An automatic driving decision dangerous scene generation method, system, device and medium

Abstract

The application relates to an automatic driving decision dangerous scene generation method, system, device and medium, which comprises the following steps: extracting static element information and dynamic element information of a scene when a driver takes over an automatic driving vehicle in real vehicle road testing; establishing a dynamic element behavior model according to the extracted static element information and dynamic element information; setting a parameter space of the dynamic element behavior model, determining a parameter combination of the dynamic element behavior model in each simulation test, and constructing a dangerous scene; clustering and fitting the parameter combination of the constructed dangerous scene, and generating a dangerous scene region under the parameter combination; and the application can realize efficient generation of a dangerous scene and can be widely applied in the field of intelligent vehicles.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicles, and in particular to a method, system, device, and medium for generating hazardous scenarios for autonomous driving decision-making. Background Technology

[0002] In recent years, the rapid development of intelligent vehicles has attracted increasing attention, with high levels of autonomous driving requiring no passenger intervention. As autonomous driving systems become more complex, their safety has become a primary concern. Decision-making, as a crucial component of autonomous driving systems, necessitates effective testing before development and deployment.

[0003] In general, testing methods for autonomous driving include real-vehicle road testing and simulation testing. The former is more representative, but it is severely limited by safety and cost issues; the latter is lower in cost and does not have safety issues, therefore it is also an essential process before real-vehicle testing. Utilizing simulation testing to evaluate the performance of autonomous driving decision-making algorithms and identify their failure points is of great significance. The core of simulation testing is building a simulation testing environment. For the decision-making algorithm development stage, the aim is to use the simulation testing environment to identify failure scenarios, also known as hazardous scenarios, of the decision-making algorithm, thereby laying the foundation for algorithm optimization and improvement.

[0004] There are generally two methods for generating hazardous scenarios. The first is to design adversarial behavior models of traffic participants in the environment without data, thereby constructing hazardous scenarios. However, due to the lack of real-world data, the scenario design and initial state lack interpretability. The second method utilizes massive amounts of driving data, searching for and constructing hazardous scenarios by changing the sampling of the data. This method can efficiently generate hazardous scenarios, but the massive amounts of data are difficult to obtain in real-world road testing. Therefore, efficiently and reasonably constructing hazardous scenarios for decision-making testing in simulation environments still requires further exploration and research. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a method, system, device, and medium for generating hazardous scenarios for autonomous driving decision-making, which can efficiently and reasonably construct hazardous scenarios for decision-making testing in a simulation environment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, it provides a method for generating hazardous scenarios for autonomous driving decision-making, comprising:

[0007] Extract static and dynamic element information of the scene when the driver takes over in an autonomous vehicle during real-world road testing;

[0008] Based on the extracted static and dynamic element information, a dynamic element behavior model is established.

[0009] Define the parameter space of the dynamic element behavior model, determine the parameter combination of the dynamic element behavior model in each simulation test, and construct the dangerous scenario;

[0010] Clustering and fitting are performed on the parameter combinations of the constructed hazardous scenarios to generate hazardous scenario regions under that parameter combination.

[0011] Furthermore, the scenarios in which the autonomous vehicle takes over during the real-vehicle road test include the real-time position and speed of the autonomous vehicle and the surrounding vehicles, as well as the type of the surrounding vehicles and the position of the lane lines.

[0012] Furthermore, the step of establishing a dynamic element behavior model based on the extracted static and dynamic element information includes:

[0013] Based on the extracted static and dynamic element information, an initial state space for the scene is established.

[0014] Based on the initial state space of the scene, establish a dynamic element behavior model for each vehicle in the scene.

[0015] Furthermore, the initial state space S0 of the scenario is:

[0016] S0={q e ,q1,q2,...,q n ,l}

[0017] Where, q e Indicates the physical state of an autonomous vehicle; q i (i = 1, ..., n) represents the physical state of other dynamic elements in the environment besides the autonomous vehicle; l represents road structure information.

[0018] Furthermore, the physical state of each dynamic element is q = {x, y, v}. x ,v y}, where x and y represent the positions of the dynamic elements, respectively, and v x ,v y These represent the velocities of the dynamic element in two directions, respectively.

[0019] The road structure in static elements is represented as l={g,w,c,(r1,r2,...,r n )}, where g, w, and c represent the length, width, and curvature of the road, respectively, (r1, r2, ..., r n () represents a series of reference points for the lane.

[0020] Furthermore, the step of setting the parameter space of the dynamic element behavior model, determining the parameter combination of the dynamic element behavior model in each simulation test, and constructing a hazardous scenario includes:

[0021] Define the parameter space for the dynamic element behavior model;

[0022] Discretize the set parameter space with a certain step size;

[0023] An optimization search method is used to determine the parameter combination of the dynamic element behavior model in each simulation test based on the discretized parameter space, and to construct dangerous scenarios and record the test results.

[0024] Furthermore, the method employs an optimization search approach to determine the parameter combinations of the dynamic element behavior model in each simulation test based on the discretized parameter space, and constructs hazardous scenarios and records test results, including:

[0025] Using the Metropolis criterion, a set of parameter combinations is selected from the discretized parameter space, and a new parameter combination is determined within the range of adjacent parameters with a certain search radius;

[0026] Check whether the determined new parameter combination is the same as the parameters that have been tested. If they are the same, select them randomly again.

[0027] Based on the combination of parameters detected, a hazardous scenario is constructed, and the test results are recorded.

[0028] Secondly, an autonomous driving decision-making hazardous scenario generation system is provided, including:

[0029] The scene information extraction module is used to extract static and dynamic element information of the scene when the driver takes over the autonomous vehicle during real-vehicle road testing.

[0030] The model building module is used to build a dynamic element behavior model based on the extracted static and dynamic element information.

[0031] The parameter combination determination module is used to set the parameter space of the dynamic element behavior model, determine the parameter combination of the dynamic element behavior model in each simulation test, and construct the dangerous scenario.

[0032] The hazardous scene region determination module is used to cluster and fit the parameter combinations of the constructed hazardous scenes to generate hazardous scene regions under the parameter combinations.

[0033] Thirdly, a processing device is provided, including computer program instructions, wherein when the computer program instructions are executed by the processing device, they are used to implement the steps corresponding to the above-mentioned method for generating dangerous scenarios for autonomous driving decisions.

[0034] Fourthly, a computer-readable storage medium is provided, wherein computer program instructions are stored on the computer-readable storage medium, wherein the computer program instructions, when executed by a processor, are used to implement the steps corresponding to the above-described method for generating hazardous scenarios for autonomous driving decisions.

[0035] The present invention has the following advantages due to the adoption of the above technical solutions:

[0036] 1. This invention utilizes a small amount of road test takeover data to construct multiple hazardous scenarios in simulated testing. The construction of the simulation environment relies on real test data, and the static elements and initial state space in the simulation originate from real scenarios, enabling efficient generation of hazardous scenarios and thus accelerating the development and verification of decision-making algorithms.

[0037] 2. The dangerous scenarios constructed in the simulation by this invention are based on data collected from real vehicle road tests and are derived from the problem scenarios encountered in the road. The initial states of the static and dynamic elements in the simulation scenario are all derived from real takeover data, which is more reasonable and representative than artificially designed scenarios.

[0038] 3. When constructing hazardous scenarios, this invention employs an optimized search method to determine the parameter combinations of dynamic element behavior models in each simulation test. Compared with the method of traversing the parameter space, the optimized search method enhances the local exploration of behavior model parameters for hazardous scenarios while retaining random exploration. It can improve the efficiency of searching for behavior model parameters corresponding to hazardous scenarios in the entire parameter space, thereby accelerating the generation of decision-oriented hazardous scenarios.

[0039] In summary, this invention can be widely applied in the field of intelligent vehicles. Attached Figure Description

[0040] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0041] Figure 1 This is a schematic diagram of a method flow provided in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the framework of an optimized search method provided in an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the initial position state of a vehicle in a scenario provided by an embodiment of the present invention;

[0044] Figure 4 This is a comparative diagram of an optimized search method and a random search method when generating the same number of dangerous scenarios, according to an embodiment of the present invention.

[0045] Figure 5 This is a comparative diagram of an optimized search method and a random search method under the same number of trials provided in an embodiment of the present invention. Detailed Implementation

[0046] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0047] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0048] Although terms such as first, second, third, etc., may be used in this document to describe multiple elements, components, regions, layers, and / or segments, these elements, components, regions, layers, and / or segments should not be limited by these terms. These terms may be used only to distinguish one element, component, region, layer, or segment from another. Unless the context clearly indicates otherwise, terms such as "first," "second," and other numerical terms used herein do not imply order or sequence. Therefore, the first element, component, region, layer, or segment discussed below may be referred to as the second element, component, region, layer, or segment without departing from the teachings of the exemplary embodiments.

[0049] The autonomous driving decision-making hazard scenario generation method, system, device, and medium provided in this invention utilize the scenario of driver takeover in an autonomous vehicle during real-vehicle road testing. The scenario is reconstructed in a simulation test to determine the road structure and initial state space in the scenario. Then, an optimization search method is used to construct the hazard scenario. Finally, the scenario is derived and generated using the occasional takeover data encountered in real road testing to find as many hazard scenarios as possible in the state space and generate a hazard scenario region.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides a method for generating hazardous scenarios for autonomous driving decision-making, including the following steps:

[0052] 1) Extract static and dynamic element information of the scene when the driver takes over the autonomous vehicle during real-vehicle road testing.

[0053] 2) Based on the extracted static and dynamic element information, establish the initial state space of the scene and establish a dynamic element behavior model.

[0054] 3) Set the parameter space of the dynamic element behavior model, and use the optimization search method to determine the parameter combination of the dynamic element behavior model in each simulation test, and construct the dangerous scenario and record the test results.

[0055] 4) Cluster and fit the parameter combinations of the constructed hazardous scenarios to generate hazardous scenario regions under the parameter combinations.

[0056] The method of this invention actually requires the construction of a simulation environment throughout the entire process. Each step provides some information for the construction of the simulation environment. Step 1) provides the basic information required for simulation construction. Step 2) sets the initial state of the simulation environment in the simulation. Step 3) sets different parameters based on this, so as to search for dangerous scenarios more efficiently in the constructed simulation scene. Step 4) based on this, clustering and fitting are performed on the parameter combination determined in Step 3).

[0057] In step 1) above, static element information includes weather, road type, road structure and the location of static obstacles, while dynamic element information includes the location and speed of dynamic traffic participants (such as vehicles and pedestrians) at each moment.

[0058] In step 1) above, the scenario where the driver takes over the autonomous vehicle during real-vehicle road testing is real data. The occurrence of driver takeover indicates that such scenarios are difficult for autonomous vehicles to handle, i.e., dangerous scenarios.

[0059] Specifically, the real-time data includes the real-time position and speed of the autonomous vehicle and the environmental vehicle, as well as the type of environmental vehicle and lane position, etc. The physical state of each dynamic element is represented as q={x,y,v}. x ,v y}, where x and y represent the positions of the dynamic elements, respectively, and v x ,v y These represent the velocities in two directions of the dynamic element, respectively; the road structure in the static element is represented as l={g,w,c,(r1,r2,...,r n )}, where g, w, and c represent the length, width, and curvature of the road, respectively, (r1, r2, ..., r n () represents a series of reference points for the lane.

[0060] In step 2) above, if a high-precision map of the test area can be obtained, it can be directly imported for scene construction; if a high-precision map of the test area cannot be obtained, a road area similar to the real data (such as intersections, highway merging ramps, two-lane straight roads, etc.) can be selected from the maps and road structures provided by the existing simulation platform.

[0061] In step 2) above, based on the extracted static and dynamic element information, an initial state space for the scene is established, and a dynamic element behavior model is built. The specific process is as follows:

[0062] 2.1) Based on the extracted static and dynamic element information, establish the initial state space S0 of the scene:

[0063] S0={q e ,q1,q2,...,w n ,l} (1)

[0064] Where, q e This represents the physical state of the vehicle under test (i.e., the ego vehicle, an autonomous vehicle); q i (i = 1, ..., n) represents the physical state of other dynamic elements in the environment besides the autonomous vehicle, and l represents the road structure information; based on the location information of the dynamic elements and the road structure information, the state of each element at the beginning of the dangerous scenario can be determined.

[0065] 2.2) Based on the initial state space S0 of the scenario, establish a dynamic element behavior model, which includes the dynamic element behavior models of various environmental vehicles in the dangerous scenario.

[0066] Specifically, given a fixed initial state, the generation of a dynamic scene depends on a dynamic element behavior model in the environment, which can be a learning-based behavior model or a predefined typical behavior model.

[0067] Specifically, taking the following scenario as an example, the dynamic element behavior model of the vehicle in front can be defined as: driving at an initial speed v0, decelerating to the brake with a certain acceleration a when the distance to the vehicle reaches d, and then accelerating to the final driving speed v with the same acceleration a. It can also be defined as other behavior models.

[0068] In step 3) above, the parameter space of the dynamic element behavior model is set, and an optimization search method is used to determine the parameter combination of the dynamic element behavior model in each simulation test, and a dangerous scenario is constructed and the test results are recorded. The specific process is as follows:

[0069] 3.1) Define the parameter space for the dynamic element behavior model.

[0070] Specifically, the parameters of different types of dynamic element behavior models are not entirely the same and can be set according to the actual situation. For example, the parameter space of the pedestrian dynamic element behavior model can be defined as {v ped ,Δx,v veh} indicates that the pedestrian observed the vehicle's speed to be v. veh When the distance between itself and the vehicle is Δx, with v ped The vehicle crosses the road at a constant speed. The parameter space of the vehicle's dynamic behavior model varies depending on the vehicle's behavior. For example, for a vehicle traveling only in a single-direction, single-lane road, the parameter space of its dynamic behavior model can be defined as {d, a}, where d is the distance from the vehicle and a is the acceleration. For a vehicle traveling in a single-direction, two-lane road, which can perform lane-changing and overtaking behaviors, the parameter space of the dynamic behavior model for lane-changing and overtaking can be defined as {Δx, Δy, v}. veh ,v ego Let ,a} represent the distances Δx and Δy between the vehicle and the target vehicle in the same lane, with the vehicle's speed and the target vehicle's speed being p respectively. peh ,v ego When overtaking, the acceleration is a; in this case, a one-way single lane is set, so the dynamic element behavior model can be represented by {v0,d,a,v}, where p0 is the initial speed and v is the final driving speed.

[0071] Specifically, after establishing the dynamic element behavior model in step 2), the parameters corresponding to the model are needed to determine the final dynamic element behavior model. All the parameters of the model and their value range constitute the parameter space.

[0072] Specifically, for example, in the dynamic element behavior model exemplified in step 2) above, there are four parameters: initial speed v0, distance d from the vehicle, acceleration a, and final driving speed v. These four parameters can be considered to constitute the parameter space: p = {p0, d, a, v}. The initial speed v0 can take a typical value or be obtained from the initial state of real data, so it can be considered a fixed parameter in the model. Similarly, other parameters can be selected as fixed values ​​or adjustable parameters according to relevant test requirements or typical behavior models. For example, the final driving speed v can also be considered a constant parameter. Then, the parameter space of the dynamic element behavior model becomes p = {d, a}, which is a two-dimensional parameter space.

[0073] 3.2) Discretize the set parameter space with a certain step size.

[0074] Specifically, the construction of the final simulation test environment requires selecting a set of defined parameters from the parameter space to generate dynamic hazardous scenarios. To facilitate sampling selection, the parameter space can be discretized with a certain step size, thereby finiteening the parameters in the parameter space.

[0075] Specifically, for example, for the parameter space p = {d, a}, the distance d can be taken to be in the range [5, 50] m, with 1 as the distance step size; the acceleration a can be taken to be in the range [0.5, 5] m / s². 2 The distance step length is 0.1. The specific parameter range and distance step length can be determined according to relevant requirements or historical experience.

[0076] 3.3) such as Figure 2 As shown, an optimization search method is used to determine the parameter combination of the dynamic element behavior model in each simulation test based on the discretized parameter space, and to construct dangerous scenarios and record the test results.

[0077] Specifically, to avoid getting trapped in local optima in this step, the parameters are selected each time from a random space or from the vicinity of a known dangerous scenario. The selection criterion is the Metropolis criterion, the core idea of ​​which is to accept the new state with a certain probability P, where probability P is:

[0078] P = exp(-k / b) (2)

[0079] Here, k represents the current number of trials, b represents an adjustable parameter related to the parameter space range, and b = n / m, where n is the total number of trials and m is an adjustable constant parameter. Therefore, in the early stages of the experiment, when the current number of trials k is small, the probability P is greater in terms of random exploration within the entire parameter space; while as the number of trials k increases, the probability of selecting parameters around dangerous scenarios also increases; that is, overall, the determination of parameters tends to shift from random exploration to searching for dangerous scenarios.

[0080] Therefore, in this step, the specific process for selecting parameters around the local hazardous scene is as follows:

[0081] 3.3.1) Using the Metropolis criterion, a set of parameter combinations is selected from the discretized parameter space, and a new parameter combination is determined within the range of adjacent parameters with a certain search radius r. After each experiment, the experimental results are recorded, and the dangerous parameters are classified into a category and recorded.

[0082] 3.3.2) During random exploration of the entire parameter space, check whether the determined new parameter combination is the same as the tested parameters. If it is the same, then randomly select again.

[0083] 3.3.3) Based on the combination of parameters after detection, construct a hazardous scenario and record the test results.

[0084] Specifically, the definition of a hazardous scenario can be formulated according to actual needs. For example, a hazardous scenario can be defined when the distance between the autonomous vehicle and other dynamic elements is less than a certain threshold, or when the time to collision (TTC) is greater than a certain threshold. Compared to the traversal search method, the parameter movement selection based on experimental results adopted in this embodiment can greatly improve the parameter search efficiency and identify hazardous scenarios for the autonomous driving algorithm more quickly.

[0085] It should be noted that the "surroundings" in the "surroundings of known dangerous scenarios" mentioned above refers to parameters whose parameter differences from the dangerous scenario are less than a certain threshold. The specific threshold is determined based on the specific behavioral model and its parameter range. Specifically, a search radius r can be defined, and the parameters surrounding the dangerous scenario are within the range of ±r from the parameters of the dangerous scenario.

[0086] In step 4) above, different fitting methods can be used for spaces of different dimensions. For example, for the two-dimensional parameter space in the example above, a Gaussian Mixture Model can be used. This model is common in clustering applications. It uses a combination of multiple Gaussian models to represent the distribution and boundaries of clusters. Ultimately, multiple ellipses can be drawn in the two-dimensional parameter space to represent dangerous scene areas.

[0087] By following the steps above, dangerous scene regions in the parameter space can be identified, dangerous scenes can be constructed in the simulation environment, thereby identifying problems in the autonomous driving algorithm and providing direction for its further optimization.

[0088] The method for generating hazardous scenarios for autonomous driving decision-making according to the present invention will be described in detail below through specific embodiments:

[0089] 1) Based on the scenarios where driver takeover occurs in autonomous vehicles during real-world road tests, extract information from the static and dynamic elements in the scenarios:

[0090] This example utilizes one of the takeover test data points from the Shougang Park area during the China Winter Olympics. A high-precision map of the Shougang Park was imported into the CARLA simulation platform. The road area where the takeover data occurred was identified as an unprotected intersection. Based on the recorded vehicle positions and speeds, the dynamic elements in the scene were determined to be three environmental vehicles, whose initial positions are as follows: Figure 3 As shown.

[0091] 2) Based on the extracted static and dynamic element information, establish the initial state space of the scene and build a dynamic element behavior model:

[0092] At this point, the state space has a higher dimension, including the positions and speeds of the four vehicles (three environmental vehicles and the autonomous vehicle itself). Let the autonomous vehicle be denoted as EgoVehicle, the preceding vehicle as Leading Vehicle, and the vehicles coming from the right and left as Right Vehicle and Left Vehicle, respectively. Since the autonomous vehicle does not only travel along straight roads, its position cannot be represented by a single-dimensional coordinate system. The other vehicles continue to travel along the current road. Therefore, the initial state space can be represented as follows:

[0093] S0={x ego ,y ego ,v ego ,x left ,v left ,x right ,v right ,x lead ,v lead}

[0094] When an autonomous vehicle makes a left turn, vehicles approaching from the left and right are the main sources of conflict. Therefore, this embodiment mainly focuses on searching and adjusting the behavioral model parameters for vehicles approaching from the left and right. The dynamic element behavioral model for vehicles approaching from the left and right is as follows: the vehicle on the left and the vehicle on the right start from a specific position with a speed of v. ;eft v right The vehicle travels at a constant speed and changes speed with acceleration a when it is d away from itself.

[0095] 3) Define the parameter space of the dynamic element behavior model, and use an optimization search method to determine the parameter combination of the dynamic element behavior model in each simulation test, and construct hazardous scenarios and record the test results:

[0096] The acceleration in the dynamic element behavior model is transformed into the control of the vehicle's throttle and brakes. The parameter space is set as follows: d∈[10,20]m, v∈[10,30]km / h, a∈[-0.7,0.3], where a is the acceleration, representing the control of the throttle and brakes. A positive a indicates acceleration, and a negative a indicates deceleration. Its magnitude simulates the percentage of the actual throttle and brake pedals being pressed. The distances of each parameter are set to 1m, 1km / h, and 0.1, respectively. The final parameter space is:

[0097] S p ={d left ,v left ,a left ,d right ,v right ,a right}

[0098] In this scenario, Distance To Collision (DTC) is used as the indicator to measure the degree of danger. A dangerous scenario threshold of 3.6m is set; that is, if the collision distance between the vehicle and two potentially colliding vehicles is less than 3.6m during a left turn, the scenario is considered dangerous. A parameter space for the dynamic element behavior model is defined, and an optimization search method is used to determine the parameter combinations of the dynamic element behavior model in each simulation test, constructing dangerous scenarios and recording the experimental results. Figure 4 and Figure 5 As shown, the optimized search found 286 hazardous scenarios in 2000 trials; the random search required 3376 trials to find the same number of hazardous scenarios. This means the optimized search method requires approximately 60% of the number of trials as the original method, and its speed in finding hazardous scenarios is about 1.7 times that of the random search. Furthermore, with the same number of trials (2000), the random search found 151 hazardous scenarios, while the optimized search found 286, approximately 1.9 times more. Therefore, the optimized search can accelerate the discovery and generation of hazardous scenarios compared to the random search.

[0099] 4) Cluster and fit the parameter combinations of the constructed hazardous scenarios to generate hazardous scenario regions under the parameter combinations.

[0100] Example 2

[0101] This embodiment provides an autonomous driving decision-making hazardous scenario generation system, including:

[0102] The scene information extraction module is used to extract static and dynamic element information of the scene when the driver takes over the autonomous vehicle during real-vehicle road testing.

[0103] The model building module is used to build a dynamic element behavior model based on the extracted static and dynamic element information.

[0104] The parameter combination determination module is used to set the parameter space of the dynamic element behavior model, determine the parameter combination of the dynamic element behavior model in each simulation test, and construct dangerous scenarios.

[0105] The hazardous scene region determination module is used to cluster and fit the parameter combinations of the constructed hazardous scenes to generate hazardous scene regions under the parameter combinations.

[0106] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0107] Example 3

[0108] This embodiment provides a processing device corresponding to the autonomous driving decision-making hazard scenario generation method provided in Embodiment 1. The processing device can be applied to client processing devices, such as mobile phones, laptops, tablets, desktop computers, etc., to execute the method of Embodiment 1.

[0109] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores computer programs that can run on the processing device. When the processing device runs the computer programs, it executes the autonomous driving decision-making hazard scenario generation method provided in Embodiment 1.

[0110] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0111] In other implementations, the processor can be any type of general-purpose processor, such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation here.

[0112] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] Those skilled in the art will understand that the structure of the above-described computing device is only a partial structure related to the solution of this application and does not constitute a limitation on the computing device to which the solution of this application is applied. A specific computing device may include more or fewer components, or combine certain components, or have different component arrangements.

[0114] Example 4

[0115] This embodiment provides a computer program product corresponding to the autonomous driving decision-making hazardous scenario generation method provided in Embodiment 1. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the autonomous driving decision-making hazardous scenario generation method described in Embodiment 1 are loaded.

[0116] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0117] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] The above embodiments are only used to illustrate the present invention. The structure, connection method and manufacturing process of each component can be varied. All equivalent transformations and improvements made on the basis of the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A method for generating hazardous scenarios for autonomous driving decision-making, characterized in that, include: Extract static and dynamic element information of the scene when the driver takes over in an autonomous vehicle during real-world road testing; Based on the extracted static and dynamic element information, a dynamic element behavior model is established. Define the parameter space of the dynamic element behavior model, determine the parameter combination of the dynamic element behavior model in each simulation test, and construct the dangerous scenario; Clustering and fitting are performed on the parameter combinations of the constructed hazardous scenarios to generate hazardous scenario regions under that parameter combination; The process of setting the parameter space of the dynamic element behavior model, determining the parameter combination of the dynamic element behavior model in each simulation test, and constructing hazardous scenarios includes: Define the parameter space for the dynamic element behavior model; Discretize the set parameter space with a certain step size; An optimized search method is used to determine the parameter combination of the dynamic element behavior model in each simulation test based on the discretized parameter space, and to construct dangerous scenarios and record the test results. The method employs an optimization search approach to determine the parameter combinations of the dynamic element behavior model in each simulation test based on the discretized parameter space, and constructs hazardous scenarios and records test results, including: Using the Metropolis criterion, a set of parameter combinations is selected from the discretized parameter space, and a new parameter combination is determined within the range of adjacent parameters with a certain search radius; Check whether the determined new parameter combination is the same as the parameters that have been tested. If they are the same, select them randomly again. Based on the combination of parameters detected, a hazardous scenario is constructed, and the test results are recorded.

2. The method for generating hazardous scenarios for autonomous driving decision-making as described in claim 1, characterized in that, The scenarios in which the autonomous vehicle takes over during the real-world road test include the real-time position and speed of the autonomous vehicle and the surrounding vehicles, as well as the type of the surrounding vehicles and their lane positions.

3. The method for generating hazardous scenarios for autonomous driving decision-making as described in claim 1, characterized in that, The step of establishing a dynamic element behavior model based on the extracted static and dynamic element information includes: Based on the extracted static and dynamic element information, an initial state space for the scene is established. Based on the initial state space of the scene, establish a dynamic element behavior model for each vehicle in the scene.

4. The method for generating hazardous scenarios for autonomous driving decision-making as described in claim 3, characterized in that, The initial state space of the scenario for: in, Indicates the physical state of an autonomous vehicle; It represents the physical state of other dynamic elements in the environment besides the autonomous vehicle itself; This indicates road structure information.

5. The method for generating hazardous scenarios for autonomous driving decision-making as described in claim 4, characterized in that, The physical state of each dynamic element is ,in, These represent the positions of dynamic elements. These represent the velocities of the dynamic element in two directions, respectively. The road structure in static elements is represented as ,in, These represent the length, width, and curvature of the road, respectively. A series of reference points representing lanes.

6. A system for generating hazardous scenarios for autonomous driving decision-making, characterized in that, include: The scene information extraction module is used to extract static and dynamic element information of the scene when the driver takes over the autonomous vehicle during real-vehicle road testing. The model building module is used to build a dynamic element behavior model based on the extracted static and dynamic element information. The parameter combination determination module is used to set the parameter space of the dynamic element behavior model, determine the parameter combination of the dynamic element behavior model in each simulation test, and construct the dangerous scenario. The hazardous scene area determination module is used to cluster and fit the parameter combinations of the constructed hazardous scenes to generate hazardous scene areas under the parameter combinations. The process of setting the parameter space of the dynamic element behavior model, determining the parameter combination of the dynamic element behavior model in each simulation test, and constructing hazardous scenarios includes: Define the parameter space for the dynamic element behavior model; Discretize the set parameter space with a certain step size; An optimized search method is used to determine the parameter combination of the dynamic element behavior model in each simulation test based on the discretized parameter space, and to construct dangerous scenarios and record the test results. The method employs an optimization search approach to determine the parameter combinations of the dynamic element behavior model in each simulation test based on the discretized parameter space, and constructs hazardous scenarios and records test results, including: Using the Metropolis criterion, a set of parameter combinations is selected from the discretized parameter space, and a new parameter combination is determined within the range of adjacent parameters with a certain search radius; Check whether the determined new parameter combination is the same as the parameters that have been tested. If they are the same, select them randomly again. Based on the combination of parameters detected, a hazardous scenario is constructed, and the test results are recorded.

7. A processing device, characterized in that, It includes computer program instructions, wherein when the computer program instructions are executed by a processing device, they are used to implement the steps corresponding to the autonomous driving decision-making hazardous scenario generation method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, wherein when the computer program instructions are executed by a processor, they are used to implement the steps corresponding to the autonomous driving decision-making hazardous scenario generation method according to any one of claims 1-5.

Images