Scene generalization method for simulation test, test method, system and medium

Abstract

The invention relates to a scene generalization method for a simulation test, a test method, a system and a medium, and relates to the technical field of automatic driving, the method comprises the steps that a simulation scene is disassembled based on a test scene six-layer model, and the six-layer model comprises a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer and a digital information layer; splitting the simulation scene based on an automatic driving function, wherein the automatic driving function comprises a plurality of sub-functions; splitting the simulation scene based on a behavior element model, wherein the behavior element model comprises a target vehicle triggering behavior, a time sequence behavior, a distance sequence behavior and a main vehicle triggering behavior; and performing cross generalization on the six-layer model, the automatic driving function and the behavior element model to generate a plurality of simulation test scenes. According to the invention, the coverage of the simulation scene is improved, so that the performance of the automatic driving algorithm is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular, to a scenario generalization method, a test method, a system, and a medium for simulation testing. Background Art

[0002] With the rapid development of autonomous driving technology, simulation testing plays an increasingly important role in the development and verification of autonomous driving systems. Simulation testing can simulate various complex driving scenarios in a virtual environment, which helps to evaluate the performance and safety of autonomous driving systems, and at the same time greatly reduces the cost and risk of real vehicle testing.

[0003] However, there are still some limitations in the current simulation test scenario construction methods. Most test scenarios mainly come from real vehicle collected data and manual experience design, lacking systematicness and comprehensiveness. This method is difficult to cover all situations that an autonomous driving system may encounter, especially some rare but potentially dangerous scenarios. In addition, manually constructing a large number of test scenarios is also a time-consuming and laborious task.

[0004] Therefore, the industry needs a more systematic and automated simulation test scenario generation method. This method should be able to model and generalize driving scenarios from multiple dimensions, generate a large number of diverse test scenarios, and improve test coverage. At the same time, this method should also have good scalability and be able to continuously improve and update the test scenario library with the progress of autonomous driving technology. Summary of the Invention

[0005] In view of the above-mentioned disadvantages of the prior art, the purpose of the present invention is to provide a scenario generalization method, a test method, a system, and a medium for simulation testing, which improve the coverage of simulation scenarios and thus further improve the performance of autonomous driving algorithms.

[0006] To achieve the above object, the present invention adopts the following technical solutions.

[0007] In the first aspect, a scenario generalization method for autonomous driving simulation testing provided by the present invention adopts the following technical solutions: Disassemble the simulation scenario based on a six-layer model of the test scenario, where the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer; Disassemble the simulation scenario based on autonomous driving functions, where the autonomous driving functions include multiple sub-functions; Disassemble the simulation scenario based on a behavior element model, where the behavior element model includes a target vehicle trigger behavior, a time series behavior, a distance series behavior, and a host vehicle trigger behavior; Perform cross generalization on the six-layer model, autonomous driving functions, and behavior element model to generate multiple simulation test scenarios.

[0008] Further, in the above scene generalization method, when disassembling the simulation scene based on the six-layer model of the test scene, the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer, and includes: Disassemble the road layer into road geometry and topology, and road surface quality and boundaries; Disassemble the traffic infrastructure layer into structural boundaries and traffic signs, and elevated roadblocks; Disassemble the temporary operation layer into a road construction site; Disassemble the object layer into static objects, dynamic objects, and interactions and operations; Disassemble the environment layer into weather and lighting conditions; Disassemble the digital information layer into V2X information and digital ground input information.

[0009] Further, in the above scene generalization method, when disassembling the simulation scene based on the autonomous driving function, it includes: Disassemble the autonomous driving function into line tracking, cut-in, cut-out, car following, lane change, dodging, ramp merging, and / or ramp exiting.

[0010] Further, in the above scene generalization method, when disassembling the simulation scene based on the behavior element model, it includes: Disassemble the target vehicle triggering behavior into the collision time with the target vehicle, the time headway with the target vehicle, and the relative speed with the target vehicle; Disassemble the time series behavior into the target appearance time, the total scene deduction time, and the deduction time after the reference point; Disassemble the distance series behavior into the main vehicle running mileage, the target running mileage, and the distance from the reference point; Disassemble the main vehicle triggering behavior into the collision time with the main vehicle, the time headway with the main vehicle, the relative distance from the main vehicle, and the relative speed with the main vehicle.

[0011] Further, in the above scene generalization method, when cross-generalizing the six-layer model, the autonomous driving function, and the behavior element model, it includes: Combine the elements of each layer in the six-layer model; Combine the combined six-layer model elements with the autonomous driving sub-functions; Combine the combined six-layer model elements and the autonomous driving sub-functions with the behavior element model.

[0012] Further, in the above scene generalization method, it further includes: Perform parametric setting on the cross-generalized scene to generate multiple parametric scenes.

[0013] Further, in the above scene generalization method, the parametric setting includes setting road geometric parameters, traffic flow parameters, environmental parameters, and / or vehicle dynamics parameters in the scene.

[0014] In a second aspect, an autonomous driving test method provided by the present invention adopts the following technical solution: Based on the scene generalization method described in any one of the above first aspects, generate multiple simulation test scenarios; Use the generated simulation test scenarios to perform simulation tests on the autonomous driving system; and, Optimize the autonomous driving algorithm based on the simulation test results.

[0015] In a third aspect, a scene generalization system for autonomous driving simulation test provided by the present invention adopts the following technical solution: A test scenario decomposition module, at least used to decompose the simulation scenario based on a six-layer model of the test scenario, and the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environmental layer, and a digital information layer; A driving function decomposition module, at least used to decompose the simulation scenario based on the autonomous driving function, and the autonomous driving function includes multiple sub-functions; A behavior element decomposition module, at least used to decompose the simulation scenario based on a behavior element model, and the behavior element model includes a target vehicle trigger behavior, a time series behavior, a distance series behavior, and a host vehicle trigger behavior; A scene generalization generation module, at least used to perform cross generalization on the six-layer model, the autonomous driving function, and the behavior element model to generate multiple simulation test scenarios.

[0016] In a fourth aspect, an autonomous driving test system provided by the present invention adopts the following technical solution: The scene generalization system described in the above third aspect, at least used to generate multiple simulation test scenarios; A simulation test module, at least used to perform simulation tests on the autonomous driving system using the generated simulation test scenarios; and, An algorithm optimization module, at least used to optimize the autonomous driving algorithm based on the simulation test results.

[0017] In a fifth aspect, a readable storage medium provided by the present invention adopts the following technical solution: A readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method described in any one of the above first aspects or second aspects is implemented.

[0018] In summary, compared with the prior art, the present invention includes at least one of the following beneficial technical effects: The present invention significantly improves the coverage and diversity of the simulation test scenarios. Through systematic scenario decomposition and cross generalization, this method can generate a large number of test scenarios with rich types, covering various situations that an autonomous vehicle may encounter on the actual road. These scenarios can be used to comprehensively evaluate the performance of the autonomous driving algorithm, discover potential safety hazards, and thus improve the reliability and safety of the autonomous driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0020] Figure 1 It is a flowchart of a specific embodiment of a scenario generalization method for simulation testing according to the present invention.

[0021] Figure 2 It is a schematic diagram of a specific embodiment of the six-layer model in the present invention.

[0022] Figure 3 It is a flowchart of a specific embodiment of an autonomous driving test method according to the present invention.

[0023] Figure 4 It is a schematic structural diagram of a specific embodiment of a scenario generalization system for simulation testing according to the present invention.

[0024] Figure 5 It is a schematic structural diagram of a specific embodiment of an autonomous driving test system according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application. In addition, it should be understood that the specific embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0026] It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments of the present application. And in the following embodiments, each embodiment has its own emphasis. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0027] The method steps described in the embodiments of the present invention may be executed in the order described in the specific implementation manner, or the execution order of each step may be adjusted according to actual needs and on the premise that the technical problem can be solved, which will not be listed here one by one.

[0028] Reference Figure 1 The present invention provides a scenario generalization method for simulation testing, comprising the following steps.

[0029] S1, disassemble the simulation scene based on the six-layer model of the test scene. The six-layer model includes the road layer, traffic infrastructure layer, temporary operation layer, object layer, environment layer and digital information layer. By systematically disassembling these six layers, all elements in the simulation scene can be fully covered.

[0030] S2, disassemble the simulation scenario based on the autonomous driving function. The autonomous driving function includes multiple sub-functions, such as line patrol, lane change, and vehicle following. By disassembling these sub-functions, test scenarios can be designed in a targeted manner.

[0031] S3, disassemble the simulation scene based on the behavior element model. The behavior element model includes the target vehicle triggering behavior, time series behavior, distance series behavior and host vehicle triggering behavior. These behavior elements describe the dynamic characteristics of each participant in the scene.

[0032] S4, cross-generalize the above six-layer models, autonomous driving functions and behavioral element models to generate multiple simulation test scenarios. Cross-generalization refers to combining the elements of these three dimensions to generate a large number of test scenarios.

[0033] Reference Figure 2 In the six-layer model, the road layer involves the geometric characteristics and surface conditions of the road; the traffic infrastructure layer includes traffic signs and roadblocks; the temporary operation layer considers temporary situations such as road construction; the object layer includes static and dynamic obstacles; the environment layer involves weather and lighting conditions; and the digital information layer includes information such as V2X communications.

[0034] The breakdown of autonomous driving functions involves specific operations that the vehicle needs to perform in different scenarios, such as cruising, changing lanes, and overtaking on the highway.

[0035] In the behavioral element model, the target vehicle trigger behavior describes the behavioral characteristics of other vehicles; the time series behavior and distance series behavior describe the evolution of the scene over time and distance; and the host vehicle trigger behavior describes the behavioral characteristics of the autonomous driving vehicle itself.

[0036] Through cross-generalization, this method combines the elements of these three dimensions to generate a large number of test scenarios. For example, combining a highway scenario (road layer) with a lane-changing function (autonomous driving function) and a sudden deceleration of the target vehicle (behavioral element) forms a specific test scenario.

[0037] Through systematic scenario decomposition and cross-generalization, this method can generate a large number and rich types of test scenarios, covering various situations that an autonomous driving vehicle may encounter on actual roads. These scenarios can be used to comprehensively evaluate the performance of autonomous driving algorithms, discover potential safety hazards, and thus improve the reliability and safety of autonomous driving systems.

[0038] Furthermore, as an implementation manner of the present invention, in step S2, the simulation scenario is decomposed based on the six-layer model of the test scenario. The six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer, including: Decompose the road layer into road geometry and topology, and road surface quality and boundaries; Decompose the traffic infrastructure layer into structural boundaries and traffic signs, and elevated roadblocks; Decompose the temporary operation layer into road construction sites; Decompose the object layer into static objects, dynamic objects, and interactions and operations; Decompose the environment layer into weather and lighting conditions; Decompose the digital information layer into V2X information and digital ground input information.

[0039] Specifically, the road layer is decomposed into road geometry and topology, and road surface quality and boundaries. Road geometry and topology include geometric features such as the curvature, slope, and width of the road, as well as topological features such as the connection relationship of the road. Road surface quality and boundaries include quality features such as the friction coefficient and flatness of the road surface, as well as boundary features such as shoulders and guardrails.

[0040] The traffic infrastructure layer is decomposed into structural boundaries and traffic signs, and elevated roadblocks. Structural boundaries include road structures such as bridges and tunnels. Traffic signs include various indicator signs, signal lights, etc. Elevated roadblocks include viaducts, overpasses, etc.

[0041] The temporary operation layer is decomposed into road construction sites, including the scope of the construction area, the location of construction equipment, traffic control measures, etc.

[0042] The object layer is decomposed into static objects, dynamic objects, and interactions and operations. Static objects include parked vehicles, roadside buildings, etc. Dynamic objects include vehicles in motion, pedestrians, etc. Interactions and operations include interaction behaviors between vehicles, interaction behaviors between vehicles and pedestrians, etc.

[0043] The environmental layer is disassembled into weather and lighting conditions. Weather conditions include sunny days, rainy days, foggy days, etc. Lighting conditions include the lighting situations at different times such as daytime, night, and dusk.

[0044] The digital information layer is disassembled into V2X information and digital ground input information. V2X information includes information such as vehicle-to-vehicle communication and vehicle-to-road communication. Digital ground input information includes high-precision maps, real-time traffic information, etc.

[0045] By disassembling this six-layer model in detail, a test framework that comprehensively covers various road scene elements is formed. This disassembly method helps to generate diverse test scenarios, thereby comprehensively evaluating the performance of the autonomous driving algorithm.

[0046] Furthermore, as an implementation manner of the present invention, in step S2, the simulation scenario is disassembled based on the autonomous driving function, including: The autonomous driving function is disassembled into lane tracking, lane cutting in, lane cutting out, vehicle following, lane changing, dodging, ramp merging, and / or ramp exiting.

[0047] Specifically, lane tracking is the basic function for an autonomous driving vehicle to maintain its lane position on the road. The test scenarios for the lane tracking function include the lane-keeping ability under different road conditions such as straight roads, curved roads, and slopes. Lane cutting in refers to the process of an autonomous driving vehicle entering from one lane into another lane. The test scenarios for the lane cutting-in function include lane-cutting operations under different vehicle speeds and different vehicle distances. Lane cutting out is the opposite of lane cutting in, referring to the process of an autonomous driving vehicle exiting from the current lane to an adjacent lane. The test scenarios for the lane cutting-out function include lane-cutting operations in situations such as emergency avoidance and preparing to exit. Vehicle following refers to the function of an autonomous driving vehicle to keep a safe distance from the vehicle in front. The test scenarios for the vehicle-following function include the following reactions under different behaviors of the vehicle in front such as accelerating, decelerating, and sudden braking. Lane changing refers to the entire process of an autonomous driving vehicle completing a change from one lane to another lane, including decision-making, execution, and completion. The test scenarios for the lane-changing function include various situations such as single lane change, continuous lane change, and emergency lane change. Dodging refers to the function of an autonomous driving vehicle to perform an emergency avoidance of a sudden obstacle. The test scenarios for the dodging function include situations such as avoiding static obstacles and dynamic obstacles. Ramp merging refers to the process of an autonomous driving vehicle entering the main road from the ramp. The test scenarios for the ramp-merging function include operations under different traffic flow densities and different merging angles, etc. Ramp exiting refers to the process of an autonomous driving vehicle entering the ramp from the main road. The test scenarios for the ramp-exiting function include a series of operations such as early lane change, deceleration, and steering.

[0048] By disassembling these sub - functions, specific test scenarios can be designed for various situations encountered by autonomous vehicles on actual roads. For example, for the lane - changing function, multiple scenarios can be designed, including normal lane - changing, emergency lane - changing, continuous lane - changing, etc. Each scenario can be combined with different road conditions, traffic flow conditions, and other factors. This function - based disassembly method makes the test more comprehensive and targeted. By conducting detailed tests on each sub - function, the performance of the autonomous driving algorithm in various complex situations can be comprehensively evaluated, thereby improving the safety and reliability of autonomous driving technology.

[0049] Furthermore, as an implementation manner of the present invention, in step S3, disassembling the simulation scenario based on the behavior element model includes: Disassembling the target vehicle triggering behavior into the time - to - collision (TTC) with the target vehicle, the time - headway (THW) with the target vehicle, and the relative speed with the target vehicle; Disassembling the time - series behavior into the target appearance time, the total scene deduction time, and the deduction time after the reference point; Disassembling the distance - series behavior into the host vehicle running mileage, the target running mileage, and the distance from the reference point; Disassembling the host vehicle triggering behavior into the time - to - collision with the host vehicle, the time - headway with the host vehicle, the relative distance from the host vehicle, and the relative speed with the host vehicle.

[0050] Specifically, disassembling the simulation scenario based on the behavior element model is an important step. The behavior element model includes the target vehicle triggering behavior, the time - series behavior, the distance - series behavior, and the host vehicle triggering behavior. By disassembling these behavior elements in detail, a test framework that comprehensively describes the dynamic characteristics of the scenario is formed.

[0051] The target vehicle triggering behavior is disassembled into the time - to - collision (TTC) with the target vehicle, the time - headway (THW) with the target vehicle, and the relative speed with the target vehicle. TTC describes the time when a collision may occur between the autonomous vehicle and the target vehicle. THW represents the time interval between the autonomous vehicle and the target vehicle in front. The relative speed reflects the speed difference between the autonomous vehicle and the target vehicle. These parameters jointly describe the potential impact of the target vehicle on the autonomous vehicle.

[0052] The time - series behavior is disassembled into the target appearance time, the total scene deduction time, and the deduction time after the reference point. The target appearance time defines the moment when other vehicles or obstacles appear in the scenario. The total scene deduction time sets the duration of the entire test scenario. The deduction time after the reference point describes the time of scene evolution calculated from a specific event or position. These time parameters are used to control the occurrence order and duration of each event in the scenario.

[0053] The distance sequence behavior is disassembled into the main vehicle's driving mileage, the target driving mileage, and the distance from the reference point. The main vehicle's driving mileage represents the total distance traveled by the autonomous vehicle in the scenario. The target driving mileage describes the distance traveled by other vehicles or moving obstacles in the scenario. The distance from the reference point defines the positional relationship of each element in the scenario relative to a fixed reference point. These distance parameters are used to describe the spatial distribution and movement trajectories of the various participants in the scenario.

[0054] The main vehicle-triggered behavior is disassembled into the time to collision (TTC) with the main vehicle, the time headway (THW) with the main vehicle, the relative distance from the main vehicle, and the relative speed with the main vehicle. These parameters describe the interaction relationship between the autonomous vehicle and its surrounding environment from the perspective of the autonomous vehicle. TTC and THW reflect the safety interval between the autonomous vehicle and other vehicles. The relative distance and relative speed directly describe the spatial and movement relationships between the autonomous vehicle and the surrounding vehicles or obstacles.

[0055] By performing such a detailed disassembly of the behavior element model, the test scenario can more accurately simulate various complex situations on real roads. For example, by adjusting the TTC and THW of the target vehicle, the test scenario can simulate different levels of dangerous situations. By setting different time series and distance sequence parameters, the test scenario can cover various dynamic traffic flow situations. This disassembly method makes the test more comprehensive and targeted, and helps to comprehensively evaluate the performance of autonomous driving algorithms in various complex dynamic scenarios.

[0056] Furthermore, as an implementation manner of the present invention, in step S4, cross-generalization is performed on the six-layer model, the autonomous driving function, and the behavior element model, including: Combining the elements of each layer in the six-layer model; Combining the combined six-layer model elements with the autonomous driving sub-functions; Combining the combined six-layer model elements and the autonomous driving sub-functions with the behavior element model.

[0057] Specifically, the elements of each layer in the six-layer model are combined. Exemplarily, the road geometric features of the road layer are combined with the traffic signs of the traffic infrastructure layer to form a basic scenario with a specific road shape and markings. The road construction conditions of the temporary operation layer are combined with the static obstacles of the object layer to create a complex road condition containing the construction area and obstacles. The weather conditions of the environment layer are combined with the V2X communication of the digital information layer to simulate the vehicle communication situation under different weather conditions. In this way, the elements of the six-layer model are systematically combined to form a diverse basic scenario structure.

[0058] Combine the combined six - layer model elements with the autonomous driving sub - functions. Exemplarily, in the highway scenario (from the six - layer model combination), add the lane - changing function to create a specific scenario for testing the lane - changing ability of the autonomous driving system. Or, in the urban intersection scenario, add the turning function to form a test environment for evaluating the turning performance of the autonomous driving system in complex road conditions. Through this combination, each test scenario not only includes a specific road environment but also is optimized for a specific autonomous driving function.

[0059] Combine the combined six - layer model elements and autonomous driving sub - functions with the behavior element model. Exemplarily, in the aforementioned highway lane - changing scenario, add the target vehicle trigger behavior (such as sudden deceleration) and time - series behavior (such as setting the start and end times of lane - changing) to form a dynamic and challenging test scenario. Or, in the urban turning scenario, add the distance - series behavior (such as setting the distance from the turning point to other vehicles) and the host vehicle trigger behavior (such as the relative speed with other vehicles) to create a complex traffic flow test environment.

[0060] Through this multi - level cross - generalization process, this method generates a large number of diverse and targeted simulation test scenarios. Each scenario contains a combination of aspects such as road environment, traffic conditions, autonomous driving functions, and dynamic behaviors, comprehensively covering various situations that the autonomous driving system may encounter during actual operation. This method not only improves the comprehensiveness of testing but also increases the possibility of discovering potential problems and boundary cases, thus helping to improve the performance and safety of autonomous driving algorithms.

[0061] Furthermore, as an implementation manner of the present invention, the scenario generalization method further includes: Perform parametric settings on the scenarios after cross - generalization to generate multiple parametric scenarios. Among them, the parametric settings include setting the road geometric parameters, traffic flow parameters, environmental parameters, and / or vehicle dynamics parameters in the scenarios.

[0062] Specifically, after cross - generalizing the scenarios, the generated scenarios can be further parametrically set to produce more diverse test scenarios. Parametric setting means assigning variable parameter values to each element in the scenario and generating multiple scenario variants with different characteristics by adjusting these parameter values.

[0063] Exemplarily, the parametric setting includes the following steps: 1. Determine the scenario elements that need to be parameterized. These elements may include road geometric parameters, traffic flow parameters, environmental parameters, and vehicle dynamics parameters, etc. For example, road geometric parameters include road curvature, slope, lane width, etc.; traffic flow parameters include traffic density, vehicle speed distribution, etc.; environmental parameters include visibility, friction coefficient, etc.; vehicle dynamics parameters include acceleration, steering angle, etc.

[0064] 2. Define a value range and step size for each parameter. The value range should cover the reasonable variation interval of the parameter in the actual situation, and the step size determines the accuracy of the parameter change. For example, for road curvature, the value range can be set as 0 - 0.1 m^(-1), and the step size is 0.01 m^(-1); for traffic density, the value range can be set as 10 - 100 vehicles / km, and the step size is 10 vehicles / km.

[0065] 3. Use parameterization tools or algorithms to traverse or randomly sample each parameter within its value range. This process generates a large number of parameter combinations, and each set of parameter combinations corresponds to a specific scenario variant.

[0066] 4. Apply these parameter combinations to the base scenario to generate multiple parameterized scenarios. Each parameterized scenario maintains the structure of the base scenario but differs in specific details.

[0067] Through parameterization settings, a base scenario can give rise to dozens, hundreds, or even more scenario variants. These variants cover various possible situations, greatly increasing the comprehensiveness and depth of testing. For example, for a highway lane - changing scenario, by adjusting parameters such as lane width, traffic density, and lane - changing distance, multiple test scenarios with different difficulties are generated to comprehensively evaluate the lane - changing performance of the autonomous driving algorithm.

[0068] Parameterization settings not only increase the number and diversity of test scenarios but also improve the systematicness and controllability of testing. By systematically adjusting parameters, testers can explore the performance of the autonomous driving algorithm under various boundary conditions and discover potential problems and weaknesses. At the same time, the parameterization method is also convenient for sensitivity analysis to understand which parameters have the greatest impact on the algorithm performance, so as to optimize the algorithm targeted.

[0069] In summary, the scenario generalization method generates a large number of diverse simulation test scenarios by systematically disassembling and cross-generalizing the six-layer model of the test scenario, the autonomous driving function, and the behavior element model. This method significantly improves the coverage and diversity of the simulation test scenarios, covering various complex situations that autonomous driving vehicles may encounter on actual roads. By parametrically setting the generalized scenarios, including adjusting road geometry parameters, traffic flow parameters, environmental parameters, and vehicle dynamics parameters, the variability of the test scenarios is further increased. This comprehensive and systematic scenario generation method helps to comprehensively evaluate the performance of autonomous driving algorithms, discover potential safety hazards, and thus improve the reliability and safety of autonomous driving technology.

[0070] Referring to Figure 3 , based on the scenario generalization method described in the above embodiments, an embodiment of the present invention also discloses an autonomous driving test method, including the following steps.

[0071] A1. Use the scenario generalization method to generate a large number of simulation test scenarios. This scenario generalization method is based on the six-layer model of the test scenario, the autonomous driving function, and the behavior element model for systematic disassembly and cross-generalization, generating hundreds of millions of diverse simulation test scenarios. These scenarios cover various complex situations that autonomous driving vehicles may encounter on actual roads, including unknown scenarios and potential dangerous scenarios.

[0072] A2. Use the generated simulation test scenarios to conduct simulation tests on the autonomous driving system. During the test, the autonomous driving system faces various complex road environments, traffic conditions, and dynamic behaviors, and its performance and responses are comprehensively evaluated. The test results record the performance of the autonomous driving system in each scenario, including aspects such as decision-making, control, and safety.

[0073] A3. Optimize the autonomous driving algorithm based on the simulation test results. By analyzing the test results, identify the weaknesses and potential problems of the autonomous driving algorithm. In particular, for the dangerous scenarios found in the test, feedback these scenarios to the algorithm development team for algorithm iteration and improvement. In this way, the autonomous driving algorithm is continuously optimized to improve its ability to handle various complex scenarios.

[0074] This test method is applicable to the functional safety test of L3-level autonomous driving, especially for the test of unknown scenarios. By generating and testing a large number of scenarios, this method can explore the dangerous scenarios that the vehicle may encounter. These discovered dangerous scenarios are used for the iterative optimization of the algorithm, thereby achieving the repair of dangerous problems. Through this systematic testing and optimization process, this method helps to realize the development and improvement of L3-level autonomous driving algorithms. This method not only improves the safety and reliability of the autonomous driving system but also provides strong support for the further development of autonomous driving technology.

[0075] Reference Figure 4 Figure 4 , an embodiment of the present invention also discloses a scenario generalization system for simulation testing, including a test scenario disassembly module 1, a driving function disassembly module 2, a behavior element disassembly module 3, and a scenario generalization generation module 4.

[0076] The test scenario disassembly module 1 disassembles the simulation scenario based on the six-layer model of the test scenario. The six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer. The test scenario disassembly module 1 disassembles each layer in detail. For example, the road layer is disassembled into road geometry and topology, road surface quality and boundaries; the traffic infrastructure layer is disassembled into structural boundaries, traffic signs, elevated roadblocks, etc. Through this systematic disassembly, the test scenario disassembly module 1 provides comprehensive basic elements for subsequent scenario generation.

[0077] The driving function disassembly module 2 disassembles the simulation scenario based on the autonomous driving function. The autonomous driving function includes multiple sub-functions, such as lane tracking, cut-in, cut-out, car following, lane change, dodging, ramp entry, and ramp exit. The driving function disassembly module 2 refines these sub-functions and defines specific test requirements and scenario features for each sub-function. This function-oriented disassembly method ensures that the generated test scenarios can specifically evaluate the various functions of the autonomous driving system.

[0078] The behavior element disassembly module 3 disassembles the simulation scenario based on the behavior element model. The behavior element model includes target vehicle trigger behavior, time series behavior, distance series behavior, and host vehicle trigger behavior. The behavior element disassembly module 3 disassembles these behaviors in detail. For example, the target vehicle trigger behavior is disassembled into the collision time with the target vehicle, the time distance from the target vehicle, the relative speed with the target vehicle, etc. This behavior-oriented disassembly method enables the generated test scenarios to simulate complex dynamic traffic environments.

[0079] The scenario generalization generation module 4 performs cross-generalization on the six-layer model, the autonomous driving function, and the behavior element model to generate multiple simulation test scenarios. The cross-generalization process includes multiple steps: First, the elements of each layer in the six-layer model are combined; Second, the combined six-layer model elements are combined with the autonomous driving sub-functions; Finally, the combination results of the first two steps are combined with the behavior element model. Through this multi-level cross-generalization, the scenario generalization generation module 4 generates a large number of diverse and targeted simulation test scenarios.

[0080] The scenario generalization system for simulation testing generates a large number of diverse simulation test scenarios through systematic scenario decomposition and cross-generalization methods. The system not only considers multiple aspects such as road environment, traffic conditions, and dynamic behaviors, but also allows for parametric settings of the generated scenarios, including adjusting road geometry parameters, traffic flow parameters, environmental parameters, and vehicle dynamics parameters. This comprehensive and flexible scenario generation method significantly improves the coverage and depth of simulation testing, helps to comprehensively evaluate the performance of autonomous driving algorithms, discovers potential safety hazards, and thus enhances the reliability and safety of autonomous driving technology.

[0081] Referring to Figure 5 , an embodiment of the present invention also discloses an autonomous driving test system, including a scenario generalization system 10, a simulation testing module 20, and an algorithm optimization module 30.

[0082] The scenario generalization system 10 generates a large number of diverse simulation test scenarios. The scenario generalization system 10 performs systematic decomposition and cross-generalization based on the six-layer model of test scenarios, the autonomous driving function, and the behavior element model, generating test scenarios covering various complex road environments, traffic conditions, and dynamic behaviors.

[0083] The simulation testing module 20 uses the simulation test scenarios generated by the scenario generalization system 10 to conduct simulation tests on the autonomous driving system. During the testing process, the performance and reactions of the autonomous driving system are comprehensively evaluated in the face of various complex scenarios. The simulation testing module 20 records the performance of the autonomous driving system in each scenario, including aspects such as decision-making, control, and safety.

[0084] The algorithm optimization module 30 optimizes the autonomous driving algorithm based on the test results of the simulation testing module 20. The algorithm optimization module 30 analyzes the test results and identifies the weaknesses and potential problems of the autonomous driving algorithm. For the dangerous scenarios discovered during the testing, the algorithm optimization module 30 feeds these scenarios back to the algorithm development team for algorithm iteration and improvement.

[0085] An embodiment of the present invention also discloses a readable storage medium.

[0086] A readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the method described in any one of the above embodiments. The computer-readable storage medium may include: any entity or device capable of carrying the computer program, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), and a software distribution medium, etc. The computer program includes computer program code. The computer program code may be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), and a software distribution medium, etc.

[0087] Any process or method description shown in a flowchart or described in other ways herein can be understood to represent a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations, where functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed. This should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0088] The logic and / or steps represented in a flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices.

[0089] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A scenario generalization method for simulation testing, characterized in that: include: Decomposing the simulation scenario based on a six-layer model of the test scenario, wherein the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer; Disassembling the simulation scenario based on an autonomous driving function, wherein the autonomous driving function includes a plurality of sub-functions; Decomposing the simulation scenario based on the behavior element model, wherein the behavior element model includes the target vehicle triggering behavior, the time series behavior, the distance series behavior and the host vehicle triggering behavior; The six-layer model, autonomous driving function and behavior element model are cross-generalized to generate multiple simulation test scenarios.

2. The scene generalization method according to claim 1, characterized in that: The six-layer model based on the test scenario is used to decompose the simulation scenario, and the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer and a digital information layer, including: Decomposing the road layer into road geometry and topology, road surface quality and boundaries; Decomposing the transportation infrastructure layer into structural boundaries, traffic signs, and elevated roadblocks; Dismantling the temporary operating layer into a road construction site; Decomposing the object layer into static objects, dynamic objects, and interactions and operations; Decomposing the environment layer into weather and lighting conditions; The digital information layer is split into V2X information and digital ground input information.

3. The scene generalization method according to claim 1, characterized in that: The decomposition of the simulation scene based on the autonomous driving function includes: Break down the autonomous driving functions into lane following, cutting in, cutting out, following a vehicle, changing lanes, dodging, merging onto ramps and / or exiting ramps.

4. The scene generalization method according to claim 1, characterized in that: The decomposing of the simulation scene based on the behavior element model includes: The triggering behavior of the target vehicle is decomposed into the collision time with the target vehicle, the time distance with the target vehicle, and the relative speed with the target vehicle; Decompose the time series behavior into target appearance time, total scenario deduction time, and deduction time after the reference point; Disassemble the distance sequence behavior into the main vehicle mileage, target mileage, and distance to the reference point; The triggering behavior of the main vehicle is decomposed into the collision time with the main vehicle, the time distance with the main vehicle, the relative distance with the main vehicle, and the relative speed with the main vehicle.

5. The scene generalization method according to claim 1, characterized in that: The cross-generalization of the six-layer model, autonomous driving function and behavior element model includes: Combine the elements of each layer in the six-layer model; Combine the combined six-layer model elements with the autonomous driving sub-functions; Combine the combined six-layer model elements and autonomous driving sub-functions with the behavioral element model.

6. The scene generalization method according to claim 1, characterized in that: Also includes: The cross-generalized scenes are parameterized to generate multiple parameterized scenes.

7. The scene generalization method according to claim 6, characterized in that: The parameterized setting includes setting road geometry parameters, traffic flow parameters, environment parameters and / or vehicle dynamics parameters in the scene.

8. An automatic driving test method, characterized in that: The method comprises: Based on the scenario generalization method as described in any one of claims 1 to 7, generating multiple simulation test scenarios; Performing simulation tests on the autonomous driving system using the generated simulation test scenarios; and, The autonomous driving algorithm is optimized based on the simulation test results.

9. A scenario generalization system for simulation testing, characterized in that: include: A test scenario disassembly module, at least for disassembling the simulation scenario based on a six-layer model of the test scenario, wherein the six-layer model includes a road layer, a traffic infrastructure layer, a temporary operation layer, an object layer, an environment layer, and a digital information layer; A driving function disassembly module, at least for disassembling a simulation scene based on an autonomous driving function, wherein the autonomous driving function includes a plurality of sub-functions; A behavior element disassembly module, at least used to disassemble the simulation scene based on a behavior element model, wherein the behavior element model includes a target vehicle triggering behavior, a time series behavior, a distance series behavior, and a host vehicle triggering behavior; The scenario generalization generation module is at least used to cross-generalize the six-layer model, the autonomous driving function and the behavior element model to generate multiple simulation test scenarios.

10. An automatic driving test system, characterized in that: include: The scenario generalization system as claimed in claim 9, at least used to generate a plurality of simulation test scenarios; A simulation test module, at least used to perform simulation test on the autonomous driving system using the generated simulation test scenario; as well as, The algorithm optimization module is at least used to optimize the autonomous driving algorithm based on the simulation test results.

11. A readable storage medium, characterized in that: The readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images