An intelligent driving test method based on multi-background traffic participant interaction

Abstract

The application relates to an intelligent driving test method based on multi-background traffic participant interaction, which comprises the following steps: acquiring a test scene and path information of traffic participants under the test scene, wherein the traffic participants include a test vehicle and various background traffic participants, and the background traffic participants include background vehicles and background pedestrians; constructing a kinematics model of the background traffic participants based on the path information; dynamically constructing safety constraints between any two traffic participants based on the path information; controlling the background traffic participants to collide with the test vehicle based on the kinematics model and the safety constraints, and constructing a model predictive control to obtain a reference speed of the background traffic participants under collaborative interaction; collecting real driving data of the test vehicle and the background traffic participants in real time, and obtaining a test result of the test vehicle through a multi-dimensional test performance evaluation index based on the reference speed and the path information. Compared with the prior art, the application has the advantages of accurate test result, portability and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, in particular to an intelligent driving test method based on multi-background traffic participant interaction. BACKGROUND

[0002] With the development of intelligent and connected vehicle technology, vehicles can interact and coordinate with each other, enabling multi-vehicle coordinated control. Through vehicle-road cooperative automatic driving, the perception range can be expanded and the perception ability can be improved, realizing group intelligence. To ensure that vehicles can reliably achieve automatic driving in real mixed traffic environments, comprehensive automatic driving vehicle testing is needed. Such comprehensive testing can simulate real traffic scenarios and comprehensively evaluate the performance of automatic driving vehicles in diversified and complex road conditions.

[0003] Currently, there is a lack of intelligent and connected driving ecological environment, and it is difficult to implement high-level intelligent and connected vehicle coordination functions. It is urgent to use a vehicle-road-cloud integrated transportation system demonstration base to conduct comprehensive testing of automatic driving vehicles in mixed traffic environments. Testing is not limited to software function evaluation, but also includes comprehensive performance testing of automatic driving vehicles in real road environments. Therefore, complex scenarios and tasks need to be designed to test the reliability and effectiveness of automatic driving vehicles. As the level of automatic driving vehicles improves, the number of test scenarios and tasks is also increasing, and high reliability, efficiency and safety are required for intelligent and connected vehicle testing systems.

[0004] Closed route testing allows automatic driving vehicles to be tested in a controlled environment with controllable risks and real simulation of open traffic scenarios. In addition, through closed road testing, test personnel or safety personnel can familiarize themselves with the operation procedures, testing methods and takeover procedures of automatic driving in dangerous conditions, facilitating the transition to open road testing. Looking at existing mature intelligent and connected vehicle test sites, they mainly focus on performance and efficiency testing of single vehicles in simple scenarios, lacking comprehensive testing of vehicles in strongly coupled, multi-agent mixed traffic system environments. Due to the infinite variety of test scenarios, an ideal testing system should be highly flexible and intelligent, capable of conducting comprehensive testing of automatic driving in complex traffic environments by designing continuous and diverse testing conditions. The system also needs to ensure safe interaction between multiple traffic participants while conflicting with the tested automatic driving vehicle, and can automatically generate edge danger scenarios to test the comprehensive performance and potential problems of the tested automatic driving vehicle.

[0005] In addition, comprehensive testing requires a large amount of test data, including sensor data, vehicle state data, scenario information, etc. At the same time, these data need to be effectively processed and analyzed to extract valuable information and support system improvement. Therefore, the acquisition and processing of test data is a challenging task.

[0006] Real-world comprehensive testing of autonomous vehicles plays an important role in ensuring system safety and driving performance. Such testing can provide real-world scenarios and challenges, helping to verify the reliability and adaptability of autonomous driving systems under diverse road conditions and complex traffic environments. There is an urgent need for comprehensive testing of autonomous vehicles in mixed traffic environments to verify system performance, identify problems and improve them, and promote the further development and commercial application of autonomous driving technology. SUMMARY

[0007] The purpose of the present application is to provide an intelligent driving test method based on multi-background traffic participant interaction to improve test reliability.

[0008] The purpose of the present application can be achieved by the following technical solutions:

[0009] An intelligent driving test method based on multi-background traffic participant interaction, comprising the following steps:

[0010] Obtaining test scene and path information of traffic interactors under the test scene, the traffic interactors including a test vehicle and various background traffic participants, the background traffic participants including background vehicles and background pedestrians;

[0011] Building a kinematic model of the background traffic participants based on the path information;

[0012] Based on the path information, dynamically building safety constraints between any two traffic interactors;

[0013] Based on the kinematic model and safety constraints, controlling the background traffic participants to conflict with the test vehicle, and building a model predictive control to obtain the reference speed of the background traffic participants under collaborative interaction;

[0014] Real-time collection of real driving data of the test vehicle and the background traffic participants, and obtaining test results of the test vehicle through multi-dimensional test performance evaluation indexes based on the reference speed and path information.

[0015] Further, the kinematic model is a discrete form of a second-order longitudinal dynamics equation, expressed as:

[0016]

[0017] In the formula, s m (i), v m (i) and a m (i) represent the position, speed and acceleration of the mth background traffic participant at the ith time, and Δt is the length of the predefined time interval.

[0018] Further, the step of dynamically constructing the safety constraints comprises:

[0019] Based on the path information, collision areas and potential interaction areas of the test vehicle, the background pedestrians and the background vehicles are respectively constructed;

[0020] Based on the collision areas and the potential interaction areas, the overlapping interaction area of any two traffic interactors is calculated;

[0021] Based on the overlapping interaction area, the safety constraint between any two traffic interactors is constructed.

[0022] Further, the overlapping interaction area of any two traffic interactors is calculated by using the Monte Carlo method, and the calculation expression of the overlapping interaction area is:

[0023]

[0024] In the formula, is the overlapping area of the traffic interactor m and the traffic interactor n, N mn is the number of points uniformly distributed in the potential interaction overlapping area of the traffic interactor m and the traffic interactor n, N m is the total number of points uniformly distributed in the potential interaction area of the traffic interactor m, is the potential interaction area of the traffic interactor m.

[0025] Further, the safety constraint comprises a collision area overlapping area constraint and a potential interaction area overlapping area constraint, which are respectively:

[0026]

[0027]

[0028] In the formula, · (j|i) is a prediction value at a time step j+i at the current time i, is the prediction of the collision area overlapping area of the traffic interactor m and the traffic interactor n, N v is the number of background traffic participants, is the potential interaction cost in the prediction step, is the prediction of the overlapping area between the potential interaction areas of the traffic interactor m and the traffic interactor n, ω v is the relative adjustment weight of the traffic interactor speed and the potential interaction area overlapping area for the safety interaction of the two traffic interactors, v m (j|i), v n (j|i) are the predictions of the speeds of the traffic interactor m and the traffic interactor n, respectively, is a step function that takes value 1 when p < q and 0 otherwise.

[0029] Further, the step of obtaining the reference speed comprises:

[0030] calculating the Euclidean distance between the test vehicle and the background traffic participants, and controlling the background traffic participants to actively generate conflicts with the test vehicle based on a kinematic model;

[0031] based on the conflicts and safety constraints, constructing and solving a discrete form of model predictive control problem to obtain the reference speed.

[0032] Further, the calculation expression of the Euclidean distance is:

[0033]

[0034] where, · (j|i) is the predicted value at time step j+i at current time i, d col (j|i) is the Euclidean distance between the test vehicle and the nearest background traffic participant, N v is the number of background traffic participants, x test (j|i), y test (j|i) is the predicted value of the test vehicle coordinate, x m (j|i), y m (j|i) is the predicted value of the nearest background traffic participant m coordinate.

[0035] Further, the expression of the model predictive control problem is:

[0036]

[0037]

[0038] s.t.

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] where, · (j|i) is the predicted value at time step j+i at current time i, N v is the number of background traffic participants, Np To predict the horizontal step size, ω a As the adjustment weight for the comfort item, a m (j|i) represents the predicted acceleration value of background traffic participant m, ω c Adjustment weights for safety assurance items, Given the potential interaction constraints of background traffic participant m at time j+1+i, ω col d is the adjusting weight for conflict-generating terms. col (j+1|i) represents the Euclidean distance between the test vehicle and background traffic participants at time j+1+i. To provide a performance index for timely and rapid driving by traffic participant m under traffic rules. m,max Let s be the furthest distance of the m-th background traffic participant. m (N p |i) represents the m-th background traffic participant in N p The position at time +i, s c,m The origin point where background traffic participants stop. For the set of background traffic participants, s m (i), v m (i) and a m (i) represents the position, velocity, and acceleration of the m-th background traffic participant at time i, where Δt is the length of a predefined time interval. For predicting the overlap area of ​​the collision regions of traffic interactors m and n, s m (0|i), v m (0|i) represent the initial state values ​​of the m-th background traffic participant in the prediction time domain, s c,m (i), v c,m (i) represents the position and speed of the background traffic participant at the origin, x m (j|i), y m (j|i) represents the predicted location of background traffic participant m, α m (j|i) represents the predicted heading angle of background traffic participant m. The reference path of traffic participant m is defined in a geometric lookup table in the "location" space, where the location is determined by the distance s traveled along the reference path. m (j|i) represents the minimum and maximum accelerations of background traffic participant m at time j+i, where amin and amax are the minimum and maximum values, respectively. The speedometer for traffic participant m in the background.

[0046] Furthermore, the reference velocity is solved using a k-level inference method.

[0047] Further, the multi-dimensional test performance evaluation index includes a perception performance evaluation index, a decision performance evaluation index, a control performance evaluation index, an interaction performance evaluation index, an actual scene simulation index, a system robustness evaluation index, and a safety evaluation index.

[0048] Compared with the prior art, the present application has the following beneficial effects:

[0049] (1) The present application dynamically constructs safety constraints under complex traffic scenes, controls the active formation of conflicts between the test vehicle and the traffic background participants under the condition of ensuring safety among each other, predicts the reference speed and reference path under multi-vehicle interaction through model predictive control in the edge dangerous conflict scene, tests the comprehensive performance of the test vehicle in the strong coupling and multi-agent mixed traffic system environment, and improves the test reliability.

[0050] (2) The safety constraints of the present application find the sufficient and necessary conditions for vehicle lateral interaction safety through accurate modeling and judgment, provide real-time safety guarantee for decision variables in multi-vehicle cooperative control, thereby ensuring the safety of vehicle lateral interaction in the prediction time domain, and provide a key safety basis for the actual application of multi-vehicle cooperative driving.

[0051] (3) The present application optimizes the multi-vehicle cooperative speed prediction based on the k-level reasoning principle, considers the initial value of the to-be-optimized vehicle and the optimal speed of other vehicles when optimizing the speed of each vehicle, realizes the dimension reduction of the control quantity and the convergence of the global optimal speed, improves the speed optimization efficiency, and realizes real-time speed control.

[0052] (4) The present application uses multiple evaluation indexes to comprehensively evaluate from multiple dimensions, and the evaluation result is more comprehensive and accurate. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a method flowchart of the present application;

[0054] Figure 2 is a safety interaction region definition diagram of an embodiment of the present application, wherein (a) is a full interaction region definition diagram of a vehicle, and (b) is a full interaction region definition diagram of a background pedestrian;

[0055] Figure 3 is a communication protocol framework diagram of an embodiment of the present application;

[0056] Figure 4 is a data flow transmission schematic diagram of an embodiment of the present application;

[0057] Figure 5 is an iterative calculation method schematic diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0058] The application will be described in detail below with reference to the drawings and specific embodiments. The embodiments are implemented on the premise of the technical solutions of the application, and detailed implementation modes and specific operation processes are given, but the protection scope of the application is not limited to the following embodiments.

[0059] The embodiment provides an intelligent driving test method based on multi-background traffic participant interaction, as shown in the figure, the method comprises the following steps: Figure 1

[0060] Step 1: Determine the test scene, input the path information corresponding to each background traffic participant and pre-process.

[0061] Firstly, the test scene is determined, and the test path is generated based on a typical test scene or a real test accident case. The test path information includes the x coordinate of the reference path in the UTM (Universal Transverse Mercartor Grid System, Universal Transverse Mercartor Grid System) coordinate system, the y coordinate of the reference path in the UTM coordinate system and the like. After removing the repeated points of the input test scene path, the test scene path is interpolated into coordinate points uniformly distributed on the path. According to the positioning coordinates, the heading angle and the like uploaded by the mth traffic participant, the position representation on the reference path is established, and the horizontal coordinate is defined as x m , the vertical coordinate is defined as y m , and the heading angle is defined as α m , which is used to represent the position of the mth traffic participant on the reference path. x r , y r , and α r represent the predefined reference path that the mth traffic participant should track. The reference path of each traffic participant is defined as a geometric lookup table in the "position" space where the position is represented by the distance s m (i) along the reference path.

[0062]

[0063] Step 2: Kinematics model of background traffic participant.

[0064] In the embodiment, a group of controlled background participants drive along the reference path , where N v is the number of controlled background participants.

[0065] We establish a discrete form of second-order longitudinal dynamics equation for the mth traffic participant:

[0066] ​

[0067] Where Δt > 0 represents the length of a predefined time interval; s m (i), v m (i) and a m (i) represents the position, velocity, and acceleration of the m-th traffic interaction participant at time i.

[0068] Step 3: Dynamically construct security constraints.

[0069] To address the challenge of ensuring the safety of traffic interaction corresponding to decision variables in the prediction time domain during multi-vehicle cooperative control, this study investigates lateral collision avoidance criteria for multi-traffic interaction. Through precise modeling and judgment, it identifies the necessary and sufficient conditions for lateral interaction safety among multiple traffic interactors, providing real-time safety assurance for decision variables in multi-vehicle cooperative control. This ensures lateral interaction safety in the prediction time domain and provides a crucial safety foundation for the practical application of multi-traffic interaction cooperative driving.

[0070] Step 3 includes the following specific steps:

[0071] Step 3.1: Define the collision area and potential interaction area.

[0072] When assessing a potential collision between two vehicles from a bird's-eye view, the vehicles are conceptualized as two-dimensional rectangular entities with defined planar dimensions. A rectangle equal in size to the actual length and width of the vehicle is defined as the "collision zone." The distances between the vehicle sensors and the front and rear ends of the rectangle are denoted as l, respectively. c,f and l c,r The sensor is located on the vehicle's lateral centerline. The vehicle's position, heading θ (defined as north, increasing clockwise), and width w are also measured. c It is known. The collision area (c-area) of the vehicle can be represented by a 6-tuple (x, y, θ, l). c,f ,l c,r ,w c The term ) is used to represent the vehicle's "potential interaction area" (p-area). Additionally, a "potential interaction area" (p-area) is defined to represent a rectangle that shares the same longitudinal line of symmetry as the vehicle's c-area, and extends beyond the c-area with a safety margin. See [link to p-area]. Figure 2 As shown in (a) above. The width of the potential interaction area is set to w. p This distance is typically smaller than the lane width. We denote the distance between the vehicle sensor and the front / rear ends of the rectangle as L, respectively. p,f and L p,r Because the speed varies greatly depending on the direction of travel, an L-shaped safety feature should be set to ensure safe vehicle interaction. p,f and l c,f Greater than L p,r and l c,r A 6-tuple (x, y, θ, l) p,f ,lp,r ,w p ) for describing the potential interaction area of a vehicle. Similarly, a background pedestrian is defined as a two-dimensional rectangular entity of planar dimensions, and a rectangular box of size equal to the actual length and width dimensions of the background pedestrian's moving chassis is defined as the "collision area". A "potential interaction area" (p-area) of the background pedestrian is also defined to represent a rectangle that shares the same longitudinal symmetry line as the c-area of the background pedestrian and exceeds the c-area by a safety margin, as shown in (b) of FIG. 6. Figure 2

[0073] Step 3.2: Overlapping interaction area calculation.

[0074] The Monte Carlo method is used to calculate the overlapping area between two vehicles. Specifically, points uniformly distributed within the potential interaction overlapping area of the traffic interactors m and n are used to approximate the target value. The overlapping area is calculated by the formula N mn is the number of points uniformly distributed within the potential interaction overlapping area of the traffic interactors m and n, N m is the total number of points uniformly distributed within the potential interaction area of the traffic interactor m, A m is the potential interaction area of the traffic interactor m. Collision area zero overlap represents a safe interaction without actual collision. On the contrary, an overlapping area greater than zero indicates the existence of a collision. This method comprehensively evaluates the collision risk based on the spatial relationship and direction of the vehicles in a two-dimensional plane. The collision area overlapping area between the m traffic interactor and the n traffic interactor is denoted by , while the potential interaction area overlapping area is denoted by .

[0075] Step 3.3: Potential interaction cost function establishment.

[0076] In the optimization problem, different constraints are applied to the p-area and c-area to ensure that the interaction between multiple traffic interactors is safe. The overlapping area of the c-area of two traffic interactors must be strictly equal to 0, i.e. When , it means that a collision has occurred, so the penalty weight in the cost function will be very large to avoid collision. Although it is required that the overlapping area of the c-area of two traffic interactors must be equal to zero, the potential interaction area allows limited overlap. J p is a reward function used to describe the potential interaction between multiple traffic interactors.

[0077]

[0078] where, represents the prediction of the overlapping area between the p-area of traffic interactors m and n.​ is defined as an indicator function that takes the value 1 when p < q and 0 otherwise.v m (j|i) and v n (j|i) are the speeds of traffic interactors m and n in the prediction horizon, respectively. When the p-area overlap of any two traffic interactors is zero, it means there is no potential interaction, thus J p = 0. When , it means there is a risk of collision. The penalty size for collision depends on the p-area overlap area and the speeds of the traffic interactors. Specifically, the higher the absolute values of v m (j|i) and v n (j|i), the higher penalty is assigned, reflecting the possibility of potentially more severe collision. The weight ω v > 0 adjusts the relative influence between the overlap area and the speeds. In addition, the inclusion of 1 ensures a base penalty in the case of relatively low overlap area or speed.

[0079] Step 4: Multi-traffic interactor cooperative speed prediction optimization.

[0080] Step 4.1: Upload the position, speed, and heading angle data of the test car, background cars, and background pedestrians to the cloud.

[0081] The cloud server communicates with the background traffic participants through the MQTT (Message Queue Telemetry Transport) protocol, as shown in Figure 3 , which can meet the requirements of real-time control. The vehicle end and the background pedestrians upload information such as time stamp, x-coordinate in UTM coordinate system, y-coordinate in UTM coordinate system, heading angle, and actual speed to the cloud, as shown in Figure 4 .

[0082] Step 4.2: Define conflict automatically generated representation.

[0083] To achieve the expected effect of testing the autonomous vehicle, the background vehicles and pedestrians need to actively form conflicts with the test vehicle as much as possible to test the comprehensive performance of the autonomous vehicle in a strong coupling, multi-agent mixed traffic system environment.

[0084] To prevent the occurrence of deadlocks leading to traffic congestion, the test vehicle is selectively controlled to only conflict with the closest background traffic participant.d col represents the Euclidean distance between the test car and the closest background traffic participant.

[0085]

[0086] The speed of the Vehicle Under Test (VUT) and the surrounding traffic interactors is predicted according to the second-order longitudinal dynamics equation in Step 2. In this way, the set of all positions and speeds that the traffic interactors can reach is obtained. The distance d col is added to the cost function, and by minimizing the cost function, the proximity of the background traffic participants to the VUT forms a conflict.

[0087] The proximity of the distance between the background traffic participants and the VUT, i.e., the intensity of the test scenario, is adjusted by the weight ω col . It is noted that the penalty ω col associated with the distance cost function is less than the safety penalty, thereby automatically generating conflict scenarios and ensuring the safety of multi-vehicle cooperative interaction. The intensity of the test scenario can be adjusted according to ω col . The greater ω col , the more intense the conflict, thereby simulating the driving behavior of the test vehicle under different traffic conditions.

[0088] By this method, the response ability and adaptability of the vehicle under test in extreme situations can be tested, and the operating limits of the vehicle under test under different background vehicle distances and trigger times can be determined. By studying the boundary conditions for safe operation of the vehicle under test in depth, the comprehensive performance of the autonomous vehicle can be understood.

[0089] Step 4.3: Multi-vehicle cooperative speed prediction optimization.

[0090] In order to solve the problem of cooperative control and conflict generation in all possible complex scenarios, including crossroads, lane changes, ramps, and roundabouts, the control problem established is not limited to a specific scenario. The background traffic participants consider the interaction with other traffic interactors when evaluating actions. Let m represent the ego vehicle, and n represent another traffic interactor interacting with m. Considerations include safety constraints, control input constraints, and state constraints involving speed and travel distance.

[0091] A discrete form of MPC (Model Predictive Control) formula is established. At each time instant i, the following control problem is solved to obtain the optimal reference speed.

[0092]

[0093] s.t.

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100] In the formula, ·(j|i) represents the predicted value at time step j+i at the current time i. The acceleration sequence used by the model is... Where N p For predicting the time-domain step size; s m,0 ,v m,0 To predict the initial state value of the m-th background traffic participant in the time domain; a min and a max These are the minimum and maximum acceleration limits for background traffic participants. To avoid discomfort caused by sudden bumps or violent movements, we can minimize a. m (j|i) 2 This makes the movement of background traffic participants smoother.

[0101] The current maximum speed limit is based on the current location by looking up the speedometer. Received. v max (m,s m (i) represents the m-th background traffic participant at location s. m The maximum speed at (i) is calculated offline based on the path curvature to improve real-time computation speed.

[0102] To prevent deadlocks caused by multiple background traffic participants, J dis This item is used to improve driving efficiency and is defined as a performance indicator for timely and rapid driving under traffic rules, namely...

[0103]

[0104] Among them, s m (N p |i) represents the m-th background traffic participant in N p Position and velocity at time +i For j = 1, 2, ..., N p That is, the m-th background traffic participant travels at the maximum speed permitted by traffic rules and road restrictions, ignoring other background traffic participants. Note J dis ∈[0,1]. Where, J dis =0 indicates the most active driving, reaching the farthest distance s. m,max J dis =1 indicates that it remains at the origin s. c,m .

[0105] Step 4.4: Cloud-based multi-vehicle collaborative optimization based on k-level inference.

[0106] Considering the initial values ​​of the background traffic participants to be optimized and the optimal speeds of other background traffic participants, the speed of each background traffic participant is optimized to achieve convergence of the globally optimal speed. In the optimization process, we initially assume that each background traffic participant has an initial speed value. In each iteration, the first step involves optimizing the speed of the first background traffic participant. This optimization considers the initial values ​​of other background traffic participants, thus determining the optimal speed of the first background traffic participant. Subsequently, in the second step, the second background traffic participant optimizes its speed based on the optimal speed of the first background traffic participant and the initial values ​​of all other background traffic participants. This process is repeated iteratively, with each subsequent background traffic participant optimizing its speed by considering the optimal speeds of previous background traffic participants and the initial values ​​of the remaining background traffic participants.

[0107] In the optimization process for the m-th background traffic participant, term J p {m} and d col It depends on the states of other traffic interactors. The cost function is expressed as:

[0108]

[0109] Among them, a m This represents the m-th background traffic participant being optimized; the state variables have been omitted for brevity. This represents the optimal values ​​a1*,...,a obtained for background traffic participants 1 to m-1 in the previous steps. m-1 *With the current iteration from m+1 to N v The initial value a of the background traffic participants m+1 ini ,..., The mixture, that is J p {m} and d col The optimization involves solving for a for the m-th background traffic participant. m ,and All terms in J are fixed. Therefore, J p {m} and d col Dimensionality reduction of control variables was achieved.

[0110] This sequential optimization continues until all N are reached. v Each background traffic participant is individually optimized to obtain their own optimal speed. The next iteration begins with each background traffic participant using their optimal speed obtained in the previous round as their new initial value. This iterative process gradually converges towards the global optimum. The iterative process is as follows:Figure 5 As shown in the figure, the circles represent the mth background traffic participant being optimized, the m+1 to Nth background traffic participants obtaining initial values or optimal values in the last iteration, and the 1st to m-1th background traffic participants that have been optimized and obtained optimal values. v

[0111] Based on the k-level reasoning principle, when optimizing the speed of each background traffic participant, the initial value of the background traffic participant to be optimized and the optimal speed of other background traffic participants are considered, realizing dimension reduction of the control quantity and convergence of the global optimal speed, improving the speed optimization efficiency, and realizing real-time speed control.

[0112] Step 4.5: The cloud issues the reference speed and reference path of the background traffic participant.

[0113] The cloud issues control information such as the reference speed and reference path of the background vehicle and the background pedestrian through the MQTT protocol, including the timestamp, the x-coordinate of the reference path in the UTM coordinate system, the y-coordinate of the reference path in the UTM coordinate system, the heading angle, and the reference speed.

[0114] Step 5: Background traffic participant speed and path tracking.

[0115] First, a lateral dynamics model is established based on the state parameters and uncertainty parameters of the background vehicle and the background pedestrian. This model considers the motion characteristics of different background traffic participants in the lateral direction, including lateral position, lateral speed, lateral acceleration, etc. Based on the established lateral dynamics model of the background traffic participant, a path tracking kinematics model is established. This model defines the lateral position deviation and yaw angle error of path tracking, and expresses the deviation of lateral position, the deviation of yaw angle, and their derivatives as a function relationship of the state of the background traffic participant and the expected path information.

[0116] Then, path tracking servo equation constraints are established using the path tracking kinematics model. This includes defining the equation constraints that the lateral position deviation and yaw angle error tend to zero, and converting the path tracking goal of the background traffic participant into a servo constraint control task. Based on the path tracking servo equation constraints, the background traffic participant lateral dynamics model is controlled to obtain a path tracking controller. This controller can make the lateral position deviation and yaw angle error of the background traffic participant tend to zero, achieving the goal of path tracking. The system obtains feedback information in real time through positioning sensors, and corrects and optimizes path planning and speed planning. This enables the system to adapt to different road conditions and traffic situations, improving the robustness and adaptability of path tracking.

[0117] ​Path tracking control algorithms, including but not limited to: control algorithms based on road geometry principles, such as pure tracking control, Stanley control, Alice control, etc.; path tracking control algorithms based on classical control theory, such as PID control, linear feedback control, etc.; path tracking control algorithms based on modern control theory, such as MPC control, LQR control, etc. These algorithms can be selected and switched according to the actual scene and demand.

[0118] Step 6: Real driving data collection, build scene library.

[0119] First, through the vehicle positioning sensor, the scene path in the actual driving is collected. This process includes using vehicle positioning sensors to obtain the vehicle's position, speed, direction and other information in geographical space. Use in-vehicle sensors to monitor driving behavior, including acceleration, deceleration, sudden braking and other dynamic behaviors. These information can be obtained in real time through the connection with the vehicle's electronic control unit (ECU) or other related sensors.

[0120] Then process and label the collected scene data to ensure the accuracy and usability of the data. The processing process may include denoising, interpolation, data alignment and other steps, while labeling involves annotating specific driving scenes or behaviors. Store the processed and labeled scene data in the database to ensure that it can be retrieved and used at any time during testing. The database uses json or other formats to support effective data management and query.

[0121] Finally, by continuously updating and accumulating scene data in actual driving, the system can reflect driving behavior under different time periods and conditions, increasing the authenticity and representativeness of the test. The test system can use these data to test reproduction or test scene expansion, evaluate the performance and robustness of the autonomous driving system under different scenarios.

[0122] Step 7: Test multi-dimensional performance evaluation.

[0123] The evaluation of autonomous vehicles mainly includes seven aspects: perception performance evaluation, decision performance evaluation, control performance evaluation, interaction performance evaluation, actual scene simulation, system robustness evaluation and safety evaluation.

[0124] In terms of perception performance evaluation, we use the root mean square error (RMSE) to calculate the accuracy difference between the perception system output and the actual data. Coverage and error rate are used to measure the recognition coverage of the target by the perception system, to evaluate the perception range and angle.

[0125] Decision performance evaluation includes planning path accuracy, which is quantified by calculating path deviation and path tracking error. Path error Where, d iis the distance deviation of the i-th point on the path, and N represents all coordinate points on the path. In addition, decision execution and timeliness are measured by decision response time to measure the rapid response of the vehicle to external environment changes.

[0126] Control performance evaluation considers vehicle stability, using the standard deviation of vehicle lateral and longitudinal acceleration for evaluation. Acceleration and braking performance are quantified using average acceleration and braking distance.

[0127] In the interactive performance evaluation, the interaction between vehicles is considered, and the minimum safety distance and interaction time are used to evaluate the vehicle cooperation. Intersection behavior is evaluated by the percentage of following traffic rules and the percentage of safe passing.

[0128] In terms of actual scenario simulation, the diversity and authenticity of the test scenario are evaluated by scenario coverage and driving behavior fidelity. The influence of weather and lighting conditions considers the stability of sensor performance under different environmental conditions.

[0129] System robustness evaluation includes fault recovery capability, using the success rate and time of system automatic recovery to evaluate the fault handling capability of the system. Robustness test evaluates the robustness of the system by the performance stability of the system under different environmental and communication disturbances.

[0130] In terms of safety evaluation, emergency braking and risk avoidance operations are considered, and braking time and risk avoidance success rate are used to evaluate the safety response of the system in emergency situations. Risk prediction evaluates the system's ability to identify potential risks through risk prediction accuracy and false alarm rate.

[0131] Step 8: System reset and scenario scheduling global path planning.

[0132] First, confirm the reset signal, because the scene test is complete or because the cloud receives a reset request signal sent by the background traffic participant (this signal can be a preset hardware trigger, software instruction or abnormal detection triggered signal), and then the cloud judges and initiates the reset instruction.

[0133] The system stops the current task: after receiving the reset signal, the system stops the test task being executed, including path and speed tracking, obstacle avoidance, etc. On the basis of ensuring the safety of each traffic participant, the cloud controls the stop, ensuring that the background traffic participant is in a safe state. The background traffic participant uploads the current state information including position, speed, heading angle and other key parameters to the cloud.

[0134] Then confirm the next test scenario information, receive the scene information from the external environment or internal database, including the next test scenario preset path (including path coordinates, heading angle, etc.) information.

[0135] For scene information analysis, based on map data (OpenDRIVE format, including road network, traffic rules, traffic signs, etc.) and scene definition parameters, a test scene model is established. Select the appropriate path planning algorithm, which can be a graph search-based algorithm, optimization algorithm or deep learning method, etc. According to the dynamics and kinematics characteristics of different traffic interactors, such as motion chassis, dummy motion platform, considering the map and vehicle dynamics, as well as the safety distance and minimum turning radius of the vehicle, etc. Dynamic constraints, ensure that the global path planning method is effectively implemented under dynamic constraints and traffic rules after given random starting point and endpoint.

[0136] Execute the selected path planning algorithm to generate a global path that meets the task requirements and obstacle avoidance conditions. And according to the current real-time location of the background traffic participant, the generated global path will be rolled out in real time to the background traffic participant for execution. In the case of path re-planning when information is lost for a short time and test failure, a robust planning method for moving body driving path is adopted. In the face of information loss, it can actively respond to ensure that the vehicle can still safely and efficiently re-plan the driving path in a dynamically changing environment.

[0137] Finally, the background traffic participant sends a reset completion confirmation signal to the cloud, indicating that the background traffic participant has completed the reset and reached the starting point of the next test scene path, and the next round of testing can be performed.

[0138] Step 9: Standardized extension definition of test scene.

[0139] In order to realize the generalization and transplantation application of path planning algorithm in different test sites and environments, this patent provides a general path planning system and its parameter self-calibration method. The design of this system aims to ensure that under the conditions of given map data format, input parameters, and motion body shape and motion characteristics, path planning for different motion bodies and map data is realized. The following is the specific implementation of the patent:

[0140] First of all, we use a general path planning architecture, which includes global path planning, middle-level local path planning, and lower-level trajectory planning modules. Global path planning is responsible for finding a high-level path from the starting point to the target point in the entire map. The middle-level local path planning generates a more adaptive local path based on the current state of the vehicle and the dynamic changes in the environment. The lower-level trajectory planning is responsible for converting the local path into a specific trajectory that the vehicle can execute.

[0141] Secondly, we introduce a parameter self-calibration method that can adapt to different geographical environments and vehicle characteristics. Through online calibration of input parameters, the system can adjust the parameters of the algorithm to adapt to different map data and motion body conditions, thereby improving the robustness and performance of path planning.

[0142] The key technologies involved also include vehicle end planning control based on regular working conditions, vehicle end planning control based on irregular working conditions, and vehicle end planning control considering time delay uncertainty. These technologies ensure the reliability and safety of the path planning system under various working conditions and environments.

[0143] With the universal path planning architecture and parameter self-calibration method, the path planning algorithm is realized in different geographical environments and vehicle characteristics. This universality and adaptability makes the path planning system more portable, suitable for different test scenarios, and can promote the comprehensive verification and application of automatic driving technology.

[0144] The present application collects real driving data of autonomous vehicles, evaluates the comprehensive behavior of autonomous vehicles, and constructs a scenario library, which helps to enrich the test scenario library, realize the standardized expansion definition of test scenario, help the test system to generate more challenging edge dangerous test scenarios, and help to expand the application to different test sites and different test working conditions.

[0145] The above functions, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the prior art that essentially contributes or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0146] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as object-oriented programming language Java and interpreted scripting language JavaScript.

[0147] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0148] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0150] While preferred embodiments of the application have been described, modifications and variations can be apparent to those skilled in the art once aware of the general underlying concepts. Accordingly, the appended claims intend to embrace all such modifications and variations as fall within the scope of the application.

[0151] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for intelligent driving test based on multi-background traffic participant interaction, characterized in that, The method comprises the following steps: acquiring a test scene and path information of traffic interactors in the test scene, the traffic interactors including a test vehicle and various background traffic participants, the background traffic participants including background vehicles and background pedestrians; constructing a kinematic model of the background traffic participants based on the path information; dynamically constructing safety constraints between any two traffic interactors based on the path information; the step of dynamically constructing the safety constraints comprises: constructing collision regions and potential interaction regions of the test vehicle, the background pedestrians and the background vehicles respectively based on the path information; calculating overlapping interaction region areas between any two traffic interactors based on the collision regions and the potential interaction regions; constructing safety constraints between any two traffic interactors based on the overlapping interaction region areas; the overlapping interaction region areas between any two traffic interactors are calculated by using a Monte Carlo method, and a calculation expression of the overlapping interaction region areas is: wherein is the area of the overlap interaction region for traffic actor m and traffic actor n, N mn is the number of points evenly distributed within the area of the potential interaction overlap region for traffic actor m and traffic actor n, N m is the total number of points evenly distributed within the potential interaction region for traffic actor m, is the area of the potential interaction region for traffic actor m; the safety constraints include collision region overlapping area constraints and potential interaction region overlapping area constraints, and the constraints are respectively: wherein • (j|i) is the prediction of the time step j+i at the current time i, c m，n (j|i) is the prediction of the overlapping area of the collision region of traffic actor m and traffic actor n, N v is the number of background traffic participants, is the potential interaction cost within the prediction horizon, is the prediction of the overlapping area between the potential interaction region of traffic actor m and traffic actor n, ω v is the relative adjustment weight of the traffic actor speed and the potential interaction region overlapping area for the influence size of the safe interaction of the two traffic actors, v m (j|i), v n (j|i) are the predictions of the speed of traffic actor m and traffic actor n, respectively, is an indicator function that takes the value 1 when p Based on the kinematic model and safety constraints, the background traffic participants are controlled to conflict with the test vehicle, and a model predictive control is constructed to obtain the reference speed of the background traffic participants under collaborative interaction. real driving data of the test vehicle and the background traffic participants are collected in real time, and test results of the test vehicle are obtained by using multi-dimensional test performance evaluation indexes based on the reference speeds and the path information. 2.The intelligent driving test method based on multi-background traffic participant interaction of claim 1, wherein, the kinematic model is a discrete second-order longitudinal dynamics equation, and an expression is: where s m (i), v m (i) and a m (i) represents the position, speed and acceleration of the mthbackground road user at the ith time instant, and Δt is the length of the predefined time interval. 3.The intelligent driving test method based on multi-background traffic participant interaction of claim 1, wherein, the step of obtaining the reference speeds comprises: calculating Euclidean distances between the test vehicle and the background traffic participants, and controlling the background traffic participants to actively generate conflicts with the test vehicle based on the kinematic model; constructing a discrete model predictive control problem based on the conflicts and the safety constraints and solving the problem to obtain the reference speeds.

4. The intelligent driving test method based on multi-background traffic participant interaction according to claim 3, characterized in that, a calculation expression of the Euclidean distances is: where • (j|i) is the predicted value at time step j+i at current time i, d col (j|i) is the Euclidean distance between the test vehicle and the closest background traffic actor, N v is the number of background traffic actors, x test (j|i), y test (j|i) is the predicted value of the test vehicle coordinates, x m (j|i), y m (j|i) is the predicted value of the closest background traffic actor m coordinates.

5. The intelligent driving test method based on multi-background traffic participant interaction according to claim 3, characterized in that, an expression of the model predictive control problem is: where • (j|i) is the predicted value at time step j+i at current time i, N v is the number of background traffic participants, p is the prediction horizon, ω a is the adjustment weight of the comfort term, a m (j|i) is the acceleration prediction value of the background traffic participant m, ω c is the adjustment weight of the safety guarantee term, is the potential interaction constraint of the background traffic participant m at j+1+i, ω col is the adjustment weight of the conflict generation term, d col (j+1|i) is the Euclidean distance between the test vehicle and the background traffic participant at j+1+i, is the performance index of the background traffic participant m driving at a timely and fast manner under the traffic rules, s m,max is the farthest distance of the mth background traffic participant, s m (N p |i) is the position of the mth background traffic participant at N p +i, s c,m is the original position where the background traffic participant stays, is the set of background traffic participants, s m (i), v m (i) and a m (i) represent the position, speed and acceleration of the mth background traffic participant at the ith time, and Δt is the length of the predefined time interval, is the prediction of the overlapping area of the collision region of the traffic participant m and the traffic participant n, s m (0|i), v m (0|i) are the initial state values of the mth background traffic participant in the prediction time domain, s c,m (i), v c,m (i) are the original position and speed of the background traffic participant, x m (j|i), y m (j|i) is the position prediction value of the background traffic participant m, α m (j|i) is the heading angle prediction value of the background traffic participant m, is a geometric lookup table where the position is defined by the distance s traveled along the reference path of the background traffic participant m, where m (j|i) represents, a min , a max are the minimum and maximum values of the acceleration of the background traffic participant m at j+i, respectively, is the speed table of the background traffic participant m.

6. The intelligent driving test method based on multi-background traffic participant interaction of claim 1, wherein, the reference speeds are solved by using a k-level reasoning method.

7. The intelligent driving test method based on multi-background traffic participant interaction of claim 1, wherein, the multi-dimensional test performance evaluation indexes include perception performance evaluation indexes, decision performance evaluation indexes, control performance evaluation indexes, interaction performance evaluation indexes, actual scene simulation indexes, system robustness evaluation indexes and safety evaluation indexes.

Images