, μ, σ 2 ,θ l and p l (l=1,2,…,d, d represents the dimension of the design value), the maximum likelihood estimate of the above parameters can be obtained by using the response value Y of the observed data.

[0033] S33. Based on the prior hyperparameters estimated in step S32, a Kriging prediction model for the input traffic element independent variables and the driver takeover performance dependent variable is established, and the model can be used to predict the response value of the unknown data point. The unknown traffic scene element data is denoted as x ★ , ★ Indicates the response value Y(x ★ ) is unknown, and the corresponding traffic element value has not been tested on the collaborative simulation platform.

[0034] Among them, Corr T (x * ) is the basic data x of unknown traffic scene elements input ★ With m' known sample points X={x 1 ,x 2 ,…,x k ,…,x m The rest of the characters in the row vector describing the correlation between '} have the same meaning as #Formula 3: Corr T (x * )=[Corr(x 1 ,x * ),Corr(x 2 ,x * ),..,Corr(x m′ ,x * )] (#Formula7) According to the above formula, the basic data x* of new traffic scene elements is input to predict the driver's takeover performance under other scene parameter conditions, thereby verifying the transferability of the decoupling regulatory effect of the analyzed traffic elements on the takeover performance in other unknown scenarios.

[0035] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention has built a collaborative simulation platform that realizes the real-time takeover judgment and human-machine control switching of L3 autonomous driving vehicles. The platform realizes data transmission and time synchronization between traffic environment, manual driving, autonomous driving, and L3 semi-autonomous driving subsystems through ROS communication combined with SCANeR, Matlab & Simulink software, and realizes real-time switching of human-machine control, with a high degree of openness and compatibility. The takeover judgment prompt module supports custom ODD and trigger events, which is convenient for simulation testing of L3 autonomous driving takeover in different scenarios.

[0036] 2. This invention proposes a decoupling test method for L3 autonomous driving takeover performance based on collaborative simulation, analyzes the decoupling characterization of the impact of key factors on takeover performance and verifies the transferability in unknown scenarios. This method can discretize the impact of a single factor on takeover performance from the complex coupling relationship of many factors affecting takeover performance, identify the key causes affecting takeover performance, and thus analyze the decoupling regulation effect of a single factor on performance.

[0037] Compared with the large-scale scenario testing with high cost, the efficiency of autonomous driving simulation testing is improved on the basis of comprehensive consideration of the internal and external validity of the analysis conclusions, meeting the growing demand for autonomous driving testing. According to the proposed method, not only can the transferability of the decoupling regulation effect of a single factor on performance in different scenarios be verified, but also the road traffic environment with poor takeover performance can be screened out, and driving simulation experiments or computer simulation experiments can be carried out in a targeted manner to improve the verification efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a collaborative simulation platform; Figure 2 It is the time synchronization mechanism for each subsystem; Figure 3 Analyze traffic scene elements; Figure 4 Output an example of a time-varying curve graph of driver behavior to the platform; Figure 5 This is an example of the process of using the Kriging model to predict test results in other scenarios; Figure 6 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0039] In order to better understand the technical solution of this specification, the embodiments of this specification are described in detail below in conjunction with the accompanying drawings. The embodiments described in the present invention are only some examples of the present invention, not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] The embodiment of the present invention provides a L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation. Figure 1 The experimental platform shown in Figure 1 is constructed by collaborative simulation using SCANeR, Robot Operating System (ROS) and Matlab&Simulink. It integrates the manual driving subsystem (including human-machine interaction equipment), the autonomous driving subsystem and the L3 semi-autonomous driving subsystem, and realizes the takeover judgment, takeover prompt and real-time human-machine control switching of the L3 autonomous driving vehicle.

[0041] Furthermore, the L3 semi-automatic driving subsystem takeover judgment prompt module is defined in this embodiment as: when a vehicle in the side lane cuts in urgently, if the headway is less than 0.8s, it is determined that the driver needs to be reminded to take over. The host vehicle number is recorded as 1, the cut-in vehicle number is 2, and the experimental platform records the host vehicle position in real time as a two-dimensional coordinate (x 1 ,y 1 ), with a velocity of v 1 ; The vehicle position in the side lane is (x 2 ,y 2 ), with a velocity of v 2 ; The subscript corresponds to the vehicle number; The headway between the two vehicles is THW; Further calculate the relative distance between the two vehicles in real time By defining a custom function in SCANeR, it is determined whether THW is less than 0.8s. When it is determined that the driver needs to take over, a signal that the vehicle needs to take over is sent to the ROS system, and the following is displayed in the lower right corner of the multi-screen display: Figure 1 The request takeover window shown in the "Human-Machine Interaction Information Interface" appears, and a beeping tone is emitted for 3 seconds.

[0042] Furthermore, for the L3 level semi-automatic driving subsystem control switching module, the takeover time threshold is set to 3s in this embodiment, and the specific logic of the switching is: the platform starts timing from the moment the takeover prompt is triggered. If the set time threshold is exceeded and the driver has not received a signal that the driver has stepped on the brake pedal and turned the steering wheel more than 2°, it is judged that the driver chooses not to take over the vehicle. At this time, the driving control right is not switched, and the vehicle still makes decisions and controls according to the automatic driving mode and adopts corresponding emergency strategies; if a signal corresponding to the driver's response is received within the time threshold, it is judged that the driver chooses to take over the vehicle, and the vehicle control command is switched to be input by the driving simulation device, and the lower right corner of the multi-screen display is displayed as follows. Figure 1 The manual driving mode window shown in the "Human-machine interaction information interface" indicates that the vehicle is controlled by the driver.

[0043] like Figure 6 As shown, the L3 autonomous driving takeover performance decoupling test method based on collaborative simulation combines scene element cluster sampling to generate a human-vehicle-environment full-element test scenario, and conducts tests based on the collaborative simulation platform. According to the test results, the impact of the single element of human-vehicle-environment on the takeover performance is analyzed, and the transferability of the analysis conclusions in other unknown scenarios is verified.

[0044] Step S1: Based on the collaborative simulation platform, combined with the cluster sampling of scene elements, a full-element test scene of human-vehicle-environment is generated. The specific process is as follows: S11. Sort out the elements of the autonomous driving simulation test scenario and different values ​​according to the needs. Figure 3 As shown, in this embodiment, two continuous variable elements, headway and relative distance, are selected and discretized. The headway is 0.2s, 0.4s, 0.6s, and 0.8s, and the relative distance is 15m, 20m, and 25m.

[0045] Optionally, the importance indexes of the two scene elements and their different values ​​are analyzed based on the hierarchical analysis method. A hierarchical structure model is constructed according to the scene elements in step S11, a judgment matrix is ​​constructed based on the hierarchical affiliation using the Delphi method, etc., and a hierarchical single sorting and consistency test are performed, so as to finally obtain the importance indexes of the headway, relative distance and their different values.

[0046] Optionally, an element occurrence probability constraint is introduced so that specific elements cannot coexist in the same scenario at the same time. For example, in this embodiment, if there is a speed limit on the road, the combination of relative distance and headway time needs to consider the speed issue to avoid the generation of simulation scenarios of unrealistic scenarios.

[0047] S12. Since the importance indexes of different elements and their values ​​are not in the same reference system, a relative importance index is introduced to put all scene elements into the same reference system to obtain an intuitive comparison of the importance indexes of all factors or their values. In addition, it is necessary to introduce element occurrence probability constraints so that specific elements cannot coexist in the same scene at the same time, so as to avoid the generation of simulation scenarios of unrealistic scenarios.

[0048] S13. Use a two-factor randomized experimental design to find a test case set to cover all possible scenario element pairs. The test case set is represented by the following matrix: The test case set includes 12 test scenarios, each involving 2 traffic elements, among which, f i1 and f i2 Represents the values ​​of the first and second traffic elements in the i-th test scene generated, where i is 1, 2, …, 12.

[0049] Step S2: Based on the generated traffic risk scenario test case set, test on the collaborative simulation platform, collect takeover performance data, decouple the impact of key factors on takeover performance, and analyze the decoupling regulation effect of a single factor on takeover performance. The specific process is as follows: S21. Build the traffic environment subsystem, including static traffic environment and dynamic traffic environment settings. In the driving simulation software SCANeR, according to the user-defined test road line shape, set it as a two-way four-lane road, import and add signs, markings, roadside buildings and other object elements to achieve simulation modeling of the static traffic environment; import the basic simulation environment vehicle group with vehicle dynamics into the simulated traffic environment, set the vehicle cut-in trigger mode and path and trajectory planning rules according to the test case set, and achieve simulation modeling of the dynamic traffic environment.

[0050] Optionally, based on test requirements, the time series data such as position, speed, acceleration, heading angle, etc. of each vehicle in the same traffic environment and the driver's behavior data such as steering wheel angle, accelerator pedal force, brake pedal force, etc. can be exported from SCANeR in the form of a csv file.

[0051] S22, such as Figure 4 As shown in the figure, the driver takeover performance indicators are calculated based on the experimental data collected, including the braking reaction time and steering reaction time of the takeover, and according to the formula Calculate the minimum pre-collision time before takeover. In this embodiment, the evaluation of takeover performance is defined as follows: when the braking reaction time or the steering reaction time is less than 3s, that is, when the L3 level semi-automatic driving subsystem control switching module determines that the driver has taken over the vehicle and transfers the vehicle control to the driver, if the minimum pre-collision time TTC before takeover MinIf the time is less than 1.5s, the driver is recorded as successfully taking over the vehicle, otherwise it is recorded as the driver's takeover failure. Further, let the total number of drivers participating in the scenario test be N, and the number of drivers who failed to take over be N', then the takeover failure probability in this scenario is The driver takeover performance in different test scenarios is evaluated by the takeover failure probability P.

[0052] S23. Based on the test results, taking the traffic elements involved in the scenario as the input independent variables and the driver takeover failure probability as the output dependent variable, analyze the coupling relationship of the human-vehicle-environment interaction in the human-machine co-driving system, identify the influencing factors of the takeover performance, decouple the impact of key elements on the takeover performance, analyze the decoupling and regulatory effect of a single traffic element on the takeover performance, and further focus on identifying the key elements that affect poor takeover performance.

[0053] Step S3: Based on the spatial interpolation method of the Kriging model, the observed data in the existing test scenario are used to predict the response in the unknown experimental scenario to verify the transferability of the decoupling relationship analyzed in the second step to other scenarios. The specific process is as follows: S31. Construct a training sample library for the Kriging model based on the decoupling test. 1 ,x 2 ,…,x i ,…,x 12 They are respectively the test scenario use case set matrix T 12×2 The 1st, 2nd, ..., i, ..., 12th rows of , correspond to the twelve test scenario use cases: T 1,: ,T 2,: ,…,T i,: ,…,T 12,: , construct the scene element independent variable matrix X = {x 1 ,x 2 ,…,x i ,…,x 12 Similarly, according to the test results, the corresponding sample response value (driver takeover failure probability in different scenarios) matrix Y = {y 1 ,y 2 ,…,y i ,…,y 12}. The subscript i of x and y both represent the i-th test scenario use case. In this embodiment, x i is a 1×2 row vector matrix, including the values ​​of two key scene elements: vehicle relative distance R and headway THW. For example, in test scenario 1, the vehicle relative distance R = 25m, and the headway THW = 0.8s, then x 1 ={25,0.8}. Some example values ​​of the independent variable scenario elements and the dependent variable driver takeover failure probability are as follows: Figure 5 (a) shown.

[0054] S32. Estimate model hyperparameters based on the training sample library and construct a Kriging model. The Kriging model assumes that all observed data have a mean of μ and a variance of σ. 2 The Kriging model is expressed as follows: Where wF(X) and wf i (X) represents the polynomial regression part, which provides a global approximation of the fitting model, and w is the weight of the polynomial regression term; the nonparametric random part Z(X) represents the Gaussian stationary random process model, which provides a local approximation of the fitting model.

[0055] Consider two points x k ,x q , the subscripts k and q represent the kth test scene and the qth test scene respectively. When the relative distance between two points is small, it is considered that y(x k ) and y(x q ) are highly correlated, and the function describing the correlation is as follows: The covariance is then defined as: From this we can see that the hyperparameters of the model are w, μ, σ 2 ,θ l and p l (l=1,2,…,d, d represents the dimension of the design value), the maximum likelihood estimate of the above parameters can be obtained by using the response value Y of the observed data, and m' is the number of known samples.

[0056] S33. Based on the prior hyperparameters estimated in step S32, a Kriging prediction model for the input traffic element independent variables and the driver takeover failure probability dependent variable is established. The unknown traffic scene element data is denoted as x*. The model can be used to predict the response value Y(x*) of the unknown data point, where * indicates that the response value Y(x*) of the scene is unknown. In this embodiment, x* is a 1×2 matrix, including the values ​​of two key scene elements: the relative distance R of the vehicle and the headway THW, and the value has not been tested on the collaborative simulation platform.

[0057] Among them, Corr T (x*) is the basic data x of the unknown traffic scene element input ★ With m' known sample points X={x 1 ,x 2 ,…,x k ,…,xm '}, the meaning of other characters is the same as #Formula 3: Corr T (x * )=[Corr(x 1 ,x * ),Corrr((x 2 , x * ),...,Corr(x m′ ,x * )] (#Formula7) According to the above formula, input the new basic data of traffic scene elements x ★ , the driver takeover failure probability under the scenario parameter conditions is predicted. Similarly, different untested values ​​of relative distance and headway are input for prediction, and the final result is as follows Figure 5 The prediction results shown in (b) are used to further verify the transferability of the decoupling regulatory effect of the analyzed traffic factors on takeover performance to other unknown scenarios.

[0058] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A L3 autonomous driving takeover performance decoupling test method based on collaborative simulation, characterized in that: Includes steps: S1: Combine scene element cluster sampling to generate a set of test scenarios for all elements of human-vehicle-environment; S2: Conduct decoupling tests based on the collaborative simulation platform and analyze the decoupling relationship; S3: Verify the transferability of the decoupling relationship in step S2 in unknown scenarios.

2. A method for decoupling performance of L3 autonomous driving takeover based on collaborative simulation as claimed in claim 1, characterized in that: The collaborative simulation platform is a test system built by collaborative simulation of SCANeR, robot operating system ROS and Matlab & Simulink, including manual driving subsystem, autonomous driving subsystem and L3 semi-autonomous driving subsystem; The collaborative simulation platform establishes a time synchronization mechanism in a traffic environment, realizes high-fidelity, low-latency human-vehicle collaborative control and control switching of L3 autonomous driving vehicles, and real-time interaction between driving simulation vehicles and environmental vehicles.

3. A collaborative simulation-based L3 autonomous driving takeover performance decoupling test method as claimed in claim 2, characterized in that: The manual driving subsystem of the collaborative simulation platform includes a human-computer interaction device, which transmits the signal of the manual driving operation hardware device back to the simulation platform system through ROS Toolbox, determines the operation of the vehicle in the manual driving mode in SCANeR, and participates in real-time interaction.

4. The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation as claimed in claim 2, characterized in that: The autonomous driving subsystem of the collaborative simulation platform performs risk state perception, decision planning and vehicle control based on environmental vehicle data around the autonomous driving vehicle, including a perception module, a decision planning module and a control module; The ROS system can transmit environmental perception data and control algorithm instructions between nodes to complete the control of autonomous driving vehicles.

5. The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation as claimed in claim 2, characterized in that: The L3 semi-automatic driving subsystem of the collaborative simulation platform integrates the functions of the automatic driving subsystem and the manual driving subsystem, and adds a takeover determination prompt module and a control right switching module; the takeover determination prompt module and the control right switching module set the trigger event for the vehicle to send a prompt requesting the driver to take over according to the vehicle ODD, or calculate the trigger determination index of the takeover prompt for risk assessment. When it is determined that the driver needs to take over the vehicle control, the prompt is triggered and the control right is switched to the driver based on the driver's response.

6. The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation as claimed in claim 1, characterized in that: In step S1, the generation of a set of use cases for the full-factor test scenarios of human-vehicle-environment includes the following steps: S11. Analyze n autonomous driving simulation test scenario elements and different values; S12. Ensure that the interactions among n factors are fully covered through a combined testing method, so as to ensure that the impact of the interactions of these elements on the L3 autonomous driving takeover performance is detected during the simulation experiment; S13. Find a test case set as small as possible to cover all possible scenario element pairs, and obtain a smaller test case set, thereby reducing the cost of testing and analysis.

7. A collaborative simulation-based L3 autonomous driving takeover performance decoupling test method as claimed in claim 6, characterized in that: The step S11 further includes: Analyze the importance index of each scene element and its value; A relative importance index is introduced for intuitive comparison, and element occurrence probability constraints are introduced so that specific elements cannot coexist in the same scenario at the same time, avoiding the generation of simulation scenarios that are not realistic.

8. The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation as claimed in claim 1, characterized in that: In step S2, a decoupling test is carried out on the collaborative simulation platform, including the following steps: S21. Build the traffic environment subsystem, set up static traffic environment and dynamic traffic environment, and conduct tests; S22, calculating the driver takeover performance index based on the experimentally collected data; S23. Taking traffic factors as input independent variables and driver takeover performance as output dependent variables, analyze the coupling relationship of human-vehicle-environment interaction in the human-machine co-driving system, decouple the impact of key factors on takeover performance, and analyze the decoupling regulatory effect of a single traffic factor on takeover performance.

9. The L3 level autonomous driving takeover performance decoupling test method based on collaborative simulation as claimed in claim 1 is characterized in that: In step S3, verifying the portability in an unknown scenario includes the following steps: S31, constructing a training sample library of a Kriging model based on a decoupling test, including an independent variable matrix representing traffic elements in a tested scenario and a corresponding sample response value matrix; S32, estimating model hyperparameters according to the training sample library and constructing a Kriging model; S33. Based on the hyperparameters estimated in step S32, a Kriging prediction model for the input traffic element independent variables and the driver takeover performance dependent variable is established. The model is used to predict the driver takeover performance under other scenario parameter conditions, thereby verifying the transferability of the decoupling regulatory effect of the analyzed traffic elements on the takeover performance in unknown scenarios.