Method and device for generating automatic driving safety boundary model based on simulation environment

Abstract

The invention discloses a method for generating an automatic driving safety boundary model based on a simulation environment, and belongs to the technical field of vehicle automatic driving. A specific embodiment of the method comprises the following steps: performing safety boundary search processing on environmental perception data of automatic driving by using a safety boundary model, so as to efficiently identify safety boundaries of an automatic driving system under different working conditions; testing the test case forming the safety boundary in a simulation environment, reversely adjusting the safety boundary model based on automatic driving simulation data generated by the test, and obtaining a quasi-safety boundary model until the safety boundary model obtained through safety boundary search processing meets a reliability requirement; therefore, the test quantity of the automatic driving system in a real scene is reduced, the test cost and test time of the automatic driving system are reduced, and the accuracy and robustness of safety assessment of the automatic driving system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle autonomous driving, and in particular relates to a method and device for generating an autonomous driving safety boundary model based on a simulation environment. Background Art

[0002] With the gradual development of the field of artificial intelligence, automatic control systems have gradually expanded to all walks of life, such as: automobiles, sorting robots, sweeping robots and other robots and machinery equipped with automatic control systems. In particular, in the field of vehicles, more and more companies have begun to invest in the research and development of autonomous driving systems. The research and development of existing autonomous driving systems is generally based on simulation systems. The reliability of the current autonomous driving system is determined by running the code of the autonomous driving system on the simulation system, controlling the vehicle in the simulation system, and feeding back various parameters. The shortcomings of the autonomous driving system are analyzed and summarized through the above process, and after determining a reliable autonomous driving system, it is input into the actual vehicle for on-road testing.

[0003] Dangerous scenario simulation is the basis for the safety verification and evaluation of autonomous driving systems. Existing autonomous driving system tests are only conducted on a few specific scenarios, so the autonomous driving data obtained based on existing test scenarios is not sufficient to support the reliability requirements of autonomous driving systems. However, actual test scenarios may include complex traffic conditions, special weather conditions or emergency events, etc.

[0004] If all possible test scenarios are tested, the testing cost and time of the autonomous driving system will increase, which is obviously unrealistic.

[0005] Therefore, there is an urgent need to provide a method and device for generating an autonomous driving safety boundary model based on a simulation environment, so as to comprehensively evaluate the emergency response capability, fault tolerance, etc. of the autonomous driving system by simulating various dangerous scenarios and potential risks, and effectively optimize and improve the autonomous driving system based on the evaluation results. Summary of the invention

[0006] In response to the above-mentioned problems existing in the prior art, an embodiment of the present invention provides a method and device for generating an autonomous driving safety boundary model based on a simulation environment; this method not only reduces the testing cost of the autonomous driving system, but also improves the accuracy of the autonomous driving system in safety boundary scenario testing, and further improves the accuracy and robustness of the safety assessment of the autonomous driving vehicle.

[0007] According to a first aspect of an embodiment of the present invention, a method for generating an autonomous driving safety boundary model based on a simulation environment is provided, which at least includes: obtaining a scene variable range corresponding to environmental perception data of an autonomous driving vehicle, and a logical scene model corresponding to the scene variable range; performing a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value; testing a first boundary scene test case created by the first scene variable value and the logical scene model using an autonomous driving simulation device to generate first autonomous driving simulation data; optimizing the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value until the optimized safety boundary model meets a preset condition, then terminating the safety boundary search process to generate a quasi-safety boundary model.

[0008] Optionally, the scene variable range corresponding to the environmental perception data of the autonomous driving vehicle and the logical scene model corresponding to the scene variable range include: obtaining the environmental perception data of the autonomous driving vehicle; performing object extraction processing on the environmental perception data to generate a dynamic and static object extraction table; performing scene recognition processing on the dynamic and static object extraction table to generate a scene-related data stream; performing scene modeling processing using the scene-related data stream to output the scene variable range and the logical scene model corresponding to the scene variable range.

[0009] Optionally, the initial safety boundary model is optimized based on the first autonomous driving simulation data and the first scene variable value until the optimized safety boundary model meets the preset conditions, then the safety boundary search process is terminated to generate a quasi-safe boundary model; including: based on the first autonomous driving simulation data and the first scene variable value, the initial safety boundary model is optimized for the first time; a first optimized safety boundary model is generated; a reliability assessment is performed on the first optimized safety boundary model; if the reliability assessment result indicates that the first optimized safety boundary model meets the preset conditions, then the safety boundary search process of the scene variable range using the first optimized safety boundary model is terminated to generate a quasi-safe boundary model; if the reliability assessment result indicates that the first optimized safety boundary model does not meet the preset conditions, then the safety boundary search process of the scene variable range is performed using the first optimized safety boundary model until a quasi-safe boundary model is generated and then the safety boundary search process is terminated.

[0010] Optionally, based on the first autonomous driving simulation data and the first scenario variable value, the initial safety boundary model is optimized for the first time; and a first optimized safety boundary model is generated; including: based on the first autonomous driving simulation data, the driving state of the autonomous driving vehicle in the first boundary scenario test case is evaluated to generate a driving state evaluation result; based on the driving state evaluation result and the first scenario variable value, the initial safety boundary model is optimized for the first time to generate a first optimized safety boundary model.

[0011] Optionally, the method further includes: performing a joint distribution calculation on the scene variable range to generate a joint distribution of scene variables; based on the quasi-safety boundary model and the joint distribution of scene variables, evaluating the safety of the autonomous driving vehicle to generate a safety evaluation report.

[0012] Optionally, the first optimized safety boundary model is used to perform safety boundary search processing on the scene variable range until a quasi-safe boundary model is generated, and the safety boundary search processing is terminated; including: based on a Gaussian process searcher, the first optimized safety boundary model is used to perform safety boundary search processing on the scene variable range to generate a second scene variable value; based on the second scene variable value and the logical scene model, a second boundary scene test case is created; the second boundary scene test case is tested using an autonomous driving simulation device to generate second autonomous driving simulation data; based on the second autonomous driving simulation data and the second scene variable value, the first optimized boundary scene model is optimized using a Gaussian process safety scenario predictor; and the generated second optimized safety boundary model is reliability evaluated using a reliability evaluator; until the reliability evaluation result indicates that the optimized safety boundary model meets the preset conditions, a quasi-safe boundary model is generated and the safety boundary search processing is terminated.

[0013] According to a second aspect of an embodiment of the present invention, there is also provided an apparatus for generating an autonomous driving safety boundary model based on a simulation environment, comprising: an acquisition module for acquiring a scene variable range corresponding to environmental perception data of an autonomous driving vehicle, and a logical scene model corresponding to the scene variable range; a search processing module for performing a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value; a testing module for testing a first boundary scene test case created by the first scene variable value and the logical scene model using an autonomous driving simulation device to generate first autonomous driving simulation data; and an optimization processing module for optimizing the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value until the optimized safety boundary model meets a preset condition, then terminating the safety boundary search process to generate a quasi-safety boundary model.

[0014] Optionally, the optimization processing module includes: an optimization processing unit, used to perform a first optimization processing on the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value; generate a first optimized safety boundary model; a reliability evaluation unit, used to perform reliability evaluation on the first optimized safety boundary model; a generation unit, used to terminate the safety boundary search processing of the scene variable range by the first optimized safety boundary model if the reliability evaluation result indicates that the first optimized safety boundary model meets preset conditions, and generate a quasi-safety boundary model; a safety boundary search processing unit, used to perform safety boundary search processing on the scene variable range using the first optimized safety boundary model if the reliability evaluation result indicates that the first optimized safety boundary model does not meet preset conditions, and terminate the safety boundary search processing until a quasi-safety boundary model is generated.

[0015] According to a third aspect of an embodiment of the present invention, there is further provided a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect is implemented.

[0016] According to a fourth aspect of an embodiment of the present invention, an electronic device is also provided, comprising: a processor; a memory for storing instructions executable by the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in the first aspect.

[0017] An embodiment of the present invention provides a method for generating an autonomous driving safety boundary model based on a simulation environment, comprising: first, obtaining a scene variable range corresponding to environmental perception data of an autonomous driving vehicle, and a logical scene model corresponding to the scene variable range; second, using an initial safety boundary model to perform a safety boundary search process on the scene variable range to generate a first scene variable value; then, using an autonomous driving simulation device to test a first boundary scene test case created by the first scene variable value and the logical scene model to generate first autonomous driving simulation data; finally, based on the first autonomous driving simulation data and the first scene variable value, optimizing the initial safety boundary model until the optimized safety boundary model meets a preset condition, then terminating the safety boundary search process to generate a quasi-safety boundary model. This embodiment uses a safety boundary model to perform safety boundary search processing on the environmental perception data of autonomous driving, so as to efficiently identify the safety boundary of the autonomous driving system under different working conditions; then, the test cases that form the safety boundary are tested in a simulation environment, and then the safety boundary model is reversely adjusted based on the autonomous driving simulation data generated by the test, until the safety boundary model obtained through the safety boundary search processing meets the reliability requirements, then a quasi-safety boundary model is obtained; thereby, not only the test amount, test cost and test time of the autonomous driving system in real scenarios are reduced, but also the accuracy and robustness of the safety assessment of the autonomous driving system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:

[0019] Figure 1 A schematic diagram of a flow chart of a method for generating an autonomous driving safety boundary model based on a simulation environment provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of the structure of an apparatus for generating an autonomous driving safety boundary model based on a simulation environment provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, it is a flowchart of a method for generating an autonomous driving safety boundary model based on a simulation environment provided by one embodiment of the present invention.

[0023] The method for generating an autonomous driving safety boundary model based on a simulation environment includes at least the following steps:

[0024] S101, obtaining a scene variable range corresponding to the environmental perception data of the autonomous driving vehicle, and a logical scene model corresponding to the scene variable range;

[0025] S102, performing a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value;

[0026] S103, using an autonomous driving simulation device to test a first boundary scenario test case created by the first scenario variable value and the logic scenario model to generate first autonomous driving simulation data;

[0027] S104, based on the first autonomous driving simulation data and the first scenario variable value, the initial safety boundary model is optimized until the optimized safety boundary model meets the preset conditions, then the safety boundary search process is terminated to generate a quasi-safety boundary model.

[0028] In S101, the corresponding scene variables, scene variable ranges, and logical scene models corresponding to the scene variable ranges are determined based on the environmental perception data of the autonomous driving vehicle. For example, the environmental perception data is input into the model for prediction, and the scene variables and scene variable ranges corresponding to the environmental perception data are output; then the model is constructed based on the scene variables to obtain the logical scene model. Alternatively, the environmental perception data is extracted and processed, and the scene variables and scene variable ranges are output; then the model is constructed based on the scene variables to obtain the logical scene model.

[0029] Exemplarily, the environmental perception data of the autonomous driving vehicle is obtained; the environmental perception data is subjected to object extraction processing to generate a dynamic and static object extraction table; the dynamic and static object extraction table is subjected to scene recognition processing to generate a scene-related data stream; the scene-related data stream is used to perform scene modeling processing, and a scene variable range and a logical scene model corresponding to the scene variable range are output. For example: environmental perception data is obtained through the environmental perception module of the autonomous driving vehicle; the environmental perception data is input into the scene recognition module; the scene recognition module includes: a dynamic and static object extractor and a scene recognizer. The environmental perception data is extracted and processed by the dynamic and static object extractor to output a dynamic and static object extraction table; the dynamic and static object extraction table is subjected to scene recognition processing by the scene recognizer to output a scene-related data stream. The scene-related data stream is input into a scene model that combines data drive and physical drive for data processing, and scene variables, scene variable ranges, and logical scene models are output. Joint distribution calculation is performed on the scene variable range to generate a scene variable joint distribution.

[0030] Among them, the environment perception module includes laser radar, millimeter wave radar, camera, etc. The static object extraction table includes road signs, traffic lights, etc. The dynamic object extraction table includes other vehicles on the road, etc.

[0031] Here, the logical scenario model is a traffic scenario model, which is used to indicate the interaction between the autonomous vehicle and other vehicles beside it; for example, the vehicle in front of the autonomous vehicle suddenly brakes, and the vehicle in the left lane suddenly cuts in front of the autonomous vehicle. Scenario variables include variables such as the distance, speed, and acceleration of the lane-changing vehicle from the autonomous vehicle.

[0032] In S102, the safety margin model is a pure mathematical statistical model, which is used to indicate the intrinsic mathematical relationship between scene variables (for example, the statistical relationship between speed, acceleration and distance). The initial safety margin model is an initialized safety margin model.

[0033] Here, the security boundary search process can be performed based on a Gaussian process searcher.

[0034] The initial safety boundary model and the scene variable range are input into the Gaussian process searcher, a safety boundary search process is performed, and the first scene variable value is output.

[0035] In S103 and S104, the scenario generator creates a first boundary scenario test case based on the logical scenario model and the first scenario variable value. The autonomous driving simulation system tests the first boundary scenario test case to generate the interaction state between the autonomous driving vehicle and the environment, i.e., the first autonomous driving simulation data. Based on the first autonomous driving simulation data and the first scenario variable value, the initial safety boundary model is optimized for the first time to generate a first optimized safety boundary model. The first optimized safety boundary model is subjected to a reliability assessment; if the reliability assessment result indicates that the first optimized safety boundary model meets the preset conditions, the safety boundary search process of the first optimized safety boundary model on the scene variable range is terminated to generate a quasi-safety boundary model; if the reliability assessment result indicates that the first optimized safety boundary model does not meet the preset conditions, the first optimized safety boundary model is used to perform a safety boundary search process on the scene variable range until a quasi-safety boundary model is generated before the safety boundary search process is terminated.

[0036] Based on the first autonomous driving simulation data, the driving state of the autonomous driving vehicle in the first boundary scenario test case is evaluated using a state evaluation classifier to generate a driving state evaluation result; based on the driving state evaluation result and the first scenario variable value, the initial safety boundary model is optimized to generate a first optimized safety boundary model. Determine whether the characteristic parameters of the first optimized safety boundary model reach a preset threshold; if so, end the safety boundary search process and generate a quasi-safety boundary model. The driving state evaluation results include: state evaluation results such as collision, no collision, dangerous or not dangerous.

[0037] If not, the first optimized safety boundary model and the scene variable range are input into the Gaussian process searcher, a safety boundary search process is performed, and a second scene variable value is output. The scene generator creates a second boundary scene test case based on the logical scene model and the second scene variable value. The autonomous driving simulation system tests the second boundary scene test case and generates second autonomous driving simulation data. Based on the second autonomous driving simulation data, the state evaluation classifier is used to evaluate the driving state of the autonomous driving vehicle in the second boundary scene test case to generate a driving state evaluation result; based on the driving state evaluation result and the second scene variable value, the first optimized safety boundary model is optimized to generate a second optimized safety boundary model. Determine whether the characteristic parameters of the second optimized safety boundary model reach the preset threshold; if so, the safety boundary search process is terminated to generate a quasi-safety boundary model.

[0038] If not, the second optimized safety boundary model and the scene variable range are input into the Gaussian process searcher to perform safety boundary search processing until the characteristic parameters of the generated nth optimized safety boundary model reach the preset threshold, and then a quasi-safety boundary model is generated.

[0039] This embodiment uses the simulation environment to perform parameterized modeling on the test case, and verifies the effectiveness of the quasi-boundary scene model generated by the simulation environment by comparing it with the actual measurement data. This parameterized simulation method driven by multi-point actual data not only improves the accuracy of the simulation test, but also provides strong support for multi-point data sampling of the autonomous driving system.

[0040] In a preferred implementation of this embodiment, a Gaussian process-based searcher uses the first optimized safety boundary model to perform safety boundary search processing on the scene variable range to generate a second scene variable value; based on the second scene variable value and the logical scene model, a second boundary scene test case is created; the second boundary scene test case is tested using an autonomous driving simulation device to generate second autonomous driving simulation data; based on the second autonomous driving simulation data and the second scene variable value, the first optimized boundary scene model is optimized using a Gaussian process safety scenario predictor; and the generated second optimized safety boundary model is reliability evaluated using a reliability evaluator; until the reliability evaluation result indicates that the optimized safety boundary model meets the reliability requirements, a quasi-safety boundary model is generated and the safety boundary search processing is terminated.

[0041] Therefore, the iterative exit condition based on the predicted probability distribution enables the Gaussian algorithm to automatically stop when the preset accuracy is reached. This exit condition not only ensures that the quasi-safe boundary model is closer to the real boundary, but also avoids unnecessary iterative calculations, reducing computing costs and time.

[0042] In a preferred implementation of this embodiment, the quasi-safety boundary model and the scene variable joint distribution are input into the autonomous driving vehicle safety assessor for safety assessment, and a safety assessment report is output.

[0043] Therefore, this embodiment evaluates the safety of the autonomous driving system based on the quasi-safety boundary model generated by the simulation environment and the joint distribution of scene variables in the real scene.

[0044] Quasi-safety scenarios are the basis for safety verification and evaluation. Through simulation, test scenarios containing dangerous elements such as emergency obstacle avoidance, vehicle loss of control, and system failure can be generated to fully verify the emergency response capabilities of the autonomous driving system. The conventional algorithm is that the autonomous driving system tests all safety scenarios to determine the probability of an accident occurring in the autonomous driving system under each safety scenario. Several safety scenarios will be generated for the same safety scenario combination. If road testing is performed on each safety scenario based on an autonomous driving vehicle, the test mileage of the autonomous driving vehicle will increase unbearable, resulting in test results that may be unrealistic or irrelevant.

[0045] The following is a detailed description of the method for evaluating the safety of autonomous driving based on a simulation environment in combination with specific application scenarios.

[0046] The method for evaluating the safety of autonomous driving based on a simulation environment includes at least the following steps:

[0047] S1, obtain the environmental perception data of the autonomous driving vehicle; perform object extraction processing on the environmental perception data to generate a dynamic and static object extraction table; perform scene recognition processing on the dynamic and static object extraction table to generate a scene-related data stream; use the scene-related data stream to perform scene modeling processing, and output the scene variable range and the logical scene model corresponding to the scene variable range.

[0048] S2, using the initial safety boundary model to perform safety boundary search processing on the scene variable range to generate a first scene variable value.

[0049] S3, using an autonomous driving simulation device to test a first boundary scenario test case created by a first scenario variable value and a logical scenario model to generate first autonomous driving simulation data.

[0050] S4, based on the first autonomous driving simulation data, evaluate the driving state of the autonomous driving vehicle in the first boundary scenario test case to generate a driving state evaluation result; based on the driving state evaluation result and the first scenario variable value, perform a first optimization processing on the initial safety boundary model to generate a first optimized safety boundary model.

[0051] S5, perform reliability evaluation on the first optimized safety boundary model; if the reliability evaluation result indicates that the first optimized safety boundary model meets the preset conditions, then terminate the safety boundary search process of the first optimized safety boundary model on the scene variable range and generate a quasi-safety boundary model; if the reliability evaluation result indicates that the first optimized safety boundary model does not meet the preset conditions, then based on the Gaussian process searcher, perform safety boundary search process on the scene variable range using the first optimized safety boundary model to generate a second scene variable value; based on the second scene variable value and the logical scene model, create a second boundary scene test case; use the autonomous driving simulation device to test the second boundary scene test case to generate second autonomous driving simulation data; based on the second autonomous driving simulation data and the second scene variable value, use the Gaussian process safety scene predictor to optimize the first optimized boundary scene model; and use the reliability evaluator to perform reliability evaluation on the generated second optimized safety boundary model; until the reliability evaluation result indicates that the optimized safety boundary model meets the preset conditions, a quasi-safety boundary model is generated and the safety boundary search process is terminated.

[0052] S6, based on the scene variable joint distribution algorithm, performs joint distribution calculation on the scene variable range to generate the scene variable joint distribution; based on the quasi-safety boundary model and the scene variable joint distribution, evaluates the safety of the autonomous driving vehicle and generates a safety evaluation report.

[0053] Exemplarily, the safety boundary model is a prediction algorithm to predict the safety boundary in the autonomous driving test scenario. The safety boundary is a specific variable value combination of multiple scenario variables. In one dimension (i.e., when there is only one scenario variable), it is a point, in two dimensions it is a curve, in three dimensions it is a surface, and so on. Gaussian processes or vector machines are used to estimate the probability that a scenario test case based on a specific variable value combination is a safe scenario, as well as the reliability of this estimate.

[0054] S61, initial safety margin model: prepare an initial input (training data) for the model based on the scenario variables (generally physical quantities) and the scenario variable range, that is, randomly select a combination of several sets of scenario variable values ​​and obtain their test results through simulation or other tests. At the same time, select the safety margin confidence threshold (the threshold is a percentage range).

[0055] S62, train the initial safety boundary model based on the training data, and generate safety assessment predictions for all variable combinations (i.e., the probability of being safe), as well as a reliability factor of the prediction result, with a value range of 0 to 1. The larger the value, the higher the prediction reliability.

[0056] S63, generates a safety boundary (a combination of all variable values ​​that separate safe and unsafe scenarios) based on the probability that the prediction result generated in step S62 is safe, and the safety boundary confidence threshold. The safety boundary confidence threshold is an adjustable variable. It affects the aggressiveness (convergence speed) of the entire prediction algorithm. In tests based on simulation environments, the algorithm can be adjusted to be more aggressive (more unsafe combinations of variable values ​​to be tested may appear), while if real vehicle tests are introduced, the algorithm will generally be adjusted to be more conservative (more safe combinations of variable values ​​to be tested appear).

[0057] S64, the safety boundary generated in step S63 selects a combination of variable values ​​of a safety boundary scenario to be verified according to the reliability factor.

[0058] S65, generate scenario test cases based on the combination of variable values, and input the simulation software test to obtain the safety of the safety scenario and the autonomous driving vehicle to be tested (i.e., whether it is safe). When the algorithm is more aggressive in adjusting the safety boundary threshold, there will be more unsafe variable value combinations, i.e., on the unsafe side of the safety boundary.

[0059] S66, generating safety assessment predictions for all variable combinations based on the safety boundary model, that is, the probability of being safe (value range 0 to 1), and generating its statistical distribution.

[0060] S67, the reliability evaluator evaluates the reliability of the safety margin model according to the statistical distribution of the prediction result. When the statistical distribution meets the preset threshold, the reliability of the safety margin model meets the requirements.

[0061] S68, the reliability evaluator will also evaluate the stability of the current safety boundary model based on the past N times and the current evaluation results of the safety boundary model to avoid the shock effect in the safety boundary model. When the average value of N+1 times is less than the preset threshold, the stability of the current safety boundary model meets the requirements.

[0062] S69, when the confidence and stability of the safety boundary model meet the requirements, the safety boundary model can be used for autonomous driving safety assessment, and the safety boundary model update is terminated. If the confidence or stability of the safety boundary model does not meet the requirements, it is required to add the test results to the training set and return to step S62.

[0063] Therefore, this embodiment optimizes and improves the safety margin model through a large amount of real driving data.

[0064] For example: an autonomous vehicle travels 10,000 kilometers on the road, and the environmental perception module of the autonomous vehicle collects a large amount of environmental perception data; the environmental perception data is input into the scene recognition module for recognition, and the scene variables corresponding to several vehicle lane change scenarios are output. For example, the vehicle lane change scenario is: the vehicle in the left lane changes lanes in front of the autonomous vehicle; the scene variables are: the distance between the lane-changing vehicle and the autonomous vehicle, the lane-changing speed, and the time taken to change lanes. The scene modeling and scene joint distribution modules are used to model and process several vehicle lane change scenarios in the environmental perception data, and output the logical scene model, scene variable range, and scene variable joint distribution. The logical scene model is used to indicate the movement trajectory of the autonomous vehicle at different times, and the probability of a specific combination occurring at the same time. The scene variable range is, for example: the distance between the lane-changing vehicle and the autonomous vehicle is 0-50m, the lane-changing speed is 0-120 kilometers / hour, and the lateral acceleration is 0-2.5m / s 2 After the Gaussian process searcher is initialized, the distance between the lane-changing vehicle and the autonomous vehicle is 20m, the lane-changing speed is 60km / h, and the lateral acceleration is 1.1m / s 2 ” combination into the scenario generator to generate test cases. Use the simulated autonomous driving system to test the test cases and output autonomous driving simulation data. For example, autonomous driving simulation data: timely braking instructions to avoid rear-end collisions; if the brakes are not effective, a rear-end collision may occur. The autonomous driving simulation data is input into the state evaluation classifier for state evaluation, where a rear-end collision is 1 and no rear-end collision is 0; then the driving state evaluation results are compared with "the distance between the lane-changing vehicle and the autonomous driving vehicle is 20m, the lane-changing speed is 60km / h, and the lateral acceleration is 1.1m / s 2 "The combination is put into the initial safety boundary model for prediction; and the initial safety boundary model is updated based on the prediction results. Then the next variable combination is selected from the scene variable range; the next variable combination is, for example, the distance between the lane-changing vehicle and the autonomous driving vehicle is 9m, the lane-changing speed is 65km / h, and the lateral acceleration is 1.2m / s 2 . Even if the distance is shorter, a collision still occurs. The three values ​​construct a spatial cube, and the boundary is the surface. The surface distinguishes between collision and non-collision. Until the reliability evaluator indicator is met, it means that the safety boundary model is reliable, that is, a quasi-safety boundary model is obtained. At this time, the obtained safety boundary model and the real model are infinitely close to the same, so the above cycle ends. Finally, based on the joint distribution of the quasi-safety boundary model and the scenario variables, the probability of rear-end collision of the autonomous driving system under the quasi-safety boundary model is determined, so as to evaluate the safety of the autonomous driving system.

[0065] This embodiment uses a safety boundary model to perform safety boundary search processing on the environmental perception data of autonomous driving, so as to efficiently identify the safety boundary of the autonomous driving system under different working conditions; then, the test cases that form the safety boundary are tested in a simulation environment, and then the safety boundary model is reversely adjusted based on the autonomous driving simulation data generated by the test, until the safety boundary model obtained through the safety boundary search processing meets the reliability requirements and a quasi-safety boundary model is obtained; thereby, not only the test amount of the autonomous driving system in real scenarios is reduced, but also the test cost and test time of the autonomous driving system are reduced, and the accuracy and robustness of the safety assessment of the autonomous driving system are improved.

[0066] like Figure 2 As shown, it is a schematic diagram of the structure of a device for generating an autonomous driving safety boundary model based on a simulation environment provided by one embodiment of the present invention.

[0067] A device for generating an autonomous driving safety boundary model based on a simulation environment, the device 200 comprising: an acquisition module 201, for acquiring a scene variable range corresponding to environmental perception data of an autonomous driving vehicle, and a logical scene model corresponding to the scene variable range; a search processing module 202, for performing a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value; a test module 203, for testing a first boundary scene test case created by the first scene variable value and the logical scene model using an autonomous driving simulation device to generate first autonomous driving simulation data; an optimization processing module 204, for optimizing the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value until the optimized safety boundary model meets a preset condition, then terminating the safety boundary search process to generate a quasi-safety boundary model.

[0068] In a preferred implementation of this embodiment, the acquisition module includes: an acquisition unit, which is used to acquire the environmental perception data of the autonomous driving vehicle; an extraction unit, which is used to perform object extraction processing on the environmental perception data to generate a dynamic and static object extraction table; a scene recognition processing unit, which is used to perform scene recognition processing on the dynamic and static object extraction table to generate a scene-related data stream; and a scene modeling processing unit, which is used to perform scene modeling processing using the scene-related data stream, and output a scene variable range and a logical scene model corresponding to the scene variable range.

[0069] In a preferred implementation manner of this embodiment, the optimization processing module includes: an optimization processing unit, which is used to perform a first optimization processing on the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value; generate a first optimized safety boundary model; a reliability evaluation unit, which is used to perform reliability evaluation on the first optimized safety boundary model; a first generation unit, which is used to terminate the safety boundary search processing of the scene variable range by the first optimized safety boundary model if the reliability evaluation result indicates that the first optimized safety boundary model meets the preset conditions, and generate a quasi-safety boundary model; a second generation unit, which is used to perform safety boundary search processing on the scene variable range using the first optimized safety boundary model if the reliability evaluation result indicates that the first optimized safety boundary model does not meet the preset conditions, and terminate the safety boundary search processing until a quasi-safety boundary model is generated.

[0070] In a preferred implementation manner of this embodiment, the optimization processing unit includes: an evaluation subunit, used to evaluate the driving state of the autonomous driving vehicle in the first boundary scenario test case based on the first autonomous driving simulation data, and generate a driving state evaluation result; a generation subunit, used to perform a first optimization processing on the initial safety boundary model based on the driving state evaluation result and the first scene variable value, and generate a first optimized safety boundary model.

[0071] In a preferred implementation manner of this embodiment, the device also includes: a joint distribution calculation module, used to perform joint distribution calculation on the scene variable range to generate a joint distribution of scene variables; a generation module, used to evaluate the safety of the autonomous driving vehicle based on the quasi-safety boundary model and the joint distribution of the scene variables, and generate a safety assessment report.

[0072] In a preferred implementation manner of this embodiment, the second generation unit includes: a first generation subunit, which is used to perform a safety boundary search process on the scene variable range based on a Gaussian process searcher using the first optimized safety boundary model to generate a second scene variable value; a creation subunit, which is used to create a second boundary scene test case based on the second scene variable value and the logical scene model; a second generation subunit, which is used to test the second boundary scene test case using an autonomous driving simulation device to generate second autonomous driving simulation data; a third generation subunit, which is used to optimize the first optimized boundary scene model based on the second autonomous driving simulation data and the second scene variable value using a Gaussian process safety scene predictor; and use a reliability evaluator to perform reliability evaluation on the generated second optimized safety boundary model; until the reliability evaluation result indicates that the optimized safety boundary model meets the preset conditions, a quasi-safety boundary model is generated and the safety boundary search process is terminated.

[0073] The above device can execute the method for generating an autonomous driving safety boundary model based on a simulation environment provided by an embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method for generating an autonomous driving safety boundary model based on a simulation environment. For technical details not fully described in this embodiment, please refer to the method for generating an autonomous driving safety boundary model based on a simulation environment provided by an embodiment of the present invention.

[0074] The present invention also provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the method of generating an autonomous driving safety boundary model based on a simulation environment described in the present invention.

[0075] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0076] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0077] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the method according to the following embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0078] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0079] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0080] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0081] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0082] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0083] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

[0084] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0085] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0086] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for generating an autonomous driving safety boundary model based on a simulation environment, characterized in that: include: Obtaining a scene variable range corresponding to the environmental perception data of the autonomous driving vehicle and a logical scene model corresponding to the scene variable range; Performing a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value; Using an autonomous driving simulation device to test a first boundary scenario test case created by the first scenario variable value and the logic scenario model to generate first autonomous driving simulation data; Based on the first autonomous driving simulation data and the first scenario variable value, the initial safety boundary model is optimized until the optimized safety boundary model meets the preset conditions, then the safety boundary search process is terminated to generate a quasi-safety boundary model.

2. The method according to claim 1, characterized in that The step of obtaining the scene variable range corresponding to the environmental perception data of the autonomous driving vehicle and the logical scene model corresponding to the scene variable range comprises: Acquiring environmental perception data of the autonomous driving vehicle; Performing object extraction processing on the environmental perception data to generate a dynamic and static object extraction table; Performing scene recognition processing on the dynamic and static object extraction table to generate a scene-related data stream; The scene-related data stream is used to perform scene modeling processing, and a scene variable range and a logical scene model corresponding to the scene variable range are output.

3. The method according to claim 1, characterized in that The method includes optimizing the initial safety boundary model based on the first autonomous driving simulation data and the first scenario variable value until the optimized safety boundary model meets a preset condition, then terminating the safety boundary search process and generating a quasi-safety boundary model; comprising: Based on the first autonomous driving simulation data and the first scenario variable value, performing a first optimization process on the initial safety boundary model to generate a first optimized safety boundary model; Performing reliability evaluation on the first optimized safety boundary model; If the reliability evaluation result indicates that the first optimized safety boundary model meets the preset condition, then the safety boundary search process of the first optimized safety boundary model on the scene variable range is terminated to generate a quasi-safety boundary model; If the reliability assessment result indicates that the first optimized safety boundary model does not meet the preset conditions, the first optimized safety boundary model is used to perform a safety boundary search process on the scene variable range until a quasi-safety boundary model is generated and the safety boundary search process is terminated.

4. The method according to claim 3, characterized in that performing a first optimization process on the initial safety boundary model based on the first autonomous driving simulation data and the first scenario variable value; Generate a first optimized safety boundary model; including: Based on the first autonomous driving simulation data, evaluating the driving state of the autonomous driving vehicle in the first boundary scenario test case to generate a driving state evaluation result; Based on the driving state evaluation result and the first scenario variable value, the initial safety margin model is optimized for the first time to generate a first optimized safety margin model.

5. The method according to claim 1, characterized in that Also includes: Performing joint distribution calculation on the scene variable range to generate a scene variable joint distribution; Based on the quasi-safety boundary model and the joint distribution of the scenario variables, the safety of the autonomous driving vehicle is evaluated and a safety evaluation report is generated.

6. The method according to claim 3, characterized in that: The step of performing a safety boundary search process on the scene variable range by using the first optimized safety boundary model until a quasi-safety boundary model is generated and then terminating the safety boundary search process comprises: A Gaussian process-based searcher uses the first optimized safety boundary model to perform safety boundary search processing on the scene variable range to generate a second scene variable value; Creating a second boundary scenario test case based on the second scenario variable value and the logical scenario model; Testing the second boundary scenario test case using an autonomous driving simulation device to generate second autonomous driving simulation data; Based on the second autonomous driving simulation data and the second scenario variable value, the first optimized boundary scenario model is optimized using a Gaussian process safety scenario predictor; and the generated second optimized safety boundary model is reliability evaluated using a reliability evaluator; until the reliability evaluation result indicates that the optimized safety boundary model meets the preset conditions, a quasi-safety boundary model is generated and the safety boundary search process is terminated.

7. A device for generating an autonomous driving safety boundary model based on a simulation environment, characterized in that: include: An acquisition module, used to acquire a scene variable range corresponding to the environmental perception data of the autonomous driving vehicle, and a logical scene model corresponding to the scene variable range; A search processing module, configured to perform a safety boundary search process on the scene variable range using an initial safety boundary model to generate a first scene variable value; A testing module, configured to test a first boundary scenario test case created by the first scenario variable value and the logic scenario model using an autonomous driving simulation device to generate first autonomous driving simulation data; The optimization processing module is used to optimize the initial safety boundary model based on the first autonomous driving simulation data and the first scene variable value until the optimized safety boundary model meets the preset conditions, then terminate the safety boundary search processing and generate a quasi-safety boundary model.

8. The device according to claim 6, characterized in that The optimization processing module includes: an optimization processing unit, configured to perform a first optimization process on the initial safety boundary model based on the first autonomous driving simulation data and the first scenario variable value, and generate a first optimized safety boundary model; A reliability evaluation unit, configured to perform reliability evaluation on the first optimized safety margin model; a generating unit, configured to terminate the security boundary search process of the first optimized security boundary model on the scene variable range and generate a quasi-security boundary model if the reliability evaluation result indicates that the first optimized security boundary model satisfies a preset condition; A safety boundary search processing unit is used to use the first optimized safety boundary model to perform safety boundary search processing on the scene variable range if the reliability assessment result indicates that the first optimized safety boundary model does not meet the preset conditions, and the safety boundary search processing is terminated only after a quasi-safety boundary model is generated.

9. An electronic device, comprising: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method according to any one of claims 1 to 6.

10. A computer readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images