Edge scene searching method and system based on particle swarm, storage medium and equipment

Abstract

The invention provides an edge scene searching method based on a particle swarm, and the method is characterized in that the method comprises the following steps: the particle swarm is randomly initialized in a global parameter space, each particle represents a group of automatic driving scene parameter configurations, and the parameter configurations at least comprise vehicle speed, vehicle spacing, environmental conditions and road type parameters; inputting scene configuration represented by each particle in the particle swarm into an analogue simulation environment, and carrying out simulation test to obtain a time-to-collision (TTC) value of each particle in each scene; and dynamically adjusting the search direction and the search step size of at least part of particles in the particle swarm according to the collision time value, and carrying out search optimization iteration to obtain an edge scene causing the high failure risk of the automatic driving system. The invention further discloses a corresponding system, a storage medium and computer equipment. By implementing the method and the device, the effectiveness and the efficiency of edge scene search in the automatic driving system test can be improved, and the cost can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving scenario testing technology, and in particular to a particle swarm edge scene search method, system, storage medium, and device. Background Art

[0002] With the rapid development of autonomous driving technology, ensuring its safety and reliability in complex and changing traffic environments has become a major challenge for research and industry.

[0003] Existing technologies primarily rely on real-world data or test scenarios generated through simulated environments to test and validate autonomous vehicles. While these methods can demonstrate the performance and safety of autonomous driving systems to a certain extent, they often require significant time and computing resources and struggle to cover all potentially hazardous scenarios, especially those involving extreme or uncommon edge cases.

[0004] Among the existing technologies, there are testing methods such as Scenario Replication Technology, Simulation Testing, and Random and Grid Search, which are attempted to be used for searching edge scenarios. However, these methods require a lot of computing resources and have low accuracy. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to propose an edge scene search method, system, storage medium and device based on particle swarm, which can improve the effectiveness and efficiency of edge scene search in autonomous driving system testing and reduce costs.

[0006] To solve the technical problem, as one aspect of the present invention, a particle swarm-based edge scene search method is proposed, which includes the following steps:

[0007] A particle swarm is randomly initialized in the global parameter space. Each particle represents a set of autonomous driving scenario parameter configurations, including vehicle speed, vehicle spacing, environmental conditions, and road type parameters.

[0008] Input the scene configuration represented by each particle in the particle swarm into the simulation environment, perform simulation tests, and obtain the Time To Collision (TTC) value of each particle in each scene;

[0009] The search direction and search step size of at least some particles in the particle swarm are dynamically adjusted according to the time to collision (TTC) value, and the search optimization iteration is performed to obtain edge scenarios that lead to a high failure risk of the autonomous driving system.

[0010] Before initializing the particle swarm, the steps of constructing a simulation environment according to the autonomous driving scenario are also included, including:

[0011] Collect traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors;

[0012] Clean and feature extract the collected data to obtain processed data;

[0013] In the simulation software, the scene is reconstructed based on the processed data to generate dynamic traffic flow and interactive environment, forming a simulation environment.

[0014] The scene configuration represented by each particle in the particle swarm is input into the simulation environment for simulation testing to obtain the collision time (TTC) value of each particle in each scene, including:

[0015] The time-to-collision (TTC) value of each particle in each scene can be calculated by dynamically tracking the relative positions and velocities of the vehicles according to the following formula:

[0016]

[0017] Among them, p background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object vehicle, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, L cm for l background and l Ego of and, is the time to collision (TTC) between the vehicle and the following vehicle at time t0.

[0018] The step of dynamically adjusting the search direction and search step size of at least some particles in the particle swarm according to the time to collision (TTC) value includes:

[0019] Comparing the time-to-collision (TTC) value of each particle with a predetermined evaluation threshold, and determining particles with a time-to-collision (TTC) value less than the evaluation threshold as high-risk particles;

[0020] The following formula is used to adjust the speed of each high-risk particle in the particle swarm:

[0021]

[0022] in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration. It is worth noting that when the first iteration is performed, is a randomly assigned vector, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, which is a hyperparameter, c1 and c2 are the individual learning factor and group learning factor of the particles respectively. The former is the influence of the local optimal solution on the search direction, and the latter is the influence of the global optimal solution on the search direction. r1 and r2 are the corresponding correction weights;

[0023] The following formula is used to adjust the position of each high-risk particle in the particle swarm:

[0024]

[0025] in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

[0026] Which further includes:

[0027] In the search optimization iterative process, after each iteration is completed, it is determined whether the iteration termination condition is met, and the iteration termination condition is that a preset convergence condition is met or a preset number of iterations has been reached;

[0028] If not, continue to the next round of iterative optimization; if satisfied, output the edge scenario determined according to the preset criteria.

[0029] As another aspect of the present invention, there is also provided a particle swarm-based edge scene search system, which includes:

[0030] The particle swarm initialization module is used to randomly initialize the particle swarm in the global parameter space. Each particle represents a set of autonomous driving scenario parameter configurations, including vehicle speed, vehicle spacing, environmental conditions, and road type parameters.

[0031] The simulation module is used to input the scene configuration represented by each particle in the particle swarm into the simulation environment, perform simulation tests, and obtain the collision time (TTC) value of each particle in each scene;

[0032] The particle update module is used to dynamically adjust the search direction and search step size of at least some particles in the particle swarm according to the collision time (TTC) value, and perform search optimization iterations to obtain edge scenarios that lead to a high risk of failure of the autonomous driving system.

[0033] Which further includes:

[0034] A simulation environment construction module is used to construct a simulation environment according to the autonomous driving scenario before initializing the particle swarm; the simulation environment construction module includes:

[0035] A data acquisition unit, used to collect traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors;

[0036] A pre-processing unit is used to clean and extract features from the collected data to obtain processed data;

[0037] The scene reconstruction unit is used to reconstruct the scene according to the processed data in the simulation software, generate dynamic traffic flow and interactive environment, and form a simulation environment.

[0038] The simulation module is further used to calculate the time to collision (TTC) value of each particle in each scene by dynamically tracking the relative positions and velocities of the vehicles according to the following formula:

[0039]

[0040] Among them, p background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object vehicle, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, L cm for l background and l Ego of and, is the time to collision (TTC) between the vehicle and the following vehicle at time t0.

[0041] Wherein, the particle update module further includes:

[0042] a risk assessment unit, configured to compare a time-to-collision (TTC) value of each particle with a predetermined assessment threshold, and determine particles having a time-to-collision (TTC) value less than the assessment threshold as high-risk particles;

[0043] The particle velocity adjustment unit is used to adjust the velocity of each high-risk particle in the particle swarm using the following formula:

[0044]

[0045] in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration. It is worth noting that when the first iteration is performed, is a randomly assigned vector, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, which is a hyperparameter, c1 and c2 are the individual learning factor and group learning factor of the particles respectively. The former is the influence of the local optimal solution on the search direction, and the latter is the influence of the global optimal solution on the search direction. r1 and r2 are the corresponding correction weights;

[0046] The particle position adjustment unit is used to adjust the position of each high-risk particle in the particle swarm using the following formula:

[0047]

[0048] in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

[0049] Which further includes:

[0050] The iterative optimization control unit is used to determine whether the iteration termination condition is met after each iteration in the search optimization iterative process. The iteration termination condition is that the preset convergence condition is met or the preset number of iterations has been reached; if not, the next round of iterative optimization is continued; if satisfied, the edge scenario determined according to the preset standard is output.

[0051] As another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0052] As another aspect of the present invention, there is also provided a computer device comprising:

[0053] one or more processors;

[0054] a memory for storing one or more computer programs;

[0055] When the one or more computer programs are executed by the one or more processors, the one or more processors are caused to implement the aforementioned method.

[0056] The implementation of the present invention has the following beneficial effects:

[0057] The present invention provides a particle swarm-based edge scene search method, system, storage medium and device. The edge scenes in the field of autonomous driving are searched by adopting particle swarm optimization (PSO) technology. Specifically, a particle swarm is randomly initialized in the global parameter space, and each particle represents a set of autonomous driving scene parameter configurations; the scene configuration represented by each particle in the particle swarm is input into a simulation environment, and a simulation test is performed to obtain the collision time value of each particle in each scene; the speed and position of each particle in the particle swarm are dynamically adjusted according to the collision time value, thereby increasing the search tendency for these high-risk particle positions, and reducing the step size at the same time, so as to explore the edge scene area more carefully and quickly to ensure that key dangerous scenes are not missed. Through this search optimization iteration, the efficiency and accuracy of the edge scene search are improved, and the cost can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, without inventive work, other drawings derived from these drawings still fall within the scope of the present invention.

[0059] Figure 1 A schematic diagram of the main process of an embodiment of a particle swarm-based edge scene search method provided by the present invention;

[0060] Figure 2 A schematic diagram of the interaction principle of the simulation system involved in the present invention;

[0061] Figure 3 A schematic diagram of the generation principle of the collision time involved in the present invention;

[0062] Figure 4 A schematic diagram of the principle of calculating collision time in scene simulation according to the present invention;

[0063] Figure 5 A schematic diagram of simulation result data involved in an example of the method of the present invention;

[0064] Figure 6 A schematic structural diagram of an embodiment of a particle swarm-based edge scene search system provided by the present invention;

[0065] Figure 7 for Figure 6 Schematic diagram of the structure of the simulation module environment building module;

[0066] Figure 8 for Figure 6 Schematic diagram of the particle update module. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in further detail below with reference to the accompanying drawings.

[0068] like Figure 1 FIG. 1 shows a schematic diagram of the main process of an embodiment of a particle swarm-based edge scene search method provided by the present invention; and FIG. Figures 2 to 5 As shown, in this embodiment, the edge scene search method based on particle swarm includes at least the following steps:

[0069] Step S10: randomly initialize a particle swarm in the global parameter space, where each particle represents a set of autonomous driving scenario parameter configurations, including vehicle speed, vehicle spacing, environmental conditions, and road type parameters;

[0070] Step S11, inputting the scene configuration represented by each particle in the particle swarm into a simulation environment, performing simulation testing, and obtaining the collision time (TTC) value of each particle in each scene;

[0071] In step S12, the search direction and search step size of at least some particles in the particle swarm are dynamically adjusted according to the collision time (TTC) value, the corresponding collision time (TTC) value is recalculated, and the search optimization iteration is performed to obtain edge scenarios that cause a high risk of failure of the autonomous driving system.

[0072] The following is a detailed description of the steps involved in the method of the present invention with reference to specific examples.

[0073] It is understood that, in the method of the present invention, before initializing the particle swarm in step S10, the method further includes:

[0074] Step S00 is a step of constructing a simulation environment based on the autonomous driving scenario. It can be understood that in the testing and verification of the autonomous driving system, accurately quantifying the driving environment in the real world and converting it into a controllable simulation test scenario is a relatively critical step to ensure that the test results are consistent with the actual situation.

[0075] The step S00 further includes:

[0076] Step S01, data collection and preliminary quantification: collecting traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors;

[0077] Specifically, it is necessary to first collect various traffic environment data through advanced sensing technologies (such as LIDAR, cameras, GPS, and IMU). This data includes but is not limited to road characteristics (such as road type, markings, and width), traffic signals (traffic light status, traffic signs), vehicle dynamics (speed, acceleration, direction), and environmental factors (weather conditions, time of day, and light intensity).

[0078] Road and traffic signal data: Identify and classify various road conditions, such as the presence of lane separations, the number of lanes, and the type and location of traffic signs. This data is automatically extracted from videos or images using image recognition technology and converted into digital information.

[0079] Vehicle dynamics and environmental factors: Vehicle-mounted sensors record the vehicle's motion and surrounding environment in real time. For example, the vehicle's speed sensor acquires speed data, and a weather station or onboard weather equipment acquires weather data.

[0080] Step S02: cleaning and feature extraction of the collected data to obtain processed data;

[0081] Specifically, the quantified data needs to be processed and analyzed to construct an accurate scenario model that can be used for simulation. This includes:

[0082] The collected raw data is cleaned to remove outliers and irrelevant data to ensure the quality and accuracy of the data.

[0083] Machine learning algorithms process data to extract key features, such as the relative positions and speeds between vehicles, which are crucial for the realism of the scene.

[0084] Step S03: reconstructing the scene based on the processed data in the simulation software, generating dynamic traffic flow and interactive environment, and forming a simulation environment.

[0085] This step involves scene reconstruction: using the processed data to reconstruct the scene within the simulation software. For example, using Geographic Information System (GIS) data and road mapping data, the road network can be accurately laid out within the simulation environment, ensuring that the geometric characteristics of each road match those in the real world.

[0086] Simultaneously, the simulation scenario is constructed. Specifically, after the data processing is completed, the next step is to construct the scenario on the simulation platform. For example, in one example, the X8V simulation platform can be used to convert the processed data into a dynamic traffic flow and interactive environment. The specific interactive flow chart is as follows: Figure 2 shown. Figure 2 The interactive principle of a general simulation system is shown, in which a simulation request and corresponding data are sent to the simulation system through a third-party system, and the simulation system's data processing system stores the result data generated by the simulation process in the middleware after processing. The third-party system can query the corresponding simulation result data from the middleware.

[0087] Conduct dynamic traffic simulations. Specifically, the vehicle's behavior patterns, including starting, accelerating, braking, and steering, are set based on the quantified data. All of these behaviors need to be adjusted based on the quantified actual driving data to reflect the complexity of the real driving environment.

[0088] Environmental effects are achieved. Specifically, environmental factors such as weather changes and lighting changes need to be included in the simulation. These factors may affect the sensor performance and decision-making of the autonomous driving system.

[0089] Examples:

[0090] Consider a simulation test of an urban traffic scenario, where the goal is to test the performance of an autonomous vehicle at a complex intersection. Quantitative data includes the intersection's detailed layout, traffic light cycles, and surrounding vehicle traffic and behavior patterns. Within the simulation software, this data is used to construct a highly realistic intersection model. Vehicle behavior is simulated based on actual observed data to test the effectiveness and reliability of the autonomous driving algorithm in handling complex traffic conditions.

[0091] Through this detailed process of quantification and simulation scenario construction, we ensure that autonomous driving system testing not only reflects real-world driving conditions but also enables in-depth analysis and optimization for specific driving challenges. This approach strongly supports the development of autonomous driving technology, helping R&D teams anticipate and address potential safety issues.

[0092] In a specific example, in step S10, a particle swarm is randomly initialized in the global parameter space, and each particle represents a set of autonomous driving scenario parameter configurations;

[0093] In simulation testing of autonomous driving systems, the particle swarm optimization (PSO) algorithm can be used to identify scenario configurations that may cause system failure. The first step in the PSO algorithm is to randomly initialize a swarm of particles within a defined parameter space. Each particle represents a set of possible configurations for the autonomous driving scenario, including but not limited to vehicle speed, vehicle spacing, environmental conditions, and road type.

[0094] Parameter setting: First, identify the key parameters that affect the test results and set reasonable ranges for these parameters. These ranges are based on previous experimental data, expert experience, or predefined safety standards.

[0095] Implementing particle representation: Specifically, in a high-dimensional parameter space, the position of each particle represents a specific set of scenario configurations. The initial positions of the particles are generated by random sampling from a set range of parameters, ensuring a wide coverage of various possible scenarios.

[0096] When the PSO algorithm is initialized, all particles are assigned a random position.

[0097]

[0098] In a specific example, in step S11, the scene configuration represented by each particle in the particle swarm is input into a simulation environment, and a simulation test is performed to obtain the collision time (TTC) value of each particle in each scene, specifically including:

[0099] After initialization, the scenario configuration represented by each particle is fed into the simulator. The simulator dynamically simulates these configurations to evaluate the autonomous driving system's performance under specific conditions. Each scenario simulation involves generating interactions between the vehicle and the environment and evaluating the autonomous driving system's responses, such as collision avoidance and speed adjustments.

[0100] Among them, time to collision (TTC) is an important indicator to measure the danger of a scene. It refers to the time when two vehicles or a vehicle and an obstacle are expected to collide under the current path and speed. Figure 3 A schematic diagram of TTC is shown.

[0101] During the simulation, the system calculates the TTC for each scenario by dynamically tracking the relative positions and speeds of vehicles. Figure 4 The TTC value can be calculated based on the positional relationship between the two vehicles at the time of collision shown in .

[0102] Specifically, the time-to-collision (TTC) value of each particle in each scene can be calculated by dynamically tracking the relative positions and velocities of the vehicles according to the following formula:

[0103]

[0104] Combine Figure 3 and Figure 4 The meaning of each parameter is as follows: background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object vehicle, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, Lcm for l background and l Ego of and, is the time to collision (TTC) between the vehicle and the following vehicle at time t0.

[0105] After the simulation is complete, the TTC results for each scenario are fed back to the PSO algorithm. This data is used to assess the risk level of each particle configuration, with configurations with lower TTC values ​​being considered high-risk scenarios.

[0106] In an embodiment of the present invention, the PSO algorithm needs to update the position and velocity of particles based on the TTC results returned by the simulation to optimize the search process and find scenario configurations that may lead to higher risks.

[0107] The update rules involved in the present invention are: the speed and position of the particle are updated according to the individual and group historical optimal positions, using a specific update formula, considering the individual optimal solution (pbesx) and the group optimal solution (g best ) to guide particles to move to potentially higher risk scenarios, where the position of the i-th particle when the number of iterations is t is expressed as Indicates that this is a multidimensional vector whose dimension is equal to the dimension of the scene package after quantization, that is, the number of variables. When the PSO algorithm is initialized, all particles are given a random position.

[0108]

[0109] Examples:

[0110] Assume that in a test, the parameter space includes vehicle speed (30-100 km / h), vehicle distance (5-50 meters) and ambient light (bright, dark). Initialization particles may randomly generate a set of conditions such as speed 85 km / h, vehicle distance 10 meters, and ambient light condition dark. Figure 3 、 4 The simulation of driving scenarios under these conditions is shown. Figure 5 , shows a calculated value curve of a simulation result. If the calculated TTC is less than 2 seconds, it indicates that this is a high-risk scenario. The PSO algorithm will focus on such parameter configurations to continue exploring and optimizing the test.

[0111] In simulation testing of autonomous driving systems, adjusting the search speed and updating particle positions based on the time-to-collision (TTC) results transmitted back from the simulation is a core step in the particle swarm optimization (PSO) algorithm. This process involves dynamically adjusting the particle search strategy to more effectively explore high-risk scenarios that could lead to system failure.

[0112] In a specific example, in step S12, dynamically adjusting the search direction and search step size of at least part of the particles in the particle swarm according to the collision time (TTC) value includes the following steps:

[0113] In step S120, the collision time (TTC) value of each particle is compared with a predetermined evaluation threshold, and particles with a collision time (TTC) value less than the evaluation threshold are determined to be high-risk particles; for example, if the configuration of a particle shows a TTC value of 3 seconds (less than the evaluation threshold of 4 seconds), it indicates that the area is a high-risk area.

[0114] Subsequently, the PSO algorithm needs to adjust the speed and position of each particle based on the TTC results. This adjustment process aims to optimize the search efficiency of the particles, making them more inclined to explore scenarios that may reveal potential defects in the system; the PSO algorithm evaluates the effectiveness of the scenario configuration represented by each particle in the particle swarm during each iteration. By measuring key performance indicators such as collision time (TTC), PSO automatically adjusts the search direction and step size based on the simulation results. For example, when the configuration of a particle shows an extremely low TTC value, indicating a high-risk situation, the algorithm will increase the search tendency towards the position of the particle, while reducing the step size to explore the area more carefully to ensure that key dangerous scenarios are not missed; reducing the step size can be achieved by reducing the speed of the particle.

[0115] Step S121: Use the following formula to adjust the speed of each high-risk particle in the particle group:

[0116]

[0117] in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration. It is worth noting that when the first iteration is performed, is a randomly assigned vector, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, which is a hyperparameter, c1 and c2 are the individual learning factor and group learning factor of the particles respectively. The former is the influence of the local optimal solution on the search direction, and the latter is the influence of the global optimal solution on the search direction. r1 and r2 are the corresponding correction weights;

[0118] Step S122: Adjust the position of each high-risk particle in the particle swarm using the following formula:

[0119]

[0120] in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

[0121] It can be seen from the above steps S121 and S122 that, in an embodiment of the present invention, the particle velocity update is adjusted according to the current velocity, the individual historical optimal position (pbest) and the group historical optimal position (gbest). The update formula usually includes an inertia weight (w), an individual learning factor (c1) and a social learning factor (c2), and is combined with random factors (r1, r2) to adjust the velocity vector. When the configuration of a particle shows an extremely low TTC value, indicating that the scene is a high-risk situation, the algorithm will increase the search tendency toward the position of the particle, while reducing the step size to explore the area more carefully. This means that when the velocity is updated, the size and direction of the velocity vector will be adjusted to move the particle toward the high-risk area, and the step size of the movement will become smaller to explore the area more accurately.

[0122] At the same time, the particle's position is updated based on the new velocity vector. This new position is based on the particle's exploration of parameter space, aiming to approach or enter unexplored high-risk areas. When the TTC value determines that a particle's scene is high-risk, during the position update phase, the particle moves in the direction of higher risk, thereby further exploring the high-risk scene area.

[0123] More specifically, the method of the present invention further comprises:

[0124] Step S13, in the search optimization iterative process, after each iteration is completed, it is determined whether the iteration termination condition is met, wherein the iteration termination condition is that a preset convergence condition is met (the particle position change is less than a predetermined threshold), or a preset number of iterations has been reached;

[0125] If not, continue to the next round of iterative optimization; if satisfied, output the scene configuration corresponding to the current group's historical optimal position as the edge scene.

[0126] It is understandable that in the present invention, the updating of the particle position and velocity is an iterative process, and each iteration is adjusted based on the latest simulation results. best and g best The iteration process may be updated to reflect new individual and group optimal solutions. These updates guide the movement direction and speed of particles in subsequent iterations. The particle convergence is monitored during the iteration process. If the change in particle position is less than a certain threshold or the preset number of iterations has been reached, the iteration process may be terminated.

[0127] As can be understood, the present invention aims to enhance the testing efficiency and accuracy of autonomous driving systems in complex traffic environments through the particle swarm optimization (PSO) algorithm. The solution provided by this invention primarily involves the quantitative construction of scenarios, random initialization and simulation testing of the PSO algorithm, and dynamic adjustment based on the simulation test results. Through this detailed dynamic adjustment process, the PSO algorithm can effectively target critical high-risk scenarios, allowing for more accurate evaluation and improvement of the performance and safety of autonomous driving systems during the R&D phase.

[0128] like Figure 6 FIG. 1 shows a schematic diagram of the structure of an embodiment of a particle swarm-based edge scene search system provided by the present invention. Figures 7 and 8 As shown, in this embodiment, the particle swarm-based edge scene search system 1 at least includes:

[0129] A particle swarm initialization module 10 is used to randomly initialize a particle swarm in a global parameter space, where each particle represents a set of autonomous driving scenario parameter configurations, including vehicle speed, vehicle spacing, environmental conditions, and road type parameters;

[0130] The simulation module 11 is used to input the scene configuration represented by each particle in the particle swarm into the simulation environment, perform simulation testing, and obtain the collision time (TTC) value of each particle in each scene;

[0131] The particle update module 12 is used to dynamically adjust the search direction and search step size of at least some particles in the particle swarm according to the collision time (TTC) value, recalculate the corresponding collision time (TTC) value, and perform search optimization iterations to obtain edge scenarios that cause a high risk of failure of the autonomous driving system.

[0132] More specifically, combined with Figure 7 As shown, the system 1 provided by the present invention further includes:

[0133] The simulation environment construction module 13 is used to construct a simulation environment according to the autonomous driving scenario before initializing the particle swarm. The simulation environment construction module 13 includes:

[0134] The data collection unit 130 is used to collect traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors;

[0135] The pre-processing unit 131 is used to clean and extract features from the collected data to obtain processed data;

[0136] The scene reconstruction unit 132 is used to reconstruct the scene according to the processed data in the simulation software, generate dynamic traffic flow and interactive environment, and form a simulation environment.

[0137] More specifically, the simulation module 11 is further configured to calculate the time to collision (TTC) value of each particle in each scene by dynamically tracking the relative positions and velocities between vehicles according to the following formula:

[0138]

[0139] Among them, P background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object vehicle, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, L cm for l background and l Ego of and, is the time to collision (TTC) between the vehicle and the following vehicle at time t0.

[0140] In a specific example, such as Figure 8 As shown, the particle updating module 12 further includes:

[0141] a risk assessment unit 120 for comparing a time-to-collision (TTC) value of each particle with a predetermined assessment threshold, and determining particles having a time-to-collision (TTC) value less than the assessment threshold as high-risk particles;

[0142] The particle speed adjustment unit 121 is used to adjust the speed of each high-risk particle in the particle group using the following formula:

[0143]

[0144] in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration. It is worth noting that when the first iteration is performed, is a randomly assigned vector, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, which is a hyperparameter; c1 and c2 are the individual learning factor and group learning factor of the particles, respectively. The former is the influence of the local optimal solution on the search direction, and the latter is the influence of the global optimal solution on the search direction; r1 and r2 are the corresponding correction weights;

[0145] The particle position adjustment unit 122 is used to adjust the position of each high-risk particle in the particle group using the following formula:

[0146]

[0147] in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

[0148] More specifically, the system 1 provided by the present invention further comprises:

[0149] The iterative optimization control unit 14 is used to determine whether the iteration termination condition is met after each iteration in the search optimization iterative process. The iteration termination condition is that the preset convergence condition is met or the preset number of iterations has been reached; if not, the next round of iterative optimization is continued; if satisfied, the scene configuration corresponding to the current group's historical optimal position is output as the edge scene.

[0150] For more details, please refer to and combine the above Figures 1 to 6 The description is not repeated here.

[0151] As another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, Figures 1 to 5 For more details, please refer to and combine the above Figures 1 to 5 The description is not repeated here.

[0152] As another aspect of the present invention, there is also provided a computer device comprising:

[0153] one or more processors;

[0154] a memory for storing one or more computer programs;

[0155] When the one or more computer programs are executed by the one or more processors, the one or more processors are caused to implement the following Figures 1 to 5 For more details, please refer to and combine the above Figures 1 to 5 The description is not repeated here.

[0156] The implementation of the present invention has the following beneficial effects:

[0157] The present invention provides a particle swarm-based edge scene search method, system, storage medium, and device. Edge scenes in the field of autonomous driving are searched by using particle swarm optimization technology. Specifically, a particle swarm is randomly initialized in a global parameter space, with each particle representing a set of autonomous driving scenario parameter configurations. The scenario configurations represented by each particle in the particle swarm are input into a simulation environment, and simulation tests are performed to obtain the collision time value of each particle in each scenario. The speed and position of each particle in the particle swarm are dynamically adjusted according to the collision time value, thereby increasing the search tendency for these high-risk particle positions, while reducing the step size to explore the edge scene area more carefully and quickly, ensuring that key dangerous scenarios are not missed. Through this search optimization iteration, the efficiency and accuracy of edge scene search are improved, and costs can be reduced.

[0158] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0159] The above disclosure is only a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A particle swarm-based edge scene search method, characterized in that: The following steps are involved: Randomly initialize a particle swarm in the global parameter space, where each particle represents a set of autonomous driving scenario parameter configurations, including at least vehicle speed, vehicle spacing, environmental conditions, and road type parameters; The scene configuration represented by each particle in the particle swarm is input into the simulation environment, and simulation tests are performed to obtain the collision time (TTC) value of each particle in each scene; The search direction and search step size of at least some particles in the particle swarm are adjusted according to the collision time value, and the search optimization iteration is performed to obtain edge scenarios that lead to a high failure risk of the autonomous driving system.

2. The method according to claim 1, characterized in that Before initializing the particle swarm, the steps of building a simulation environment based on the autonomous driving scenario are also included, including: Collect traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors; Clean and feature extract the collected data to obtain processed data; In the simulation software, the scene is reconstructed based on the processed data to generate dynamic traffic flow and interactive environment, forming a simulation environment.

3. The method according to claim 1 or 2, wherein: The scene configuration represented by each particle in the particle swarm is input into the simulation environment, and simulation tests are performed to obtain the collision time (TTC) value of each particle in each scene, including: The time-to-collision (TTC) value of each particle in each scene can be calculated by dynamically tracking the relative positions and velocities of the vehicles according to the following formula: Among them, p background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object vehicle, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, L cm l background and l Ego of and, is the collision time between the vehicle and the following vehicle at time t0.

4. The method according to claim 3, wherein The dynamically adjusting the search direction and search step size of at least part of the particles in the particle swarm according to the collision time value includes: Comparing the time-to-collision (TTC) value of each particle with a predetermined evaluation threshold, and determining particles with a time-to-collision (TTC) value less than the evaluation threshold as high-risk particles; The following formula is used to adjust the speed of each high-risk particle in the particle swarm: in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, c1 and c2 are the individual learning factor and group learning factor of the particles respectively, and r1 and r2 are the corresponding correction weights; The following formula is used to adjust the position of each high-risk particle in the particle swarm: in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

5. The method according to claim 4, wherein Further including: In the search optimization iterative process, after each iteration is completed, it is determined whether the iteration termination condition is met, and the iteration termination condition is that a preset convergence condition is met or a preset number of iterations has been reached; If not, continue to the next round of iterative optimization; if satisfied, output the edge scenario determined according to the preset criteria.

6. A particle swarm-based edge scene search system, characterized in that: include: A particle swarm initialization module is used to randomly initialize a particle swarm in the global parameter space, where each particle represents a set of autonomous driving scenario parameter configurations, including at least vehicle speed, vehicle spacing, environmental conditions, and road type parameters; The simulation module is used to input the scene configuration represented by each particle in the particle swarm into the simulation environment, perform simulation tests, and obtain the collision time (TTC) value of each particle in each scene; The particle update module is used to dynamically adjust the search direction and search step size of at least some particles in the particle swarm according to the collision time value, and perform search optimization iterations to obtain edge scenarios that cause a high risk of failure of the autonomous driving system.

7. The system according to claim 6, characterized in that Further including: The simulation environment construction module is used to build a simulation environment according to the autonomous driving scenario before initializing the particle swarm; The simulation environment building module includes: A data acquisition unit, used to collect traffic environment data, including at least: road characteristics, traffic signals, vehicle dynamics and environmental factors; A pre-processing unit is used to clean and extract features from the collected data to obtain processed data; The scene reconstruction unit is used to reconstruct the scene according to the processed data in the simulation software, generate dynamic traffic flow and interactive environment, and form a simulation environment.

8. The system according to claim 6 or 7, characterized in that The simulation module is further used to calculate the collision time value of each particle in each scene by dynamically tracking the relative positions and velocities between vehicles according to the following formula: Among them, p background is the current position of the target vehicle, p Ego is the current position of the vehicle, v background is the current speed of the object car, v Ego is the current speed of the vehicle, l background is the length from the rear boundary of the target vehicle to the center of the vehicle, l Ego is the length from the front edge of the vehicle to the center of the vehicle, L cm for l background and l Ego of and, is the collision time between the vehicle and the following vehicle at time t0.

9. The system according to claim 8, wherein The particle update module further includes: a risk assessment unit, configured to compare the collision time value of each particle with a predetermined assessment threshold, and determine particles whose collision time value is less than the assessment threshold as high-risk particles; The particle velocity adjustment unit is used to adjust the velocity of each high-risk particle in the particle swarm using the following formula: in, is the next velocity vector of the i-th particle calculated based on the result of the t-th iteration, that is, the velocity vector of the t+1-th iteration, is the velocity vector of the i-th particle at the t-th iteration, is the local optimal solution of the i-th particle at the t-th iteration, is the global optimal solution of all particles at the tth iteration, w is the inertia weight, c1 and c2 are the individual learning factor and group learning factor of the particles respectively, and r1 and r2 are the corresponding correction weights; The particle position adjustment unit is used to adjust the position of each high-risk particle in the particle swarm using the following formula: in, is the next position vector of the i-th particle calculated based on the t-th iteration result, is the position vector of the i-th particle at the t-th iteration.

10. The system according to claim 9, wherein: Further including: An iterative optimization control unit is used to determine whether an iteration termination condition is satisfied after each iteration in the search optimization iterative process, wherein the iteration termination condition is that a preset convergence condition is satisfied or a preset number of iterations has been reached; If not, continue to the next round of iterative optimization; if satisfied, output the edge scenario determined according to the preset criteria.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

12. A computer device, characterized in that: include: one or more processors; a memory for storing one or more computer programs; When the one or more computer programs are executed by the one or more processors, the one or more processors are caused to implement the method according to any one of claims 1 to 5.

Images