Automatic driving multi-vehicle interactive test scene generation method and device and storage medium

Abstract

The invention discloses an automatic driving multi-vehicle interactive test scene generation method, and the method comprises the steps: carrying out the decomposition of a scene according to the design requirements of an automatic driving test scene, and extracting necessary static environment information and the dynamic parameter distribution of traffic participants; static environment information and dynamic parameters of traffic participants are sampled, a scene is extracted into a plurality of scene sections, and the scene sections are divided into test scene samples and verification scene samples according to a preset proportion; inputting the test scene sample and the verification scene sample into a multi-agent model for training to obtain a trained multi-agent model; and deploying the trained multi-agent model to a test scene sample, carrying out an interaction test with the tested automatic driving system, and obtaining an automatic driving multi-vehicle interactive test scene, so that the target vehicle is controlled through the multi-agent model, and interaction with the automatic driving system is carried out. Therefore, the performance of the automatic driving system in a high-risk environment can be reflected more truly, and the safety and robustness evaluation of the automatic driving system are ensured to be more comprehensive.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving, and specifically to a method, device and storage medium for generating a multi-vehicle interactive test scenario for autonomous driving. Background Art

[0002] In the existing generation of autonomous driving test scenarios, although the generated model can capture the trajectory of the vehicle, there is a large gap between the preset trajectory and the actual environment. In reality, other vehicles will make adjustments based on the behavior of the autonomous driving vehicle, while in traditional simulations, the behavior of other vehicles is preset and cannot respond to the decisions of the autonomous driving system in real time, resulting in test results that are inconsistent with reality and affecting the accuracy of the test. In complex traffic scenarios, autonomous driving systems must consider the dynamic interaction with other vehicles, especially in high-risk scenarios, where this interaction is particularly important.

[0003] However, the existing multi-vehicle interaction research is still in its early stages and lacks in-depth analysis of complex scenarios, especially in accident scenarios, where there is little research and the test scenarios cannot fully reflect the actual performance of the autonomous driving system. Therefore, a dynamic test scenario generation method that can simulate multi-vehicle interaction is urgently needed. Summary of the invention

[0004] In response to the above technical problems, the present application provides a method, device and storage medium for generating an interactive test scenario for multiple vehicles in an autonomous driving system, which more realistically reflects the performance of the autonomous driving system in a high-risk environment and ensures a more comprehensive evaluation of its safety and robustness.

[0005] The present application provides a method for generating a multi-vehicle interactive test scenario for autonomous driving, the method comprising:

[0006] Step S1: Decompose the scene according to the design requirements of the autonomous driving test scene, and extract the necessary static environment information and dynamic parameter distribution of traffic participants;

[0007] Step S2, sampling static environmental information and dynamic parameters of traffic participants, extracting the scene into a number of scene sections, and dividing them into test scene samples and verification scene samples according to a preset ratio;

[0008] Step S3, inputting the test scene sample and the verification scene sample into the multi-agent model training to obtain a trained multi-agent model;

[0009] Step S4: deploy the trained multi-agent model to the test scene sample, perform interactive testing with the autonomous driving system under test, and obtain an autonomous driving multi-vehicle interactive test scene.

[0010] Preferably, the static environmental information includes: vehicle type of traffic participants, road type, road width, number of lanes, lighting conditions, weather conditions, traffic control method, and number of background vehicles.

[0011] Preferably, the dynamic parameters of the traffic participant are the state of the traffic participant at each time step, including: the longitudinal coordinate, the transverse coordinate, the speed, acceleration, direction, turn, and relative position difference at each coordinate within each step.

[0012] Preferably, the sampling of the static environment information and the dynamic parameters of the traffic participants, extracting the scene into a plurality of scene sections, and dividing them into test scene samples and verification scene samples according to a preset ratio, includes:

[0013] Static environmental information and dynamic parameters of traffic participants are sampled at preset time intervals to generate several independent scene sections. The data of each scene section must include static information, dynamic information, and the relative position and behavioral characteristics of each participating vehicle, and are divided into test scene samples and verification scene samples according to a preset ratio.

[0014] Preferably, the step of inputting the test scene sample and the verification scene sample into a multi-agent model training to obtain a trained multi-agent model comprises:

[0015] Input the test scene samples into the multi-agent model training to generate a dynamic driving scene set;

[0016] Based on the dynamic driving scenario set and the verification scenario samples, a multi-agent model is input, a reward function is set, and when the training loss value reaches convergence, a trained multi-agent model is obtained.

[0017] Preferably, the trained multi-agent model is deployed to a test scene sample, and interactive testing is performed with the tested autonomous driving system to obtain an autonomous driving multi-vehicle interactive test scene, including:

[0018] Deploy the trained multi-agent model to the test scenario sample and set up the tested autonomous driving system to ensure that its parameters and states match the test scenario sample;

[0019] In the test scene sample, the target vehicle controlled by the trained multi-agent model and the autonomous driving system under test together constitute an interactive environment;

[0020] An interactive test is performed based on the test scenario samples to generate an autonomous driving multi-vehicle interactive test scenario.

[0021] Preferably, after performing interactive testing based on the test scenario sample to generate an autonomous driving multi-vehicle interactive test scenario, the method further includes:

[0022] Obtain the response data of the autonomous driving system and the behavior data of the target vehicle, including the perception accuracy, decision time, response delay, and distance and speed changes of the autonomous driving system from the target vehicle.

[0023] The embodiment of the present invention further provides a device for generating a multi-vehicle interactive test scenario for autonomous driving, comprising:

[0024] The acquisition module is used to decompose the scene according to the design requirements of the autonomous driving test scene and extract the necessary static environment information and dynamic parameter distribution of traffic participants;

[0025] A sampling module is used to sample static environmental information and dynamic parameters of traffic participants, extract the scene into several scene sections, and divide them into test scene samples and verification scene samples according to a preset ratio;

[0026] A training module, used for inputting the test scenario samples and the verification scenario samples into a multi-agent model training to obtain a trained multi-agent model;

[0027] The output module is used to deploy the trained multi-agent model to the test scene sample, and perform interactive testing with the tested autonomous driving system to obtain an autonomous driving multi-vehicle interactive test scene. The present application also provides a computer device, including: a memory and a processor, wherein the memory stores computer program instructions for execution on the processor, and when the processor executes the computer program instructions, the method for generating an autonomous driving multi-vehicle interactive test scene as described above is implemented.

[0028] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for generating a multi-vehicle interactive test scenario for autonomous driving as described above are implemented.

[0029] As described above, according to the design requirements of the autonomous driving test scenario, the scenario is decomposed to extract the necessary static environmental information and the dynamic parameter distribution of traffic participants; the static environmental information and the dynamic parameters of traffic participants are sampled, the scenario is extracted into several scenario sections, and divided into test scenario samples and verification scenario samples according to a preset ratio; the test scenario samples and the verification scenario samples are input into the multi-agent model training to obtain a trained multi-agent model; the trained multi-agent model is deployed to the test scenario samples, and interactive testing is performed with the autonomous driving system under test to obtain an autonomous driving multi-vehicle interactive test scenario. In this way, the target vehicle is controlled by the multi-agent model and interacts with the autonomous driving system to more realistically reflect the performance of the autonomous driving system in a high-risk environment, ensuring a more comprehensive evaluation of its safety and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.

[0031] Figure 1 A schematic diagram of a flow chart of a method for generating a multi-vehicle interactive test scenario for autonomous driving provided by an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of the structure of an apparatus for generating a multi-vehicle interactive test scenario for autonomous driving provided in one embodiment of the present invention.

[0033] The realization of the purpose, functional features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. The above-mentioned drawings have shown clear embodiments of this application, which will be described in more detail later. These drawings and textual descriptions are not intended to limit the scope of the concept of this application in any way, but to illustrate the concept of this application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0035] Before further describing the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following explanations.

[0036] See also Figure 1 This embodiment provides a method for generating a multi-vehicle interactive test scenario for autonomous driving, the method comprising:

[0037] Step S1: Decompose the scene according to the design requirements of the autonomous driving test scene, and extract the necessary static environment information and dynamic parameter distribution of traffic participants;

[0038] Here, by deconstructing the scene and extracting information, we can accurately obtain the key elements required for the test, including static environmental information such as road layout and traffic signals, as well as dynamic parameter distributions such as speed, acceleration, and position of various traffic participants. This refined extraction method makes the test scene construction more flexible and can quickly adjust the scene elements according to specific test requirements, thereby improving the pertinence and efficiency of the test.

[0039] Step S2, sampling static environmental information and dynamic parameters of traffic participants, extracting the scene into a number of scene sections, and dividing them into test scene samples and verification scene samples according to a preset ratio;

[0040] Here, by sampling the scene information, the complex scene can be decomposed into multiple easy-to-process sections, each of which can be used as an independent test unit. This segmented processing method can reduce the complexity of the entire scene, making subsequent generation and calculation more efficient, while ensuring the full expression of scene diversity and improving the flexibility of scene generation.

[0041] The scene is sampled with a sampling interval of 0.2 seconds to generate a scene section. Specifically, according to the system requirements, 0.2 seconds is selected as the sampling time step. This means that all static and dynamic parameters in the autonomous driving scene are collected once every 0.2 seconds to ensure that the continuous behavioral changes of traffic participants are captured. In each 0.2-second time step, all static environmental information (such as road type, road width, number of lanes, etc.) and dynamic parameters (such as the speed, acceleration, direction, etc. of each vehicle) in the scene are extracted. With this information, several independent scene sections are generated, each of which represents the traffic state at a specific time point. Each scene is decomposed into independent segments in multiple time periods to more finely describe the dynamic behavior of traffic participants. For each generated scene section, its related parameter data is sorted and encoded to ensure that the data can be used for subsequent model training. The data of each scene section must include static information, dynamic information, and the relative position and behavioral characteristics of each participating vehicle. After completing the generation of the scene section, according to the training requirements of the model, all generated scene samples are divided into training scene samples and verification scene samples in a ratio of 8:2. The training scene samples are used to train the multi-agent model, and the validation scene samples are used to evaluate the performance of the model. Ensure the randomness and uniformity of sample division to avoid overfitting or data bias.

[0042] Step S3, inputting the test scene sample and the verification scene sample into the multi-agent model training to obtain a trained multi-agent model;

[0043] Here, a multi-agent deep deterministic policy gradient algorithm (MADDPG) is used to control each target vehicle. Each target vehicle is considered as an independent agent, and its agent interacts with the autonomous driving system by adaptively adjusting its behavior. Each agent makes decisions based on environmental information and self-goals to generate complex dynamic driving scenarios. During training, agents share global information, such as the states and actions of other agents, to more accurately evaluate the value of joint state-action pairs. Each agent makes independent decisions based on its local observation information, achieving collaboration and competition in autonomous driving scenarios.

[0044] Step S4: deploy the trained multi-agent model to the test scene sample, perform interactive testing with the autonomous driving system under test, and obtain an autonomous driving multi-vehicle interactive test scene.

[0045] The above-mentioned embodiments of the present invention control the target vehicle through a multi-agent model and interact with the autonomous driving system to more realistically reflect the performance of the autonomous driving system in a high-risk environment and ensure a more comprehensive evaluation of its safety and robustness.

[0046] In one embodiment, the static environment information includes: vehicle type of traffic participants, road type, road width, number of lanes, lighting conditions, weather conditions, traffic control method, and number of background vehicles.

[0047] In one embodiment, the dynamic parameters of the traffic participant are the state of the traffic participant at each time step, including: longitudinal coordinates, transverse coordinates, speed, acceleration, direction, turn, and relative position difference at each coordinate within each step.

[0048] In one embodiment, the static environment information and the dynamic parameters of the traffic participants are sampled, the scene is extracted into a plurality of scene sections, and divided into test scene samples and verification scene samples according to a preset ratio, including:

[0049] Static environmental information and dynamic parameters of traffic participants are sampled at preset time intervals to generate several independent scene sections. The data of each scene section must include static information, dynamic information, and the relative position and behavioral characteristics of each participating vehicle, and are divided into test scene samples and verification scene samples according to a preset ratio.

[0050] In one embodiment, inputting the test scene sample and the verification scene sample into the multi-agent model training to obtain a trained multi-agent model includes:

[0051] Input the test scene samples into the multi-agent model training to generate a dynamic driving scene set;

[0052] Based on the dynamic driving scenario set and the verification scenario samples, a multi-agent model is input, a reward function is set, and when the training loss value reaches convergence, a trained multi-agent model is obtained.

[0053] Each target vehicle is controlled using a multi-agent deep deterministic policy gradient algorithm (MADDPG). Each target vehicle is treated as an independent agent, and its agent interacts with the autonomous driving system by adaptively adjusting its behavior. Each agent makes decisions based on environmental information and its own goals to generate complex dynamic driving scenarios. During training, agents share global information, such as the states and actions of other agents, to more accurately evaluate the value of joint state-action pairs. Each agent makes independent decisions based on its local observation information, achieving collaboration and competition in autonomous driving scenarios.

[0054] Here, the multi-agent model can use the multi-agent deep deterministic policy gradient algorithm (MADDPG). The training process of the multi-vehicle interactive control model based on MADDPG control is as follows:

[0055] Step 0: Initialization;

[0056] With random parameters Initialize each critic network for multiple cars Network with actors

[0057] Initialize the target network

[0058] Initialize the replay buffer

[0059] Step 1: Start each round of training;

[0060] For each training round m, initialize the random process parameters For behavioral exploration;

[0061] Receive the first observed state value (i.e. the initial state of each vehicle);

[0062] Step 2: Behavior selection and execution;

[0063] For each time step t = 1 to T, an action is selected by the actor network:

[0064]

[0065] Execute the selected behavior, determine the interaction between the host vehicle and other vehicles, and calculate the reward and the new state

[0066] Step 3: Store experience;

[0067] Transfer sequence Deposit separately

[0068] Step 4: Sample from experience;

[0069] From the replay buffer A minibatch sample of size N is randomly sampled from

[0070] Step 5: Target update and loss calculation;

[0071] Set the target value:

[0072]

[0073] Calculate the loss function L of the critic network and minimize it:

[0074]

[0075] Step 6: Update the critic and actor networks and actor networks

[0076] The critic network parameter update is done by minimizing the loss function L, and the actor network is updated through the policy gradient:

[0077]

[0078] Step 7: Update the target network;

[0079] Update the target network parameters using the soft update method:

[0080]

[0081] Here, different state spaces are set according to different vehicle types. Specifically, the state parameters of each vehicle are divided into two dimensions, lateral and longitudinal, according to its motion characteristics. The lateral state includes the lateral position and steering angle of the vehicle, while the longitudinal state consists of the longitudinal position and speed. Therefore, the state space of each target vehicle can be represented by a four-dimensional vector as follows:

[0082] S i =[s i1 ,s i2 ,s i3 ,s i4 ];

[0083] Among them, s i1 represents the lateral position of the vehicle, s i2 represents the steering angle, s i3 andi4 denote the longitudinal position and longitudinal speed, respectively. In order to simplify the calculation and facilitate the study of the relative position relationship of the vehicles in the accident, the initial position of the host vehicle is fixed as the origin of the coordinate system, and the initial position of each target vehicle is determined according to its relative position to the host vehicle in the original accident scene. After initializing the autonomous driving state, the autonomous driving takes over the control of the host vehicle and generates three driving tasks according to the corresponding map: left turn, right turn and straight driving.

[0084] The state space of the target vehicle is set according to the vehicle dynamics model. For ease of calculation, the same target vehicle is set to have the same motion parameters. The lateral and longitudinal control of the target vehicle is determined by the pedals and steering wheel, so the acceleration and yaw rate are selected as the action parameters in reinforcement learning. The behavior constraints are set based on the vehicle's dynamic parameters as follows:

[0085] ―6m / s 2 ≤a≤6m / s 2 ;

[0086] ―15° / s≤ω≤15° / s;

[0087] Among them, a is the acceleration and ω is the yaw rate of the vehicle.

[0088] Setting the reward function: The reward function is designed based on the relative distance and generalized time to collision (GTTC) between the autonomous vehicle and each target vehicle. The agent guides its behavior through the reward function, giving priority to generating realistic multi-vehicle interaction scenarios. The reward function for the main interactive target vehicles is as follows:

[0089]

[0090] where d m is the relative distance between the main vehicle and the main interactive target vehicle, GTTC m is the generalized collision time between the two, calculated as:

[0091] D i,j =[(p i ―p j ) T (p i ―p j )] 1 / 2

[0092] D′ i,j =(p i ―p j ) T (v i ―v j ) / D i,j

[0093] GTTC=―Di,j / D′ i,j

[0094] Among them, p i ,p j are the positions of the points where the two vehicles are closest to each other at a given time step; v i ,v j is the velocity vector of the two vehicles at each time step; in the formula, GTTC is only calculated when D′ i,j <0. D′ i,j ≥0 means that the distance between the two vehicles will increase or remain the same, that is, there is no risk of collision. The reward function is set by the distance d m The inverse of GTTC makes the reward value larger when the target car is close to the host car, thus encouraging the target car to reduce the distance with the host car. Similarly, a larger GTTC value means that the potential risk of a collision between the two vehicles is smaller. The reward function encourages the target car to take more aggressive actions by increasing the penalty of the GTTC value. Its essence is to punish larger relative distances and larger GTTC values, with the aim of prompting the target vehicle to get as close to autonomous driving as possible and simulate more challenging driving scenarios.

[0095] In the design of the reward function for the secondary interactive vehicle, in addition to the distance d between the main vehicle and the secondary interactive vehicle s In addition to the GTTC between them, the relative distance d between the secondary vehicle and the primary target vehicle is also specially considered. o , the reward function is set as:

[0096]

[0097] Among them, d s The distance between the main vehicle and the secondary interacting vehicle: When the secondary vehicle approaches the main vehicle, the reward value increases, encouraging the agent to handle potential interactions between the target vehicle and the main vehicle, ensuring that the agent can weigh the dynamic interactions between the main vehicle and the target vehicle. o Represents the distance between the two secondary target vehicles: ensure the interaction between the two target vehicles in the scene. The smaller d o This means that the two target vehicles are close to each other, and the agent needs to handle the interaction between the host vehicle and the two target vehicles at the same time. The complexity of this multi-vehicle interaction scenario significantly increases the difficulty of the scenario. o The penalty of urges the agent to not only pay attention to the interaction between the main vehicle and the priority target vehicle, but also reasonably handle the interaction between the secondary target vehicle and the main target vehicle. s It represents the GTTC between the main vehicle and the secondary target vehicle. By penalizing a larger GTTC value, it encourages the agent to confront the main vehicle and increases the impact of the secondary target vehicle on the scene.

[0098] The reward functions of the remaining interacting vehicles j are set as:

[0099] r j =γr s

[0100] Where γ is defined as the interaction discount coefficient, which weakens the influence of other interacting vehicles and improves the convergence ability of the model. In the process of multi-vehicle interaction, the design of the reward function needs to comprehensively consider the behavior of all target vehicles and secondary target vehicles. Therefore, the overall reward function is jointly determined by the interaction process:

[0101] r = λ1 × r m +λ2×r s +Σ ∈C\{m,s} r j

[0102] j∈C\{m,s} represents all target vehicles except the primary interaction vehicles and all other vehicles of the secondary interaction targets. The specific design of weights is shown in Table 1.

[0103] Table 1 Weight parameter design

[0104]

[0105] In one embodiment, the trained multi-agent model is deployed to a test scene sample, and interactive testing is performed with the tested autonomous driving system to obtain an autonomous driving multi-vehicle interactive test scene, including:

[0106] Deploy the trained multi-agent model to the test scenario sample and set up the tested autonomous driving system to ensure that its parameters and states match the test scenario sample;

[0107] In the test scene sample, the target vehicle controlled by the trained multi-agent model and the autonomous driving system under test together constitute an interactive environment;

[0108] An interactive test is performed based on the test scenario samples to generate an autonomous driving multi-vehicle interactive test scenario.

[0109] Here, the agent model that has been trained through multi-agent reinforcement learning is deployed to the test set. At this point, each agent has the ability to adaptively learn driving strategies in different scenarios and can make dynamic decisions in complex environments. The deployment of the agent includes loading its policy network into the test scenario so that the agent can make decisions in real time based on the new environment and target vehicle information.

[0110] Furthermore, the autonomous driving system under test is set up to ensure that its parameters and states match the scenarios in the test set. The autonomous driving system needs to be able to perceive all traffic participants in the environment (including the target vehicle controlled by the agent) and respond accordingly.

[0111] Here, in each scenario in the test set, the target vehicle controlled by the agent and the tested autonomous driving system together form an interactive environment. As a representative of other traffic participants, the agent will influence the decision-making process of the autonomous driving system through its behavior. The scenarios in the test set cover a variety of potential high-risk traffic situations, such as merging, overtaking, intersections and other complex traffic scenarios.

[0112] Here, the interactive test process is initiated, and the agent begins to control the target vehicle and interact with the autonomous driving system in real time according to the strategy learned during the training process. The agent's behavior will be continuously adjusted according to the response of the autonomous driving system. For example, when the autonomous driving system brakes, the target vehicle may accelerate closer to create a potential collision risk. The whole process is a dynamic interactive process designed to simulate multi-vehicle interaction scenarios in the real world.

[0113] Here, in interactive testing, the intelligent agent continuously challenges the responsiveness of the autonomous driving system through complex driving behaviors, generating high-risk test scenarios. The autonomous driving system must respond to these high-risk behaviors in real time through modules such as perception, planning, and control, and make safe decisions.

[0114] In one embodiment, after performing interactive testing based on the test scenario sample to generate an autonomous driving multi-vehicle interactive test scenario, the method further includes:

[0115] Obtain the response data of the autonomous driving system and the behavior data of the target vehicle, including the perception accuracy, decision time, response delay, and distance and speed changes of the autonomous driving system from the target vehicle.

[0116] Here, during the entire interactive test process, the response data of the autonomous driving system and the behavior data of the target vehicle under the control of the agent are recorded in real time. These data include the perception accuracy, decision time, response delay, and distance and speed changes between the autonomous driving system and the target vehicle. These data can provide a basis for subsequent performance analysis and safety assessment.

[0117] Furthermore, the generated scene is evaluated in multiple dimensions. The detailed steps are as follows:

[0118] 1. Safety assessment: In the generated high-risk scenarios, the focus is on assessing the safety performance of the autonomous driving system. This includes whether the autonomous driving system can successfully avoid accidents when faced with complex interactions and potential collisions, and whether it can make reasonable decisions to ensure the safety of itself and other traffic participants. Through simulated high-risk scenarios such as sudden lane changes, emergency braking, or multiple vehicles approaching, the response capabilities and safety boundaries of the autonomous driving system are verified.

[0119] 2. Robustness analysis: Evaluate the robustness of the autonomous driving system in dealing with a variety of complex scenarios. This requires the system to maintain consistent performance in different types of scenarios, facing different traffic conditions, vehicle behaviors, weather conditions and other challenges. Robustness analysis usually covers a variety of test environments, such as urban roads, highways, rainy days, nighttime, etc., to ensure that the autonomous driving system performs stably in a changing real-world environment.

[0120] 3. Ability to deal with multi-vehicle interactions: Multi-vehicle interactions are the core content of the generated scenarios. Therefore, the evaluation focus should include how the autonomous driving system responds when multiple target vehicles interact with it. The specific evaluation includes whether the system can effectively identify threats and make reasonable responses when dealing with complex interactions such as multiple vehicles approaching, changing lanes, and overtaking at the same time, and whether it can avoid potential collisions between multiple vehicles.

[0121] 4. Behavior diversity and rationality: Analyze the diversity and rationality of the target vehicle's behavior in the scenarios generated by the agent. The target vehicle should exhibit a variety of complex and realistic driving behaviors, such as sudden acceleration, braking, lane changes, etc., to challenge the decision-making ability of the autonomous driving system. At the same time, the rationality of these behaviors also needs to be confirmed to ensure that the generated scenarios have practical application value and can represent the potential risks in real driving scenarios.

[0122] 5. Autonomous driving system performance regression: Based on the test results, the performance of the autonomous driving system is regressed. By comparing the performance of the system in different scenarios, it is determined whether its performance is stable within a certain range, and whether it shows weaknesses or wrong decisions in specific types of scenarios. Through this process, the system's response strategy can be further optimized and its overall performance can be improved.

[0123] See also Figure 2 The embodiment of the present invention further provides a device for generating a multi-vehicle interactive test scenario for autonomous driving, comprising:

[0124] An acquisition module 21 is used to decompose the scene according to the design requirements of the autonomous driving test scene, and extract necessary static environment information and dynamic parameter distribution of traffic participants;

[0125] The sampling module 22 is used to sample the static environment information and the dynamic parameters of the traffic participants, extract the scene into a number of scene sections, and divide them into test scene samples and verification scene samples according to a preset ratio;

[0126] A training module 23, used for inputting the test scene sample and the verification scene sample into a multi-agent model training to obtain a trained multi-agent model;

[0127] The output module 24 is used to deploy the trained multi-agent model to the test scene sample, perform interactive testing with the tested autonomous driving system, and obtain an autonomous driving multi-vehicle interactive test scene.

[0128] The present application also provides a computer device, including: a memory and a processor, wherein the memory stores computer program instructions for execution on the processor, and when the processor executes the computer program instructions, the method for generating a multi-vehicle interactive test scenario for autonomous driving as described above is implemented.

[0129] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for generating a multi-vehicle interactive test scenario for autonomous driving as described above are implemented.

[0130] By integrating the functions of these modules, the constructed device can not only achieve efficient and accurate generation of multi-vehicle interactive test scenarios for autonomous driving, but also improve the accuracy of robustness assessment of autonomous driving vehicles, which helps to improve the safety and reliability of autonomous driving vehicles and promote the development of their practical applications.

[0131] The above examples are only reference examples. In order to avoid redundancy, they are not listed one by one here. In actual development or application, they can be flexibly combined according to actual needs, but any combination belongs to the technical solution of this application and is covered within the scope of protection of this application.

[0132] An embodiment of the present application also provides a computer device, including a processor and a memory, wherein the memory stores computer program instructions for execution on the processor, and when the processor executes the computer program instructions, the method for generating an autonomous driving multi-vehicle interactive test scenario as described above is implemented.

[0133] An embodiment of the present application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implements the method for generating a multi-vehicle interactive test scenario for autonomous driving as described above.

[0134] It is understood that the above scenarios are only examples and do not constitute a limitation on the application scenarios of the technical solutions provided in the embodiments of the present application. The technical solutions of the present application can also be applied to other scenarios. For example, it is known to those skilled in the art that with the evolution of the system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0135] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0136] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs.

[0137] The units in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0138] In the present application, the same or similar terminology concepts, technical solutions and / or application scenario descriptions are generally described in detail only the first time they appear. When they appear again later, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of the present application, for the same or similar terminology concepts, technical solutions and / or application scenario descriptions that are not described in detail later, reference can be made to the previous related detailed descriptions.

[0139] In the present application, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0140] The various technical features of the technical solution of the present application can be arbitrarily combined. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application.

[0141] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, a controlled terminal, or a network device, etc.) to execute the method of each embodiment of the present application.

[0142] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a storage medium, or transmitted from one storage medium to another storage medium. For example, computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integration. Available media can be magnetic media, (e.g., floppy disk, storage disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state storage disk Solid State Disk (SSD)), etc.

[0143] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for generating a multi-vehicle interactive test scenario for autonomous driving, characterized in that: The method comprises: Step S1: Decompose the scene according to the design requirements of the autonomous driving test scene, and extract the necessary static environment information and dynamic parameter distribution of traffic participants; Step S2, sampling static environmental information and dynamic parameters of traffic participants, extracting the scene into a number of scene sections, and dividing them into test scene samples and verification scene samples according to a preset ratio; Step S3, inputting the test scene sample and the verification scene sample into the multi-agent model training to obtain a trained multi-agent model; Step S4: deploy the trained multi-agent model to the test scene sample, perform interactive testing with the autonomous driving system under test, and obtain an autonomous driving multi-vehicle interactive test scene.

2. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 1, characterized in that: The static environment information includes: vehicle type of traffic participants, road type, road width, number of lanes, lighting conditions, weather conditions, traffic control method, and number of background vehicles.

3. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 1, characterized in that: The dynamic parameters of the traffic participant are the state of the traffic participant at each time step, including: the longitudinal coordinate, the transverse coordinate, the speed, acceleration, direction, turn, and relative position difference at each coordinate within each step.

4. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 1, characterized in that: The static environment information and the dynamic parameters of the traffic participants are sampled, the scene is extracted into a plurality of scene sections, and divided into test scene samples and verification scene samples according to a preset ratio, including: Static environmental information and dynamic parameters of traffic participants are sampled at preset time intervals to generate several independent scene sections. The data of each scene section must include static information, dynamic information, and the relative position and behavioral characteristics of each participating vehicle, and are divided into test scene samples and verification scene samples according to a preset ratio.

5. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 1, characterized in that: The step of inputting the test scene sample and the verification scene sample into a multi-agent model training to obtain a trained multi-agent model comprises: Input the test scene samples into the multi-agent model training to generate a dynamic driving scene set; Based on the dynamic driving scenario set and the verification scenario samples, a multi-agent model is input, a reward function is set, and when the training loss value reaches convergence, a trained multi-agent model is obtained.

6. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 1, characterized in that: The trained multi-agent model is deployed to the test scene sample, and interactive testing is performed with the tested autonomous driving system to obtain an autonomous driving multi-vehicle interactive test scene, including: Deploy the trained multi-agent model to the test scenario sample and set up the tested autonomous driving system to ensure that its parameters and states match the test scenario sample; In the test scene sample, the target vehicle controlled by the trained multi-agent model and the autonomous driving system under test together constitute an interactive environment; An interactive test is performed based on the test scenario samples to generate an autonomous driving multi-vehicle interactive test scenario.

7. The method for generating a multi-vehicle interactive test scenario for autonomous driving according to claim 6, characterized in that: After the interactive test is performed based on the test scenario sample to generate an autonomous driving multi-vehicle interactive test scenario, the method further includes: Obtain the response data of the autonomous driving system and the behavior data of the target vehicle, including the perception accuracy, decision time, response delay, and distance and speed changes of the autonomous driving system from the target vehicle.

8. A device for generating interactive test scenarios for autonomous driving multiple vehicles, characterized in that: include: The acquisition module is used to decompose the scene according to the design requirements of the autonomous driving test scene and extract the necessary static environment information and dynamic parameter distribution of traffic participants; A sampling module is used to sample static environmental information and dynamic parameters of traffic participants, extract the scene into several scene sections, and divide them into test scene samples and verification scene samples according to a preset ratio; A training module, used for inputting the test scenario samples and the verification scenario samples into a multi-agent model training to obtain a trained multi-agent model; The output module is used to deploy the trained multi-agent model to the test scene sample, conduct interactive testing with the autonomous driving system under test, and obtain an autonomous driving multi-vehicle interactive test scene.

9. A computer device, characterized in that: It includes a processor and a memory, the memory stores computer program instructions for execution on the processor, and when the processor executes the computer program instructions, it implements the method for generating an autonomous driving multi-vehicle interactive test scenario as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: It stores computer instructions, which, when executed by a processor, implement the method for generating an autonomous driving multi-vehicle interactive test scenario as described in any one of claims 1 to 7.

Images