Vehicle confrontation method and device, storage medium and product

By deploying a simulation platform in a cloud server and using the control model to control the target in different scenarios, the problem that the existing technology cannot comprehensively evaluate the performance of the autonomous driving system in complex traffic environments is solved, and efficient and safe autonomous driving tests are achieved .

CN118961237APending Publication Date: 2024-11-15CHONGQING CHANGAN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411086952.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing simulation methods cannot comprehensively evaluate the performance of autonomous driving systems in a diverse, complex and unpredictable real traffic environment. Traditional road tests are costly and inefficient, and pose safety risks.

Method used

By deploying a simulation platform in a cloud server, using a preset control model to input the vehicle information and environmental information of the main vehicle in different scenarios, a control signal for opposing targets is generated, based on this signal, the opposing targets are controlled in the simulation platform. Make it confrontation with the main car in various scenarios to achieve a comprehensive test of the main car.

Benefits of technology

The confrontational testing of autonomous vehicles in different scenarios is realized, and its performance in diverse and complex traffic environments is comprehensively evaluated, testing efficiency and safety is improved, and the direction and speed of the confrontation target can be accurately controlled in simulation scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118961237A_ABST
    Figure CN118961237A_ABST
Patent Text Reader

Abstract

The invention relates to a vehicle confrontation method and device, a storage medium and a product, and relates to the technical field of automatic driving. The method comprises the steps of inputting vehicle information and environment information of a main vehicle in a target scene into a preset control model to obtain a control signal of a confrontation target, and controlling the confrontation target in a simulation platform based on the control signal to enable the confrontation target to have a confrontation behavior with the main vehicle in the target scene. Therefore, the performance of the self-driving automobile (namely the main automobile) in a diversified, complex and unpredictable real traffic environment can be comprehensively evaluated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, in particular to the field of vehicle confrontation technology, and specifically to a vehicle confrontation method, device, storage medium and product. Background Art

[0002] With the continuous development of autonomous driving technology, safety and reliability have become key indicators for evaluating autonomous driving systems. Traditional autonomous driving testing methods mainly rely on road testing in real environments, which is not only costly and inefficient, but may also bring safety hazards. In order to solve these problems, simulation technology is widely used in the research and development of autonomous driving systems.

[0003] Existing simulation methods usually only simulate a single scenario or specific conditions and cannot comprehensively evaluate the performance of autonomous driving systems in diverse, complex and unpredictable real traffic environments. Summary of the invention

[0004] This application provides a vehicle confrontation method, device, storage medium and product to solve the technical problem that the related technology cannot fully evaluate the performance of the autonomous driving system in a diverse, complex and unpredictable real traffic environment. The technical solution of this application is as follows:

[0005] According to a first aspect of the present application, a vehicle confrontation method is provided, which is applied to a cloud server, and a simulation platform is deployed in the cloud server, including:

[0006] The vehicle information and environmental information of the main vehicle in the target scene are input into the preset control model to obtain the control signal of the antagonistic target. Based on the control signal, the antagonistic target is controlled in the simulation platform so that the antagonistic target performs antagonistic behavior with the main vehicle in the target scene.

[0007] The antagonistic target and the host vehicle are virtual objects pre-generated in the simulation platform. The host vehicle corresponds to different vehicle information and environmental information in different scenarios. The control signal is used to control the direction and / or speed of the antagonistic target.

[0008] Through the above technical means, the main vehicle corresponds to different vehicle information and environmental information in different scenarios. Therefore, when conducting confrontation tests on the main vehicle, the cloud server can control the confrontation target in different scenarios, so that the confrontation target will have confrontational behavior with the main vehicle in the simulation scenario, thereby realizing confrontation tests on the main vehicle. Therefore, this application can realize confrontation tests on the main vehicle in different scenarios, and then comprehensively evaluate the performance of the autonomous driving car (i.e., the main vehicle) in a diverse, complex and unpredictable real traffic environment.

[0009] In a possible implementation, the control signal includes a lateral control signal and / or a longitudinal control signal, the lateral control signal is used to control the direction of the confrontation target, and the longitudinal control signal is used to control the speed of the confrontation target.

[0010] Through the above technical solution, when controlling the confrontation target, not only the direction of the confrontation target can be controlled, but also the speed of the confrontation target can be controlled, so as to achieve more precise control of the confrontation target.

[0011] In one possible implementation, the control model includes a lateral control model and / or a longitudinal control model, the lateral control model is used to generate a lateral control signal to control the direction of the confrontation target, and the longitudinal control model is used to generate a longitudinal control signal to control the speed of the confrontation target.

[0012] On this basis, the vehicle information and environmental information of the main vehicle in the target scene are input into a preset control model to obtain a control signal for the counter target, including: inputting the vehicle information and environmental information of the main vehicle in the target scene into the lateral control model and / or the longitudinal control model respectively to obtain the lateral control signal and / or the longitudinal control signal of the counter target.

[0013] In a possible implementation, the method further includes: training the first control model based on a lateral reward function corresponding to the target scene to obtain a lateral control model; different scenes correspond to different lateral reward functions; and / or, training the second control model based on a longitudinal reward function corresponding to the target scene to obtain a longitudinal control model; different scenes correspond to different longitudinal reward functions.

[0014] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the control model includes a lateral control model; the lateral reward function corresponding to the target scene is related to the time when the main vehicle and the confrontation target arrive at the intersection area.

[0015] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the lateral reward function corresponding to the target scene is:

[0016]

[0017] Among them, t1 represents the time when the main vehicle arrives at the intersection, and t2 represents the time when the adversarial target arrives at the intersection.

[0018] In a possible implementation, the target scenario is a ramp scenario or a lane addition and reduction scenario, and the control model includes a lateral control model and a longitudinal control model; the lateral reward function corresponding to the target scenario is related to the lateral distance between the host vehicle and the adversarial target; the lateral distance refers to the distance between the host vehicle and the adversarial target in a direction perpendicular to the lane line; the longitudinal reward function corresponding to the target scenario is related to the longitudinal distance between the host vehicle and the adversarial target; the longitudinal distance refers to the distance between the host vehicle and the adversarial target in a direction parallel to the lane line.

[0019] In a possible implementation, the target scene is a ramp scene or a lane addition and subtraction scene, and the lateral reward function corresponding to the target scene is:

[0020]

[0021] Where L represents the distance between the host vehicle and the adversarial target in the direction perpendicular to the lane line;

[0022] The vertical reward function corresponding to the target scene is:

[0023]

[0024] Where X represents the distance between the host vehicle and the adversarial target in the direction parallel to the lane line.

[0025] In a possible implementation, the target scene is any one of an intersection scene, a pedestrian scene, a ramp scene, a roundabout scene, and a lane addition and reduction scene.

[0026] Through the above technical scheme, the target scenario is explained, so that the vehicle confrontation method provided in this application can be compatible with a variety of scenarios, that is, it can realize joint simulation with the simulation platform in multiple scenarios such as intersection scenarios, pedestrian scenarios, ramp scenarios, roundabout scenarios and lane addition and reduction scenarios, so as to realize comprehensive testing of autonomous driving vehicles (i.e., the main vehicle).

[0027] In a possible implementation, the target scene identifies that the host vehicle and the counter target intersect in a first area within a target lane of the host vehicle, where the target lane of the host vehicle refers to a lane in which the host vehicle intends to travel.

[0028] The above-mentioned controlling the antagonistic target in the simulation platform based on the control signal so that the antagonistic target engages in antagonistic behavior with the main vehicle in the target scene includes: controlling the antagonistic target in the simulation platform based on the control signal so that the antagonistic target arrives at the first area of ​​the target lane at the same time as the main vehicle, or so that the antagonistic target arrives at the first area of ​​the target lane before the main vehicle.

[0029] Through the above technical scheme, the antagonistic effect of controlling the antagonistic target in the scenario where the main vehicle and the antagonistic target intersect in the first area within the target lane of the main vehicle is explained, so that the antagonistic target can effectively confront the main vehicle in the first area of ​​the target lane, thereby realizing the test of the main vehicle.

[0030] In a possible implementation, the host vehicle travels in a straight line, and the lane the host vehicle is currently traveling in is the same as the lane the confrontation target is currently traveling in.

[0031] The above-mentioned controlling the antagonistic target in the simulation platform based on the control signal so that the antagonistic target engages in antagonistic behavior with the main vehicle in the target scene includes: when the position of the antagonistic target is before the position of the main vehicle, controlling the antagonistic target based on the control signal so that the antagonistic target brakes the main vehicle; or, when the position of the main vehicle is before the position of the antagonistic target, controlling the antagonistic target based on the control signal so that the distance between the antagonistic target and the main vehicle is less than a first distance threshold.

[0032] Through the above technical scheme, the antagonistic effect of controlling the antagonistic target is explained in the scenario where the main vehicle is traveling in a straight line and the lane the main vehicle is currently traveling in is the same as the lane the antagonistic target is currently traveling in, so that the antagonistic target can follow the vehicle or brake the main vehicle to achieve effective confrontation with the main vehicle, thereby realizing the test of the main vehicle.

[0033] In one possible implementation, the environmental information includes at least one of the following: vehicle information of vehicles surrounding the main vehicle; the surrounding vehicles of the main vehicle include vehicles whose distance from the main vehicle is less than a second distance threshold; road information; the road information includes at least the width and number of roads; pedestrian information; the pedestrian information includes information of pedestrians whose distance from the main vehicle is less than a third distance threshold.

[0034] Through the above technical solution, the environmental information is explained so that the environmental information and vehicle information can be combined to accurately control the confrontation target, thereby accurately achieving confrontation with the main vehicle.

[0035] In a possible implementation, the confrontation target includes any one of a confrontation vehicle and a confrontation pedestrian. The confrontation vehicle refers to a vehicle that engages in confrontational behavior with the main vehicle; the confrontation pedestrian refers to a pedestrian that engages in confrontational behavior with the main vehicle.

[0036] In a possible implementation, the method further includes:

[0037] At least one adversarial target is generated in the simulation platform based on the scenario type of the target scenario.

[0038] In a possible implementation, generating at least one adversarial target in a target scene includes:

[0039] A target lane for the host vehicle is determined, and based on the target lane for the host vehicle, at least one confrontation target is generated in an opposite lane to a lane in which the host vehicle is currently traveling, a lane adjacent to a lane in which the host vehicle is currently traveling, and / or a lane in which the host vehicle is currently traveling.

[0040] The target lane of the adversary target is the same as the target lane of the host vehicle. The target lane of the host vehicle refers to the lane that the host vehicle intends to travel.

[0041] Through the above technical solution, the target lane of the generated confrontation target can be made consistent with the target lane of the main vehicle, so that the subsequent confrontation target and the main vehicle can effectively confront each other in the target lane, thereby realizing the test of the main vehicle.

[0042] According to a second aspect provided by the present application, a vehicle confrontation device is provided, the device comprising: a generation unit and a confrontation unit, wherein:

[0043] The generation unit is used to input the vehicle information and environmental information of the host vehicle in the target scene into a preset control model to obtain a control signal of the antagonistic target; the antagonistic target and the host vehicle are virtual objects pre-generated in the simulation platform; the host vehicle corresponds to different vehicle information and environmental information in different scenes; the control signal is used to control the direction and / or speed of the antagonistic target.

[0044] The confrontation unit is used to control the confrontation target in the simulation platform based on the control signal, so that the confrontation target performs confrontation behavior with the main vehicle in the target scene.

[0045] In a possible implementation, the control signal includes a lateral control signal and / or a longitudinal control signal, the lateral control signal is used to control the direction of the confrontation target, and the longitudinal control signal is used to control the speed of the confrontation target.

[0046] In one possible implementation, the control model includes a lateral control model and / or a longitudinal control model, the lateral control model is used to generate a lateral control signal to control the direction of the confrontation target, and the longitudinal control model is used to generate a longitudinal control signal to control the speed of the confrontation target.

[0047] In a possible implementation, the device also includes a training unit, which is specifically used to: train the first control model based on the lateral reward function corresponding to the target scene to obtain a lateral control model; different scenes correspond to different lateral reward functions; and / or, train the second control model based on the longitudinal reward function corresponding to the target scene to obtain a longitudinal control model; different scenes correspond to different longitudinal reward functions.

[0048] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the control model includes a lateral control model; the lateral reward function corresponding to the target scene is related to the time when the main vehicle and the confrontation target arrive at the intersection area.

[0049] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the lateral reward function corresponding to the target scene is:

[0050]

[0051] Among them, t1 represents the time when the main vehicle arrives at the intersection, and t2 represents the time when the adversarial target arrives at the intersection.

[0052] In a possible implementation, the target scenario is a ramp scenario or a lane addition and reduction scenario, and the control model includes a lateral control model and a longitudinal control model; the lateral reward function corresponding to the target scenario is related to the lateral distance between the host vehicle and the adversarial target; the lateral distance refers to the distance between the host vehicle and the adversarial target in a direction perpendicular to the lane line; the longitudinal reward function corresponding to the target scenario is related to the longitudinal distance between the host vehicle and the adversarial target; the longitudinal distance refers to the distance between the host vehicle and the adversarial target in a direction parallel to the lane line.

[0053] In a possible implementation, the target scene is a ramp scene or a lane addition and subtraction scene, and the lateral reward function corresponding to the target scene is:

[0054]

[0055] Where L represents the distance between the host vehicle and the adversarial target in the direction perpendicular to the lane line;

[0056] The vertical reward function corresponding to the target scene is:

[0057]

[0058] Where X represents the distance between the host vehicle and the adversarial target in the direction parallel to the lane line.

[0059] According to the third aspect provided by the present application, a server is provided, comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to execute instructions to implement the vehicle countermeasure method of the above-mentioned first aspect and any possible implementation manner thereof.

[0060] According to the fourth aspect provided by the present application, a computer-readable storage medium is provided. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the vehicle countermeasure method in the above-mentioned first aspect and any possible implementation manner thereof.

[0061] According to the fifth aspect provided by the present application, a computer program product is provided, which includes computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the vehicle countermeasure method of the above-mentioned first aspect and any possible implementation manner thereof.

[0062] Therefore, the above technical features of the present application have the following beneficial effects:

[0063] (1) The main vehicle corresponds to different vehicle information and environmental information in different scenarios. Therefore, when conducting confrontation tests on the main vehicle, the cloud server can control the confrontation target in different scenarios so that the confrontation target can engage in confrontational behavior with the main vehicle in the simulation scenario, thereby realizing confrontation tests on the main vehicle. Therefore, the present application can realize confrontation tests on the main vehicle in different scenarios, and then comprehensively evaluate the performance of the autonomous driving vehicle (i.e., the main vehicle) in a diverse, complex, and unpredictable real traffic environment.

[0064] (2) Through the above technical solution, when controlling the confrontation target, not only the direction of the confrontation target can be controlled, but also the speed of the confrontation target can be controlled, so as to achieve more precise control of the confrontation target.

[0065] (3) In the scenario where the host vehicle and the antagonistic target intersect in the first area within the target lane of the host vehicle, the antagonistic effect of controlling the antagonistic target is explained, so that the antagonistic target can effectively confront the host vehicle in the first area of ​​the target lane, thereby realizing the test of the host vehicle.

[0066] (4) The environmental information is explained so that the environmental information and vehicle information can be combined to accurately control the confrontation target and thus accurately achieve confrontation with the main vehicle.

[0067] (5) The target lane of the generated adversarial target can be made consistent with the target lane of the host vehicle, so that the subsequent adversarial target and the host vehicle can effectively confront each other in the target lane, thereby realizing the test of the host vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application, and do not constitute improper limitations on the present application.

[0069] Figure 1 is a flow chart of a vehicle confrontation method according to an exemplary embodiment;

[0070] Figure 2 is a structural diagram of a cloud server according to an exemplary embodiment;

[0071] Figure 3is a schematic diagram of a structure of an adversarial model according to an exemplary embodiment;

[0072] Figure 4 is a schematic diagram of a target scene according to an exemplary embodiment;

[0073] Figure 5 is a schematic diagram of another target scene according to an exemplary embodiment;

[0074] Figure 6 is a schematic diagram of another target scene according to an exemplary embodiment;

[0075] Figure 7 is a schematic diagram showing a non-intersection scenario according to an exemplary embodiment;

[0076] Figure 8 is a schematic diagram of a T-junction intersection scene according to an exemplary embodiment;

[0077] Fig. 9 is a schematic diagram showing a lane adding and merging scenario according to an exemplary embodiment;

[0078] Fig.10 is a schematic diagram showing a lane reduction merging scenario according to an exemplary embodiment;

[0079] Fig.11 is a network structure diagram of a PPO and GAIL algorithm model according to an exemplary embodiment;

[0080] Fig.12 is a structural diagram of a conditional diffusion behavior cloning algorithm according to an exemplary embodiment;

[0081] Fig.13 is a block diagram of a vehicle countermeasure device according to an exemplary embodiment;

[0082] Fig.14 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0083] In order to enable ordinary persons in the art to better understand the technical solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0084] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0085] With the continuous development of autonomous driving technology, safety and reliability have become key indicators for evaluating autonomous driving systems. Traditional autonomous driving testing methods mainly rely on road testing in real environments, which is not only costly and inefficient, but may also bring safety hazards. In order to solve these problems, simulation technology is widely used in the research and development of autonomous driving systems.

[0086] Vehicle confrontation scenarios can be generated through simulation technology, so that autonomous vehicles can be tested on the road in vehicle confrontation scenarios. Currently, there are two main ways to generate vehicle confrontation scenarios:

[0087] Method 1: Extract scene elements and build a basic scenario library. Then generate adversarial scenarios based on adversarial elements and generate adversarial scenarios based on visual image derivative methods. Finally, evaluate and integrate the adversarial scenario library to systematically sort out all commonly used adversarial scenario elements in autonomous driving test scenarios.

[0088] Method 2: Extract the benefit elements and cost elements of the target confrontation scenario, and construct a mathematical model of the target confrontation scenario based on the benefit elements and cost elements; obtain the road environment information and the node information of the two opposing parties in the target confrontation scenario, and model the road environment information and node information based on the mathematical model to generate a graph theory model of the target confrontation scenario; obtain the state information of our nodes in the two opposing parties and the directional characteristic information and benefit characteristic information of the opponent's nodes, and optimize the graph theory model according to the directional characteristic information, benefit characteristic information, state information and preset hierarchical strategy to construct a graph theory optimization model of the target confrontation scenario.

[0089] In the above method 1, the focus is mainly on the generation of confrontation scenarios at the perception level, and confrontation under specific pictures is required to be effective for the perception system. In actual applications, it is very slow to find matching perception confrontation pictures, and there is a lack of generation of confrontation scenarios related to vehicle interaction behaviors, which cannot interfere with the autonomous driving control system. In the above method 2, the focus is mainly on generating confrontational road network structures and vehicle information from the perspective of graph theory, mainly modeling the environment to form a vehicle game between single agents, and there is no joint simulation with the existing simulation environment.

[0090] In summary, existing simulation methods usually only simulate a single scenario or specific conditions and cannot comprehensively evaluate the performance of autonomous driving systems in diverse, complex and unpredictable real traffic environments.

[0091] In order to solve the technical problem of only simulating a single scene or specific conditions in the prior art, an embodiment of the present application provides a vehicle confrontation method, which can input the vehicle information and environmental information of the main vehicle in the target scene into a preset control model to obtain a control signal of the confrontation target, and control the confrontation target in the simulation platform based on the control signal, so that the confrontation target will have confrontation behavior with the main vehicle in the target scene. The confrontation target and the main vehicle are virtual objects pre-generated in the simulation platform. The main vehicle corresponds to different vehicle information and environmental information in different scenes, and the control signal is used to control the direction and / or speed of the confrontation target.

[0092] From the above content, it can be seen that in the embodiment of the present application, the main vehicle corresponds to different vehicle information and environmental information in different scenarios. Therefore, when conducting confrontation tests on the main vehicle, the cloud server can control the confrontation target in different scenarios so that the confrontation target will have confrontational behavior with the main vehicle in the simulation scenario, thereby realizing confrontation tests on the main vehicle. Therefore, the present application can realize confrontation tests on the main vehicle in different scenarios, and then comprehensively evaluate the performance of the autonomous driving car (i.e., the main vehicle) in a diverse, complex and unpredictable real traffic environment.

[0093] Figure 1 is a flow chart of a vehicle confrontation method according to an exemplary embodiment. Figure 1 As shown, the vehicle confrontation method includes the following steps:

[0094] S101, inputting vehicle information and environmental information of the host vehicle in the target scene into a preset control model to obtain a control signal for countering the target.

[0095] The main vehicle corresponds to different vehicle information and environmental information in different scenarios.

[0096] The embodiment of the present application does not limit the vehicle information of the main vehicle. For example, the vehicle information of the main vehicle may include the position, speed, heading angle, etc. of the main vehicle.

[0097] The embodiment of the present application does not limit the environmental information of the main vehicle. For example, the environmental information of the main vehicle may include vehicle information of the vehicles surrounding the main vehicle, road information, such as road information of the road where the main vehicle is located, road information of the target lane of the main vehicle, etc., pedestrian information, etc.

[0098] The surrounding vehicles of the host vehicle include vehicles whose distances to the host vehicle are less than a second distance threshold.

[0099] The pedestrian information includes information of pedestrians whose distance from the host vehicle is less than a third distance threshold.

[0100] The embodiment of the present application does not limit the second distance threshold and the third distance. For example, the second distance threshold may be 100 meters or 50 meters, and the third distance threshold may be 50 meters or 40 meters.

[0101] The adversarial target is a virtual object generated in advance in the simulation platform.

[0102] The control signal of the adversarial target refers to the information used to control the adversarial target.

[0103] The embodiment of the present application does not limit the confrontation target. For example, the confrontation target may be a confrontation vehicle or a confrontation pedestrian. A confrontation vehicle refers to a vehicle that is in a confrontation with the main vehicle. A confrontation pedestrian refers to a pedestrian that is in a confrontation with the main vehicle.

[0104] S102, based on the control signal, controlling the antagonistic target in the simulation platform so that the antagonistic target performs antagonistic behavior with the host vehicle in the target scene.

[0105] In an optional embodiment, the control signal may include a lateral control signal and / or a longitudinal control signal. The lateral control signal is used to control the direction of the antagonistic target, and the longitudinal control signal is used to control the speed of the antagonistic target. Accordingly, the above S102 may be replaced by: based on the lateral control signal and / or the longitudinal control signal, controlling the antagonistic target in the simulation platform so that the antagonistic target performs antagonistic behavior with the host vehicle in the target scene.

[0106] In an optional implementation, it is assumed that the target scenario can identify that the host vehicle and the antagonistic target intersect in the first area of ​​the host vehicle's target lane, and the host vehicle's target lane refers to the lane that the host vehicle intends to travel in. On this basis, when the antagonistic target is controlled in the simulation platform, the cloud server can control the antagonistic target in the simulation platform based on the control signal so that the antagonistic target reaches the first area of ​​the target lane at the same time as the host vehicle, or so that the antagonistic target reaches the first area of ​​the target lane before the host vehicle.

[0107] The embodiments of the present application do not limit the lane in which the main vehicle intends to travel. For example, the lane in which the main vehicle intends to travel may be the lane in which the main vehicle is currently traveling, the lane in which the antagonist target is currently traveling, the lane adjacent to the lane in which the main vehicle is currently traveling, or the lane adjacent to the lane in which the antagonist target is currently traveling.

[0108] In an optional implementation, it is assumed that the host vehicle is traveling in a straight line, and the lane that the host vehicle is currently traveling in is the same as the lane that the antagonistic target is currently traveling in. On this basis, when controlling the antagonistic target in the simulation platform, the cloud server can control the antagonistic target based on the control signal when the position of the antagonistic target is ahead of the position of the host vehicle, so that the antagonistic target brakes the host vehicle; or, when the position of the host vehicle is ahead of the position of the antagonistic target, control the antagonistic target based on the control signal so that the distance between the antagonistic target and the host vehicle is less than the first distance threshold.

[0109] The embodiment of the present application does not limit the first distance threshold. The first distance threshold can be 50 meters or 20 meters, which is not limited in the embodiment of the present application.

[0110] In the embodiment of the present application, the main vehicle corresponds to different vehicle information and environmental information in different scenarios. Therefore, when conducting confrontation tests on the main vehicle, the cloud server can control the confrontation target in different scenarios so that the confrontation target will have confrontational behavior with the main vehicle in the simulation scenario, thereby realizing confrontation tests on the main vehicle. Therefore, the present application can realize confrontation tests on the main vehicle in different scenarios, and then comprehensively evaluate the performance of the autonomous driving car (i.e., the main vehicle) in a diverse, complex, and unpredictable real traffic environment.

[0111] In the embodiment of the present application, the cloud server may be deployed with Figure 2 The simulation platform, communication module and confrontation components shown, Figure 1 The vehicle confrontation method shown can also be implemented by the interaction of the simulation platform, the communication module and the confrontation component. Specifically, the communication module can obtain RDB information containing vehicle information and environmental information of the main vehicle in the target scene from the simulation platform through the pre-configured sensor dedicated port 4819.

[0112] The vehicle information of the main vehicle may include the speed, position, heading angle and acceleration of the main vehicle, etc. The environmental information of the main vehicle may include the vehicle information, lane information, road network information, traffic sign information and traffic light information of the surrounding vehicles of the main vehicle.

[0113] The communication module can send the received RDB information to the confrontation component, and the confrontation component can parse the received RDB information to obtain the vehicle information and environmental information of the main vehicle in the target scene. After that, the confrontation component can determine the confrontation target from the surrounding vehicles of the main vehicle based on the vehicle information and environmental information of the main vehicle, and generate a control signal for controlling the confrontation target based on the vehicle information and environmental information of the main vehicle, and send the control signal to the communication module. The communication module can send the control signal to the simulation platform through the 4819 port. After receiving the control signal, the simulation platform can control the confrontation target so that the confrontation target will engage in confrontation with the main vehicle in the target scene.

[0114] Optionally, a pre-trained adversarial model can be deployed in the adversarial component. When the adversarial component generates a control signal for controlling the adversarial target based on the vehicle information and environmental information of the main vehicle, the vehicle information of the main vehicle and the vehicle information of the adversarial target (such as the speed, position, heading angle and acceleration of the adversarial target) can be input into the adversarial model. The adversarial model can output a control signal for controlling the adversarial target.

[0115] For example, taking the case where the control signal includes an acceleration signal and a steering angle signal, and the acceleration signal is used to control the speed of the confrontation target, and the steering angle signal is used to control the direction of the confrontation target, after the confrontation component inputs the vehicle information of the main vehicle and the vehicle information of the confrontation target into the confrontation model, the confrontation model can output the acceleration and steering angle for controlling the confrontation target, and the confrontation component can generate an acceleration signal based on the acceleration, and generate a steering angle signal based on the steering angle.

[0116] Among them, the acceleration range is [-5m / s 2 ,5m / s 2 ], and the steering angle range is between [-π / 3rad,π / 3rad].

[0117] Optionally, the communication module may send a control signal to the simulation platform in the following two ways:

[0118] Method 1: Use SCP signals to achieve lateral and vertical control of the target.

[0119] Method 2: Replace the scp signal with RDB information to achieve real-time control of the adversarial target. Specifically, by re-encoding the signal DRIVER_CTRL used for vehicle control in the RDB information, and writing the corresponding python parsing module according to the specification of the RDB information, the RDB information that can be recognized by the simulation platform for lateral and longitudinal control of the adversarial target is generated.

[0120] Figure 3 A schematic diagram of a confrontation model provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the adversarial model includes a strategy model and a control model, wherein the strategy model is used to output corresponding control instructions to the control model according to the input vehicle information and environmental information of the main vehicle in the target scene, and the control model is used to generate a control signal for controlling the adversarial target according to the input control instructions.

[0121] The strategy model and control model are described below respectively.

[0122] First, taking the target scenarios of intersection scenes, pedestrian scenes, ramp scenes, roundabout scenes, and lane addition and reduction scenes as examples, the output, output, and confrontation effects of the strategy model are explained.

[0123] 1. Intersection Scene

[0124] The embodiment of the present application does not limit the intersection scene. For example, the intersection scene can be a crossroads or a T-junction.

[0125] In the intersection scenario, the vehicle information of the host vehicle input into the strategy model may include: the speed of the host vehicle, the position of the host vehicle, and the heading of the host vehicle.

[0126] Environmental information may include: whether it is possible to change lanes left or right, the type of traffic light, whether there are pedestrians on the zebra crossing ahead, the distance between the confrontation target and the stop line, the width between the current lane where the main vehicle is currently traveling and the target lane, the angle between the heading of the confrontation target and the current lane, the angle between the heading of the confrontation target and the target lane, the lateral offset between the center of the confrontation target and the current lane, the lateral offset between the center of the confrontation target and the target lane, the absolute speed of the confrontation target (in both x and y directions), the heading angle of the confrontation target, the relative speed between the confrontation target and the main vehicle, the relative distance between the confrontation target and the main vehicle, the relative heading between the confrontation target and the main vehicle, the relative distance between the confrontation target and the five nearest vehicles around it, the relative speed between the confrontation target and the five nearest vehicles around it, the relative heading between the confrontation target and the five nearest vehicles around it, and other 46-dimensional information.

[0127] The control instructions output by the strategy model may include: 0: LEFT (turn left), 1: IDLE (maintain current state), 2: RIGHT (turn right), 3: FASTER (accelerate), 4: SLOWER (decelerate).

[0128] The countermeasure effects to be achieved by the above control instructions include:

[0129] 1. If the opponent and the main vehicle have an intersection at the intersection, the opponent needs to seize the intersection and brake at the intersection or confront the main vehicle.

[0130] 2. If the main vehicle is on the side of the opponent, if the main vehicle is going straight, the opponent needs to accelerate and cut into the lane where the main vehicle is located. If the main vehicle changes lanes to the lane where the opponent is located, the opponent needs to accelerate and occupy the lane to force the main vehicle to fail to change lanes.

[0131] 3. If the host vehicle is behind the opposing vehicle, the opposing vehicle needs to brake. If the host vehicle changes lanes to the target lane, the opposing vehicle also needs to change lanes to the target lane.

[0132] 4. If the host vehicle is in front of the antagonist, the antagonist needs to follow the vehicle closely so that the distance between the antagonist and the host vehicle is less than the first distance threshold.

[0133] 2. Pedestrian Scene

[0134] In the pedestrian scenario, the vehicle information of the host vehicle input into the strategy model may include: the speed of the host vehicle, the position of the host vehicle, and the heading of the host vehicle.

[0135] Environmental information may include: type of traffic lights on the sidewalk, width of the sidewalk, angle between the pedestrian's movement direction and the sidewalk, lateral offset between the pedestrian and the sidewalk, absolute speed of the pedestrian (in both x and y directions), movement direction of the pedestrian, relative speed of the pedestrian and the host vehicle, relative distance between the pedestrian and the host vehicle, relative movement direction of the pedestrian and the host vehicle, relative distance between the pedestrian and the nearest five vehicles around, relative speed between the pedestrian and the nearest five vehicles around, relative movement direction between the pedestrian and the nearest five vehicles around, etc. 39-dimensional information.

[0136] The control instructions output by the strategy model may include: 0: LEFT (turn left), 1: IDLE (maintain current state), 2: RIGHT (turn right), 3: FASTER (accelerate), 4: SLOWER (decelerate).

[0137] The countermeasure effects to be achieved by the above control instructions include:

[0138] 1. When there is no traffic light and the direction of movement of pedestrians and the main vehicle intersects, pedestrians need to seize the intersection point and stay at the intersection.

[0139] 2. When there is a traffic light, pedestrians can run the red light, suddenly accelerate to run in front of the main vehicle and the main vehicle is blocked by background vehicles, pedestrians cross the road, and pedestrians poke their heads out.

[0140] 3. Ramp Scenario

[0141] The embodiment of the present application does not limit the ramp scene. For example, the ramp scene can be a basic ramp or an overpass ramp.

[0142] In the ramp scenario, the vehicle information of the host vehicle input into the strategy model may include: the speed of the host vehicle, the position of the host vehicle, and the heading of the host vehicle.

[0143] Environmental information may include: whether lane changes can be made left or right, whether the vehicle is on a ramp, the curvature of the ramp, the distance between the antagonistic target and the ramp entrance when going up the ramp, the distance between the antagonistic target and the ramp end when going down the ramp, the width between the lane where the host vehicle is located and the target lane, the angle between the heading of the antagonistic target and the lane where the antagonistic target is located, the angle between the heading of the antagonistic target and the target lane, the lateral offset between the center of the antagonistic target and the lane where the antagonistic target is located, the lateral offset between the center of the antagonistic target and the target lane, the absolute speed of the antagonistic target (in both x and y directions), the heading angle of the antagonistic target, the relative speed between the antagonistic target and the host vehicle, the relative distance between the antagonistic target and the host vehicle, the relative heading between the antagonistic target and the host vehicle, the relative distance between the antagonistic target and the five nearest vehicles around it, the relative speed between the antagonistic target and the five nearest vehicles around it, the relative heading between the antagonistic target and the five nearest vehicles around it, and other 45-dimensional information.

[0144] The control instructions output by the strategy model may include: 0: LEFT (turn left), 1: IDLE (maintain current state), 2: RIGHT (turn right), 3: FASTER (accelerate), 4: SLOWER (decelerate).

[0145] The countermeasure effects to be achieved by the above control instructions include:

[0146] 1. If the host vehicle is on or off the ramp and the adversary is in the host vehicle's target lane, the adversary needs to drive side by side with the host vehicle to force it to fail to go on or off the ramp.

[0147] 2. If the host vehicle is on the side of the opposing target, if the host vehicle is going straight, the opposing target needs to accelerate to cut into the lane where the host vehicle is located. If the host vehicle changes lanes to the lane where the opposing target is located, the opposing target needs to accelerate to occupy the lane and force the host vehicle to fail to change lanes.

[0148] 3. On the ramp, if the host vehicle is behind the opposing vehicle, the opposing vehicle needs to brake. If the host vehicle changes lanes to the target lane, the opposing vehicle also needs to change lanes to the target lane.

[0149] 4. On the ramp, if the host vehicle is in front of the antagonist, the antagonist needs to follow closely so that the distance between the antagonist and the host vehicle is less than the first distance threshold.

[0150] 4. Roundabout Scene

[0151] In the roundabout scenario, the vehicle information of the host vehicle input into the strategy model may include: the speed of the host vehicle, the position of the host vehicle, and the heading of the host vehicle.

[0152] Environmental information may include: whether it is possible to change lanes left or right, whether it is on a roundabout, the curvature of the roundabout, the distance between the antagonistic target and the entrance of the roundabout when entering the roundabout, the distance between the antagonistic target and the exit of the roundabout when exiting the roundabout, the width between the lane where the host vehicle is located and the target lane, the angle between the heading of the antagonistic target and the lane where the antagonistic target is located, the angle between the heading of the antagonistic target and the target lane, the lateral offset between the center of the antagonistic target and the lane where the antagonistic target is located, the lateral offset between the center of the antagonistic target and the target lane, the absolute speed of the antagonistic target (in both x and y directions), the heading angle of the antagonistic target, the relative speed between the antagonistic target and the host vehicle, the relative distance between the antagonistic target and the host vehicle, the relative heading between the antagonistic target and the host vehicle, the relative distance between the antagonistic target and the five nearest vehicles around it, the relative speed between the antagonistic target and the five nearest vehicles around it, the relative heading between the antagonistic target and the five nearest vehicles around it, and other 45-dimensional information.

[0153] The control instructions output by the strategy model may include: 0: LEFT (turn left), 1: IDLE (maintain current state), 2: RIGHT (turn right), 3: FASTER (accelerate), 4: SLOWER (decelerate).

[0154] The countermeasure effects to be achieved by the above control instructions include:

[0155] 1. When entering or exiting a roundabout, if there is an intersection between the opposing target and the main vehicle, the opposing target needs to seize the intersection, brake at the intersection, or confront the main vehicle in the direction of the intersection.

[0156] 2. When exiting a roundabout, the main vehicle is in the middle lane of the roundabout, and the opposing target is in the edge lane of the roundabout. The opposing target needs to drive side by side with the main vehicle to force it to fail to exit the roundabout.

[0157] 3. In a roundabout, if the main vehicle is on the side of the opposing target, if the main vehicle is going straight, the opposing target needs to accelerate and cut into the lane where the main vehicle is located. If the main vehicle changes lanes to the lane where the opposing target is located, the opposing target needs to accelerate to occupy the lane and force the main vehicle to fail to change lanes.

[0158] 4. In a roundabout, if the main vehicle is behind the opposing vehicle, the opposing vehicle needs to brake. If the main vehicle changes lanes to the target lane, the opposing vehicle also needs to change lanes to the target lane.

[0159] 5. In a roundabout, if the host vehicle is in front of the antagonist, the antagonist needs to follow closely so that the distance between the antagonist and the host vehicle is less than the first distance threshold.

[0160] 5. Lane Addition and Reduction Scenarios

[0161] In the lane addition and subtraction scenario, the vehicle information of the host vehicle input into the strategy model may include: the speed of the host vehicle, the position of the host vehicle, and the heading of the host vehicle.

[0162] Environmental information may include: whether it is possible to change lanes left or right, whether there is a lane reduction phenomenon in the current road section, whether there is a lane increase phenomenon in the current road section, whether the current lane, the left lane, and the right lane have lane reductions or increases, the distance between the confrontation target and the end of the lane reduction or the starting point of the lane increase, the width between the lane where the host vehicle is located and the target lane, the angle between the heading of the confrontation target and the lane where the confrontation target is located, the angle between the heading of the confrontation target and the target lane, the lateral offset between the center of the confrontation target and the lane where the confrontation target is located, the lateral offset between the center of the confrontation target and the target lane, the absolute speed of the confrontation target (in both x and y directions), the heading angle of the confrontation target, the relative speed between the confrontation target and the host vehicle, the relative distance between the confrontation target and the host vehicle, the relative heading between the confrontation target and the host vehicle, the relative distance between the confrontation target and the nearest 5 vehicles around it, the relative speed between the confrontation target and the nearest 5 vehicles around it, the relative heading between the confrontation target and the nearest 5 vehicles around it, and other 47-dimensional information.

[0163] The control instructions output by the strategy model may include: 0: LEFT (turn left), 1: IDLE (maintain current state), 2: RIGHT (turn right), 3: FASTER (accelerate), 4: SLOWER (decelerate).

[0164] The countermeasure effects to be achieved by the above control instructions include:

[0165] 1. If the lane ahead of the main vehicle is reduced and the opposing target is on the side of the main vehicle, the opposing target needs to drive side by side with the main vehicle to force it to fail to change lanes.

[0166] 2. If the host vehicle needs to change lanes to an additional lane, and the adversarial target is in the host vehicle's target lane, the adversarial target needs to drive side by side with the host vehicle to force it to fail in changing lanes.

[0167] 3. If the host vehicle is on the side of the opposing target, if the host vehicle is going straight, the opposing target needs to accelerate to cut into the host vehicle's lane. If the host vehicle changes lanes to the opposing target's lane, the opposing target needs to accelerate to occupy the lane and force the host vehicle to fail to change lanes.

[0168] 4. If the host vehicle is behind the opposing vehicle, the opposing vehicle needs to brake. If the host vehicle changes lanes to the target lane, the opposing vehicle also needs to change lanes to the target lane.

[0169] 5. If the host vehicle is in front of the antagonist, the antagonist needs to follow the vehicle closely so that the distance between the antagonist and the host vehicle is less than the first distance threshold.

[0170] After the strategy model outputs the control instruction to the control model in the above manner, if the control instruction is IDLE, the control model can output a control signal: the speed and heading angle of the confrontation target remain unchanged.

[0171] If the control instruction is FASTER or SLOWER, the control model can output a control signal: the heading angle of the confrontation target remains unchanged, and the speed increases or decreases uniformly.

[0172] If the control instruction is LEFT or RIGHT, the control model can first determine the local coordinates of the antagonist target on the target lane (such as the Frenet coordinate system), and then determine the next simulated longitudinal position of the antagonist target based on the local coordinates of the antagonist target on the target lane, the speed of the antagonist target and the simulation frequency, and obtain the heading angle of the next simulation, calculate the lateral speed according to the proportional controller and convert it into a heading angle, and finally calculate the heading angular rate, sideslip angle, and steering angle in sequence to generate continuous actions, and generate control signals based on the heading angle, heading angular rate, sideslip angle, and steering angle, and output the control signals to achieve control of the antagonist target.

[0173] In an optional embodiment, the control model may include a lateral control model and / or a longitudinal control model, the lateral control model is used to generate a lateral control signal to control the direction of the antagonistic target, and the longitudinal control model is used to generate a longitudinal control signal to control the speed of the antagonistic target. Accordingly, when the vehicle information and environmental information are input into the preset control model to obtain the control signal of the antagonistic target, the vehicle information and environmental information may be input into the lateral control model and / or the longitudinal control model respectively to obtain the lateral control signal and / or the longitudinal control signal of the antagonistic target.

[0174] Among them, the lateral control model can output lateral actions such as lane change and deviation to test the main vehicle's lane change, emergency avoidance, normal planned driving and other functions. The longitudinal control model can output longitudinal acceleration and deceleration actions to test the main vehicle's lane change, overtaking, collision prediction and obstacle avoidance and other functions.

[0175] In an optional implementation, before using the lateral control model and the longitudinal control model, the lateral control model and the longitudinal control model may be trained. According to the characteristics of different scenarios, the lateral control model and the longitudinal control model may be trained separately, so that the confrontation actions can be designed more specifically, the training difficulty can be reduced, and the confrontation efficiency can be improved.

[0176] Specifically, the first control model may be trained based on the lateral reward function corresponding to the target scene to obtain a lateral control model; and / or the second control model may be trained based on the longitudinal reward function corresponding to the target scene to obtain a longitudinal control model. Different scenes correspond to different longitudinal reward functions and lateral reward functions.

[0177] By flexibly configuring the lateral control model and the longitudinal control model, more adversarial scenarios can be generalized, and the adaptability of the autonomous driving control algorithm in adversarial scenarios can be efficiently tested, thus solving the problems of high cost, poor generalization and lack of interaction of traditional methods.

[0178] During the training process, different reward functions can be designed for the lateral control model and the longitudinal control model in different scenarios to optimize the lateral control model and the longitudinal control model to improve the accuracy of controlling the adversarial target, thereby improving the driving performance of the autonomous vehicle.

[0179] Specifically, when the target scene is an intersection scene or a roundabout scene, the control model may include a lateral control model, and the lateral reward function corresponding to the target scene is related to the time when the main vehicle and the confrontation target arrive at the intersection area.

[0180] When the target scenario is a ramp scenario or a lane addition and subtraction scenario, the control model may include a lateral control model and a longitudinal control model, and the lateral reward function corresponding to the target scenario is related to the lateral distance between the host vehicle and the adversarial target. The longitudinal reward function corresponding to the target scenario is related to the longitudinal distance between the host vehicle and the adversarial target.

[0181] Among them, the lateral distance refers to the distance between the main vehicle and the opposing target in the direction perpendicular to the lane line, and the longitudinal distance refers to the distance between the main vehicle and the opposing target in the direction parallel to the lane line.

[0182] The following will use different scenarios as examples to explain the reward functions of the lateral control model and the longitudinal control model.

[0183] In the target scene Figure 4In the scenario shown in the figure, where the host vehicle and the antagonist target have an intersection (such as a roundabout or intersection), the antagonist target does not need to change lanes or deviate from the lane. Therefore, the reward function in this scenario is the reward function of the longitudinal control model, and this reward function is mainly related to the time when the host vehicle and the antagonist target reach the intersection, that is, the smaller the time difference between the host vehicle and the antagonist target when they reach the intersection, the greater the reward. Therefore, the reward function of the longitudinal control model in this scenario can be set as:

[0184]

[0185] Among them, t1 represents the time when the main vehicle arrives at the intersection, and t2 represents the time when the opponent target arrives at the intersection. t1 can be determined according to the speed of the main vehicle and the distance from the main vehicle to the intersection, and t2 can be determined according to the speed of the opponent target and the distance from the opponent target to the intersection.

[0186] exist Figure 4 In the target scenario shown, the input of the longitudinal control model may include the speed and orientation angle of the antagonistic target, the time and distance of the antagonistic target to reach the intersection, the relative position and relative speed of the host vehicle and the antagonistic target, the orientation angle of the host vehicle, the time and distance of the host vehicle to reach the intersection, and the relative distance and relative speed between the antagonistic target and its surrounding vehicles, etc. The output of the longitudinal control model is acceleration, deceleration or maintaining the current state.

[0187] In the target scene Figure 5 The scenario shown in the figure is that the host vehicle needs to actively change lanes to cut into the lane where the antagonist vehicle target is located (such as ramps, lane addition and subtraction scenarios, and lane changes in advance for left and right turns). The antagonist target does not need to change lanes or deviate from the lane. Therefore, the reward function in this scenario is the reward function of the longitudinal control model, and the reward function is mainly related to the longitudinal distance between the host vehicle and the antagonist vehicle target along the road direction, such as Figure 5 As shown in the figure, the smaller the longitudinal distance X between the main vehicle and the opponent vehicle along the road direction, the greater the reward. Therefore, the expression of the reward function can be set as:

[0188]

[0189] exist Figure 5 In the target scenario shown, the input of the longitudinal control model may include the speed and orientation angle of the antagonistic target, the relative position, relative speed, and orientation angle of the host vehicle and the antagonistic target, the longitudinal distance between the host vehicle and the antagonistic target along the road direction, and the relative distance and relative speed between the antagonistic target and its surrounding vehicles, etc. The output of the longitudinal control model is acceleration, deceleration, or maintaining the current state.

[0190] In the target scene Figure 6In the scenario shown in FIG. 1 , the antagonist target needs to actively change lanes and cut into the lane where the host vehicle is located. The antagonist target needs to change lanes or lane deviation. Therefore, the reward function in this scenario includes the reward function of the lateral control model, and the reward function is mainly related to the lateral distance between the host vehicle and the antagonist target, such as Figure 6 As shown in Figure 2, the smaller the lateral distance L between the main vehicle and the adversarial vehicle target, the greater the reward. Therefore, the expression of the reward function can be set as:

[0191]

[0192] exist Figure 6 In the target scenario shown, the input of the longitudinal control model may include the speed and orientation angle of the antagonistic target, the relative position, relative speed, and orientation angle of the host vehicle and the antagonistic target, the lateral distance between the host vehicle and the antagonistic target along the road direction, and the relative distance and relative speed between the antagonistic target and its surrounding vehicles, etc. The output of the lateral control model is change lane to the left, shift to the left, hold, change lane to the right, or shift to the right.

[0193] Since the interaction between the antagonistic target and the host vehicle is different under different road structures, in order to ensure effective confrontation in different scenarios, it is necessary to design different antagonistic target generation strategies according to different scenarios to ensure the effectiveness of the interaction between the host vehicle and the antagonistic target, so as to achieve the purpose of testing one or more logical functions of the host vehicle (such as lane changing, overtaking, obstacle avoidance, etc.), thereby ensuring the effectiveness of the confrontation behavior.

[0194] Therefore, in the embodiment of the present application, when executing Figure 1 Before the vehicle confrontation method shown, at least one confrontation target may be generated in the simulation platform based on the scene type of the target scene.

[0195] In the embodiments of the present application, there is no limitation on the scene type of the target scene. For example, the target scene may be a non-intersection scene, and the non-intersection scene indicates that the current lane is a one-way lane, such as a highway, a ramp, a roundabout, and other scenarios. The target scene may be an intersection scene, and the intersection scene indicates that the current lane is a multi-way lane, such as an intersection with traffic lights, a T-junction, entering and exiting a roundabout, basic on- and off-ramps, on- and off-ramps of an overpass, lane addition and subtraction, and other scenarios.

[0196] Specifically, when the cloud server generates at least one antagonistic target in the target scene, it can first determine the target lane of the main vehicle, and then generate at least one antagonistic target in the opposite lane of the lane in which the main vehicle is currently traveling, the lane adjacent to the lane in which the main vehicle is currently traveling, and / or the lane in which the main vehicle is currently traveling, based on the target lane of the main vehicle, and the target lane of the antagonistic target is the same as the target lane of the main vehicle.

[0197] The target lane of the main vehicle refers to the lane where the main vehicle intends to travel.

[0198] For example, in one embodiment, assuming that the target scene is Figure 7 In the non-intersection scenario shown, vehicle A is the main vehicle, and the lane currently traveled by vehicle A is the second lane. If the target lane of vehicle A is the second lane, the cloud server can generate at least one confrontation target in the lane currently traveled by the main vehicle and / or the lane adjacent to the lane currently traveled by the main vehicle, such as Figure 7 As shown in the figure, the target lanes of vehicles C, E, B and D are also the second lane.

[0199] Through the above method, after at least one confrontation target is generated in the lane where the main vehicle is currently traveling and in the lane adjacent to the lane where the main vehicle is currently traveling, the confrontation target can be subsequently controlled so that the confrontation target gradually approaches the main vehicle and engages in confrontation with the main vehicle.

[0200] In another embodiment, assuming the target scene is Figure 8 In the T-junction intersection scenario shown, vehicle A is the main vehicle, the lane currently traveled by vehicle A is lane 1, and the target lane of vehicle A is lane 3. The cloud server can generate at least one adversarial target whose target lane is also lane 3, such as Figure 8 The vehicle B shown is configured so that the vehicle B can be controlled later and can engage in confrontational behavior with the vehicle A in lane 3.

[0201] In another embodiment, assuming that the target scene is Fig. 9 In the lane addition intersection scenario shown in the figure, vehicle A is the main vehicle and the driving trajectory of vehicle A is as follows: Fig. 9 As shown by the dotted line in , it can be seen that the lane currently traveled by vehicle A is the second lane, and the target lane of vehicle A is the third lane (i.e., the added lane). The cloud server can generate at least one adversarial target whose target lane is also the third lane, such as Fig. 9 The vehicle B shown is configured so that the vehicle B can be controlled later and can engage in confrontational behavior with the vehicle A in the third lane.

[0202] In another embodiment, assuming that the target scene is Fig.10 In the lane reduction intersection scenario shown in FIG. 1 , vehicle A is the main vehicle, and the driving trajectory of vehicle A is as follows: Fig.10 As shown by the dotted line in , it can be seen that the lane currently traveled by vehicle A is the third lane (i.e., the reduced lane), and the target lane of vehicle A is the second lane. The cloud server can generate at least one adversarial target whose target lane is also the second lane, such as Fig.10The vehicle B shown is configured so that the vehicle B can be subsequently controlled so that the vehicle B can engage in confrontational behavior with the vehicle A in the second lane.

[0203] In an embodiment of the present application, the confrontation target needs to be placed or generated outside a certain range (the detection range is calculated based on the speed of the main vehicle) to prevent the target vehicle from suddenly appearing around the main vehicle and to ensure that the main vehicle can smoothly transition to the confrontation scene, thereby triggering effective confrontation behavior.

[0204] In the embodiment of the present application, the time from the start of the simulation task to the occurrence of the confrontation behavior does not exceed 120 seconds.

[0205] In the embodiment of the present application, the timing of the arrival of the confrontation target at the intersection point or intersection area should comply with the advance principle, that is, try to appear at a position where the road conditions meet the lateral movement conditions of the main vehicle.

[0206] Before executing the vehicle confrontation method provided in the embodiment of the present application, the Highway-env simulation technology can be used to simulate and reproduce the natural traffic flow environment in the simulation platform according to the real driving trajectory data set to obtain the target scenario described above, and the artificial intelligence-based driver model can be trained according to the reproduced natural traffic flow environment to achieve modeling of natural human driving behavior. Afterwards, reinforcement learning technology can be used to generate confrontation targets in the natural traffic flow environment.

[0207] After training and generating adversarial targets, different thresholds can be set for the adversarial spatial range and adversarial duration of the adversarial targets, thereby realizing adversarial testing of the main vehicle in continuous time and space. This can not only improve the efficiency of test scene generation, but also ensure the probability of the scene occurring in the real world.

[0208] In the embodiment of the present application, when generating control information for the adversarial object or the main vehicle, in order to avoid the low dimension of the input information, a lightweight network structure can be designed using a fully connected layer and a ReLU activation function to avoid overfitting of the model training and accelerate network training. The network structure of the PPO and GAIL algorithm models is shown in Figure 1. Fig.11 As shown in the figure, the PPO algorithm model consists of an Actor network and a Critic network, and the GAIL algorithm model consists of a PPO algorithm network and a discriminator network.

[0209] For the Actor network, its input is the state S t , first through the input layer, the number of neurons in the input layer is equal to S t The dimension is then passed through two hidden layers with 128 neurons each, and further decomposed into two head networks to generate the mean μ and variance σ of the Gaussian distribution. 2 , the final output layer samples the Gaussian distribution and outputs action a t.

[0210] For the Critic network, its input is the state S t The input layer is the same as the hidden layer and the Actor network, and the output layer outputs the current state S t The value of

[0211] The discriminator network is similar to the Critic network structure, except that its input is an action-state pair, i.e., S t with a t The output is the classification probability D of the action state pair δ (S t ,a t ).

[0212] In the embodiment of the present application, the driving trajectory of the main vehicle or the adversary object can also be generated based on the network structure of the conditional diffusion behavior cloning algorithm. Fig.12 As shown, during training, a sample X0(S t ,a t ), and randomly select a time i from 1 to T, and then t and i are passed to Gaussian diffusion, which is the forward diffusion process, sampling random noise and adding it to a t On, form During training, i needs to be used as the input of the time encoding network and output T-dimensional time encoding t_emb i , then, the T-dimensional time encoding t_emb i and S t , Concatenate as input to a fully connected network to generate actions

[0213] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to achieve the above functions, the vehicle countermeasure device or electronic device includes a hardware structure and / or software module corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0214] The embodiments of the present application can exemplarily divide the functional modules of the vehicle confrontation device or electronic device according to the above method. For example, the vehicle confrontation device or electronic device can include various functional modules corresponding to the various functional divisions, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiments of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0215] Fig.13 FIG. 1 is a block diagram of a vehicle confrontation device 1300 according to an exemplary embodiment. Fig.13 , the vehicle confrontation device 1300 includes:

[0216] The generation unit 1301 is used to input the vehicle information and environmental information of the host vehicle in the target scene into a preset control model to obtain a control signal of the antagonistic target; the antagonistic target and the host vehicle are virtual objects pre-generated in the simulation platform; the host vehicle corresponds to different vehicle information and environmental information in different scenes; the control signal is used to control the direction and / or speed of the antagonistic target.

[0217] The confrontation unit 1302 is used to control the confrontation target in the simulation platform based on the control signal, so that the confrontation target performs confrontation behavior with the host vehicle in the target scene.

[0218] In a possible implementation, the control signal includes a lateral control signal and / or a longitudinal control signal, the lateral control signal is used to control the direction of the confrontation target, and the longitudinal control signal is used to control the speed of the confrontation target.

[0219] In one possible implementation, the control model includes a lateral control model and / or a longitudinal control model, the lateral control model is used to generate a lateral control signal to control the direction of the confrontation target, and the longitudinal control model is used to generate a longitudinal control signal to control the speed of the confrontation target.

[0220] In a possible implementation, the device also includes a training unit, which is specifically used to: train the first control model based on the lateral reward function corresponding to the target scene to obtain a lateral control model; different scenes correspond to different lateral reward functions; and / or, train the second control model based on the longitudinal reward function corresponding to the target scene to obtain a longitudinal control model; different scenes correspond to different longitudinal reward functions.

[0221] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the control model includes a lateral control model; the lateral reward function corresponding to the target scene is related to the time when the main vehicle and the confrontation target arrive at the intersection area.

[0222] In a possible implementation, the target scene is an intersection scene or a roundabout scene, and the lateral reward function corresponding to the target scene is:

[0223]

[0224] Among them, t1 represents the time when the main vehicle arrives at the intersection, and t2 represents the time when the adversarial target arrives at the intersection.

[0225] In a possible implementation, the target scenario is a ramp scenario or a lane addition and reduction scenario, and the control model includes a lateral control model and a longitudinal control model; the lateral reward function corresponding to the target scenario is related to the lateral distance between the host vehicle and the adversarial target; the lateral distance refers to the distance between the host vehicle and the adversarial target in a direction perpendicular to the lane line; the longitudinal reward function corresponding to the target scenario is related to the longitudinal distance between the host vehicle and the adversarial target; the longitudinal distance refers to the distance between the host vehicle and the adversarial target in a direction parallel to the lane line.

[0226] In a possible implementation, the target scene is a ramp scene or a lane addition and subtraction scene, and the lateral reward function corresponding to the target scene is:

[0227]

[0228] Where L represents the distance between the host vehicle and the adversarial target in the direction perpendicular to the lane line;

[0229] The vertical reward function corresponding to the target scene is:

[0230]

[0231] Where X represents the distance between the host vehicle and the adversarial target in the direction parallel to the lane line.

[0232] In the vehicle confrontation device provided in the embodiment of the present application, the main vehicle can correspond to different vehicle information and environmental information in different scenarios. Therefore, when conducting confrontation tests on the main vehicle, the cloud server can control the confrontation target in different scenarios, so that the confrontation target will have confrontational behavior with the main vehicle in the simulation scenario, thereby realizing confrontation tests on the main vehicle. Therefore, the present application can realize confrontation tests on the main vehicle in different scenarios, and then comprehensively evaluate the performance of the autonomous driving car (i.e., the main vehicle) in a diverse, complex and unpredictable real traffic environment.

[0233] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0234] Fig.14 FIG. 1 is a block diagram of an electronic device according to an exemplary embodiment. Fig.14 As shown, the electronic device 1400 includes but is not limited to: a processor 1401 and a memory 1402 .

[0235] The memory 1402 is used to store executable instructions of the processor 1401. It can be understood that the processor 1401 is configured to execute instructions to implement the vehicle confrontation method in the above embodiment.

[0236] It should be noted that those skilled in the art can understand that Fig.14 The electronic device structure shown in the figure does not constitute a limitation on the electronic device, and the electronic device may include Fig.14 More or fewer components may be shown, or certain components may be combined, or the components may be arranged differently.

[0237] The processor 1401 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. By running or executing software programs and / or modules stored in the memory 1402, and calling data stored in the memory 1402, it performs various functions of the electronic device and processes data, thereby monitoring the electronic device as a whole. The processor 1401 may include one or more processing units. Optionally, the processor 1401 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1401.

[0238] The memory 1402 may be used to store software programs and various data. The memory 1402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one functional module (such as a determination unit, a processing unit, etc.), etc. In addition, the memory 1402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0239] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 1402 including instructions, and the above instructions can be executed by the processor 1401 of the electronic device 1400 to implement the method in the above embodiment.

[0240] In actual implementation, Fig.13 The generation unit 1301 and the confrontation unit 1302 in Fig.14 The processor 1401 in the embodiment calls the computer program stored in the memory 1402 to implement. The specific execution process can refer to the description of the method part in the above embodiment, which will not be repeated here.

[0241] Optionally, the computer-readable storage medium may be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0242] In an exemplary embodiment, the present application also provides a computer program product including one or more instructions, and the one or more instructions can be executed by the processor 1401 of the electronic device to complete the method in the above embodiment.

[0243] It should be noted that when the instructions in the above-mentioned computer-readable storage medium or one or more instructions in the computer program product are executed by the processor of the electronic device, the various processes of the above-mentioned method embodiment are implemented, and the same technical effect as the above-mentioned method can be achieved. To avoid repetition, they will not be repeated here.

[0244] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0245] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0246] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0247] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0248] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or the full classification part or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to execute the full classification part or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM, RAM, disk or CD and other media that can store program code.

[0249] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto, and any changes or substitutions within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A vehicle confrontation method, characterized in that: Applied to a cloud server, wherein a simulation platform is deployed in the cloud server, the method comprises: Inputting vehicle information and environmental information of the host vehicle in the target scene into a preset control model to obtain a control signal of the antagonistic target; the antagonistic target and the host vehicle are virtual objects pre-generated in the simulation platform; the host vehicle corresponds to different vehicle information and environmental information in different scenes; the control signal is used to control the direction and / or speed of the antagonistic target; Based on the control signal, the antagonistic target is controlled in the simulation platform so that the antagonistic target performs antagonistic behavior with the host vehicle in the target scene.

2. The method according to claim 1, characterized in that The control signal includes a lateral control signal and / or a longitudinal control signal; the lateral control signal is used to control the direction of the confrontation target; the longitudinal control signal is used to control the speed of the confrontation target.

3. The method according to claim 1 or 2, characterized in that: The control model includes a lateral control model and / or a longitudinal control model; the lateral control model is used to generate a lateral control signal to control the direction of the confrontation target; the longitudinal control model is used to generate a longitudinal control signal to control the speed of the confrontation target.

4. The method according to claim 3, characterized in that The method further comprises: The first control model is trained based on the lateral reward function corresponding to the target scene to obtain the lateral control model; different scenes correspond to different lateral reward functions; and / or, The second control model is trained based on the longitudinal reward function corresponding to the target scene to obtain the longitudinal control model; different scenes correspond to different longitudinal reward functions.

5. The method according to claim 4, characterized in that The target scene is an intersection scene or a roundabout scene, and the control model includes a lateral control model; the lateral reward function corresponding to the target scene is related to the time when the main vehicle and the confrontation target arrive at the intersection area.

6. The method according to claim 4 or 5, characterized in that: The target scene is an intersection scene or a roundabout scene, and the lateral reward function corresponding to the target scene is: Among them, t1 represents the time when the main vehicle arrives at the intersection, and t2 represents the time when the adversarial target arrives at the intersection.

7. The method according to claim 4, characterized in that The target scene is a ramp scene or a lane increase and decrease scene, and the control model includes a lateral control model and a longitudinal control model; the lateral reward function corresponding to the target scene is related to the lateral distance between the main vehicle and the confrontation target; the lateral distance refers to the distance between the main vehicle and the confrontation target in a direction perpendicular to the lane line; the longitudinal reward function corresponding to the target scene is related to the longitudinal distance between the main vehicle and the confrontation target; the longitudinal distance refers to the distance between the main vehicle and the confrontation target in a direction parallel to the lane line.

8. The method according to claim 7, characterized in that The target scene is a ramp scene or a lane increase or decrease scene, and the lateral reward function corresponding to the target scene is: Wherein, L represents the distance between the host vehicle and the confrontation target in the direction perpendicular to the lane line; The vertical reward function corresponding to the target scene is: Wherein, X represents the distance between the host vehicle and the confrontation target in the direction parallel to the lane line.

9. A vehicle countermeasure device, characterized in that: The device comprises: a generating unit and a confrontation unit, wherein: The generating unit is used to input the vehicle information and environmental information of the host vehicle in the target scene into a preset control model to obtain a control signal of the antagonistic target; the antagonistic target and the host vehicle are virtual objects pre-generated in the simulation platform; the host vehicle corresponds to different vehicle information and environmental information in different scenes; the control signal is used to control the direction and / or speed of the antagonistic target; The confrontation unit is used to control the confrontation target in the simulation platform based on the control signal, so that the confrontation target performs confrontation behavior with the main vehicle in the target scene.

10. A server, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the vehicle countermeasure method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: When the computer-executable instructions stored in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device executes the vehicle countermeasure method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product includes computer instructions; when part or all of the computer instructions are run on a computer, the computer executes the vehicle countermeasure method according to any one of claims 1 to 8.