Automated vehicle performance boundary mining method, system, device, and medium
By combining a dual-objective monkey swarm random agent optimization algorithm with virtual simulation testing, the five elements representing autonomous driving traffic scenarios are decoupled, solving the problem of difficulty in mining the performance boundary of autonomous vehicles in high-dimensional space, and realizing fast and accurate performance boundary mining and safety performance improvement.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-03-24
AI Technical Summary
Existing technologies struggle to efficiently uncover the performance boundaries of autonomous vehicles, especially in high-dimensional spaces where it is difficult to accurately characterize and search for global or local optimal solutions, resulting in insufficient safety performance of autonomous driving systems.
A dual-objective monkey swarm random agent optimization algorithm combined with virtual simulation testing was adopted to decouple the five elements representing the traffic simulation scenario. Key parameters were screened through sensitivity analysis, a dual-objective random simulation optimization model was established, and the performance boundary was finally determined by using an autonomous driving simulation platform for evaluation.
This method rapidly uncovers the performance boundaries of autonomous vehicles with a limited number of simulation evaluations, reduces computational costs, and improves the safety performance and algorithm improvement efficiency of autonomous driving systems.
Smart Images
Figure CN121302936B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of autonomous driving technology, and in particular to a method, system, device, and medium for mining the performance boundaries of autonomous vehicles. Background Technology
[0002] Current autonomous driving systems suffer from insufficient safety performance and frequent accidents. As a tightly coupled system involving humans, vehicles, roads, environment, and tasks, the testing of autonomous vehicles encompasses multiple modules, including perception, localization, mapping, decision-making and planning, and control execution, as well as the overall system performance. Designing complex and hazardous scenarios and tasks to verify their reliability and effectiveness is a common challenge faced by scientists and engineers worldwide. To establish a comprehensive testing and verification platform, the International Automobile Manufacturers Association (IAMA) has proposed a "three-pillar" certification approach, combining virtual simulation, closed-track testing, and real-world road testing. Due to its advantages such as high efficiency, low cost, process safety, and reproducible scenarios, virtual simulation testing will account for 90% of the testing content for autonomous vehicles.
[0003] Meanwhile, due to technological and cost limitations, full-condition, all-weather autonomous driving has not yet been realized, indicating that autonomous vehicles have performance boundaries. Accurately grasping the performance boundaries of current autonomous driving systems has significant theoretical and practical economic value. On the one hand, by exploring the performance boundaries of autonomous driving systems, we can identify the safe and dangerous operating domains of current autonomous vehicles and comprehensively compare the advantages and disadvantages of autonomous vehicles on the market. On the other hand, based on the theoretical performance boundaries of autonomous driving systems, we can generate a massive number of dangerous scenarios in a targeted manner, thereby guiding and improving autonomous driving algorithms, accurately fixing the defects of autonomous driving systems, and expanding their safe operating domains in certain scenarios.
[0004] The performance boundary is characterized by the key parameter matrix or vector of the critical hazardous scenarios that the autonomous driving system can cope with. Autonomous driving safety simulation testing involves traffic scenarios covering parameters of five related elements: people, vehicles, roads, environment, and tasks. As the dimensionality of parameters increases, the complexity of the performance boundary representation may grow exponentially. Its shape and structure in high-dimensional space may be nonlinear, non-convex, multimodal, or even isolated. Furthermore, the diverse traffic simulation test scenarios lead to varied performance boundaries for autonomous vehicles. Moreover, autonomous driving virtual simulation has characteristics such as randomness and high computational cost, making it difficult for traditional global optimization algorithms to efficiently search for global or even local optima, thus hindering the accurate discovery of the performance boundary of autonomous vehicles. Summary of the Invention
[0005] To address the shortcomings of the existing technologies, this invention provides a method, system, device, and medium for mining the performance boundaries of autonomous vehicles. By combining virtual simulation testing and proxy optimization techniques, it achieves accurate mining of the performance boundaries of autonomous vehicles.
[0006] Firstly, a method for mining the performance boundaries of autonomous vehicles is provided, including the following steps:
[0007] Decouple the five elements representing autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks;
[0008] The five elements of people, vehicles, roads, environment and tasks are parameterized to be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios.
[0009] Sensitivity analysis was used to identify key parameter sets for five categories of related elements: people, vehicles, roads, environment, and tasks.
[0010] Using the selected set of key parameters as decision variables, traffic flow uncertainty and random dangerous driving behavior of background vehicles as disturbance factors, an autonomous driving simulation platform as evaluation tool, and minimizing the absolute value of the speed difference and distance between the tested autonomous vehicle and surrounding dynamic traffic participants and static objects as optimization objectives, a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining is established.
[0011] A bi-objective monkey swarm random agent optimization algorithm is used to solve the bi-objective stochastic simulation optimization model, and the Pareto stochastic optimal key parameter solution set is obtained.
[0012] Based on the Pareto-optimal key parameter solution set, the critical scenarios that the autonomous vehicle can cope with are determined, and the performance boundary of the autonomous vehicle under test is obtained.
[0013] Furthermore, the key parameter sets for the five related elements—people, vehicles, roads, environment, and tasks—include:
[0014] Key parameters related to people include travel route, departure time, and motion attributes; key parameters related to vehicles include initial relative position, initial relative speed, and start time; key parameters related to roads include road type, road channelization conditions, and intersection type; key parameters related to the environment include weather, light, and occlusion; and key parameters related to tasks include route selection, expected speed, and arrival time.
[0015] Furthermore, the process of solving the bi-objective stochastic simulation optimization model using the bi-objective monkey swarm stochastic agent optimization algorithm includes:
[0016] Using the monkey's position Characterize the key parameter set and randomly generate the initial monkey troop location set. And use an autonomous driving simulation platform to set the initial monkey troop location set. A realistic assessment was conducted to obtain a set of evaluation values for both the absolute value of the velocity difference and the distance between the two targets. and ;based on and Establish a regression kriging model for the absolute value of the speed difference. ,based on and Establish a regression kriging model for the spacing. ,use and The location of the monkeys is evaluated; then, the monkey troop is iteratively subjected to climbing, look-jump, and somersault processes to refine the troop's location set. and Perform updates; when the termination condition is met, output the set of all actually evaluated monkey troop locations and the final updated set. and and utilize the final update and The sets of monkey troop locations that were actually evaluated were re-evaluated to obtain the Pareto-random optimal set of key parameters.
[0017] Furthermore, the crawling process includes:
[0018] For each monkey, a single climbing operation includes:
[0019] use Represents the position of the i-th monkey, and a vector is randomly generated. , , Let n be the displacement of the i-th monkey in the j-th dimension, where n is the dimension number of the monkey's position; combined with and judge and Percentile dominance relationships include: Case (a). Percentile dominance in the j-th dimension ; Situation (b) Percentile dominance in the j-th dimension ; Case (c), and Between the first There is no percentile dominance relationship in terms of dimension;
[0020] based on and The percentile dominance relationship between them is updated to update the i-th monkey in the th position. Position in dimension: In case (a), In case (b), In case (c), set with a certain probability. ;in, For the updated i-th monkey in the i-th position Position in dimensions;
[0021] Perform the above operation on each dimension of the monkey's position to obtain the updated monkey position. ;if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, remain unchanged;
[0022] Repeat the above climbing operation for each monkey until all dimensions of the monkey's position are correct. and The percentile dominance relationship between them exceeds the preset proportion or is all of case (b) or case (c), or the maximum number of climbs allowed has been reached.
[0023] Furthermore, the look-jump process includes:
[0024] For each monkey, in each dimension of its position, within the interval Generate random numbers respectively Gain a new position to observe. , Characterized by the monkey's field of vision;
[0025] judge and Percentile dominance, if percentile dominance or and There is no percentile dominance relationship between them, and In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat the above steps until a suitable one is found. Or reach a certain number of observations;
[0026] For each monkey, repeat the crawling process starting from the updated position until the crawling process is completed;
[0027] Obtain and evaluate the latest set of monkey troop locations, and then... and Update.
[0028] Furthermore, the latest set of monkey troop locations is obtained and evaluated, thereby enabling [further action / action]. and The update includes:
[0029] Get the latest set of monkey troop locations Then update using any of the following strategies. and :
[0030] Strategy 1: Utilize an autonomous driving simulation platform to obtain the latest monkey troop location data. A realistic assessment was conducted to obtain a set of evaluation values for both the absolute value of the velocity difference and the distance between the two targets. and ;use , and renew and ;
[0031] Strategy 2: Adopt the approach from the previous iteration and Determine the latest set of monkey troop locations The percentile dominance relationship between each pair of solutions is used to select the solution that has the largest number of percentile dominance relationships with other solutions. The optimal monkey position is determined, and a realistic evaluation is performed using an autonomous driving simulation platform. The optimal monkey position and the corresponding realistic evaluation results are then used to update the system. and .
[0032] Furthermore, the somersault process includes:
[0033] Based on the current positions of M monkeys: , Each monkey performs the following somersault:
[0034] Step 1: Based on the monkey's location, within the interval This interval This is known as the monkey's somersault distance;
[0035] Step 2: Calculate the position of the i-th monkey after performing a somersault in the j-th dimension: , Indicates the monkey troop in the th... Geometric center in a dimension;
[0036] Step 3: After the somersault, if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat steps 1 and 2 until a feasible solution is found. .
[0037] Secondly, a system for exploring the performance boundaries of autonomous vehicles is provided, including:
[0038] The element decoupling module is used to decouple five types of elements that characterize autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks.
[0039] The parameterization module is used to parameterize five types of elements: people, vehicles, roads, environment, and tasks, so that they can be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios.
[0040] The key parameter screening module is used to screen out key parameter sets of five related elements: people, vehicles, roads, environment, and tasks through sensitivity analysis.
[0041] The optimization model building module is used to establish a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining. The selected key parameter set is used as decision variables, traffic flow uncertainty and background vehicle random dangerous driving behavior are used as disturbance factors, autonomous driving simulation platform is used as evaluation tool, and the absolute value of the speed difference and distance between the tested autonomous vehicle and the surrounding dynamic traffic participants and static objects is minimized as the optimization objective.
[0042] The model solving module is used to solve the bi-objective stochastic simulation optimization model using the bi-objective monkey swarm random agent optimization algorithm, and obtain the Pareto stochastic optimal key parameter solution set.
[0043] The performance boundary output module is used to determine the critical scenarios that the autonomous vehicle can cope with based on the Pareto random optimal key parameter solution set, and to derive the performance boundary of the autonomous vehicle under test.
[0044] Thirdly, an electronic device is provided, comprising:
[0045] A memory on which computer programs are stored;
[0046] A processor is used to load and execute the computer program to implement the previously described method for mining the performance boundaries of autonomous vehicles.
[0047] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the aforementioned method for mining the performance boundaries of autonomous vehicles.
[0048] This invention proposes a method, system, device, and medium for mining the performance boundary of autonomous vehicles, which has the following beneficial effects: As the dimensionality of parameters increases, the complexity of the performance boundary of autonomous vehicles grows exponentially; diverse traffic simulation test scenarios and traffic uncertainties pose significant challenges to the mining of the performance boundary of autonomous vehicles; simultaneously, autonomous driving virtual simulation evaluation is characterized by high computational cost and being a black box, making it difficult for gradient-based global search algorithms to efficiently find global or even local optima. To address this high-dimensional, high-computational-cost, black-box, multi-objective uncertain simulation optimization problem, this invention uses key parameters to characterize traffic virtual simulation scenarios, establishes a bi-objective stochastic simulation optimization model for mining the performance boundary of autonomous vehicles, and designs a bi-objective monkey swarm random agent optimization algorithm for efficient solution. This allows for rapid mining of the performance boundary of autonomous vehicles under limited simulation evaluation constraints, contributing to accelerating product iteration and updates in the autonomous vehicle industry. Attached Figure Description
[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0050] Figure 1 This is a flowchart of the autonomous vehicle performance boundary mining method provided in an embodiment of the present invention. Detailed Implementation
[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0052] like Figure 1 As shown, this embodiment of the invention provides a method for mining the performance boundary of an autonomous vehicle, including the following steps:
[0053] S1: Decouple the five elements of autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks.
[0054] Specifically, parameters include: 1) Person-related parameters, such as travel route, departure time, and motion attributes; 2) Vehicle-related parameters, such as initial relative position, initial relative speed, and start time; 3) Road-related parameters, such as road type, road channelization, and intersection type; 4) Environmental-related parameters, such as weather, lighting, and occlusion; and 5) Task-related parameters, such as route selection, expected speed, and arrival time.
[0055] S2: It parametrically represents five elements: people, vehicles, roads, environment, and tasks, and makes them into the OpenX series standard format so that they can be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios.
[0056] S3: Through sensitivity analysis, key parameter sets of five related elements—people, vehicles, roads, environment, and tasks—are selected.
[0057] Sensitivity analysis methods can include single-factor variation method, Morton index method, Fourier amplitude sensitivity test, etc. In this embodiment, the key parameters of the five categories of related elements—people, vehicles, roads, environment, and tasks—are as follows: key parameters related to people include travel route, departure time, and motion attributes; key parameters related to vehicles include initial relative position, initial relative speed, and start time; key parameters related to roads include road type, road channelization conditions, and intersection type; key parameters related to the environment include weather, light, and occlusion; and key parameters related to tasks include route selection, expected speed, and arrival time.
[0058] S4: Using the selected set of key parameters as decision variables, traffic flow uncertainty and random dangerous driving behavior of background vehicles as disturbance factors, an autonomous driving simulation platform as the evaluation tool, and minimizing the absolute value of the speed difference and distance between the tested autonomous vehicle and surrounding dynamic traffic participants and static objects as the optimization objective, a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining is established. The dual-objective stochastic simulation optimization model can be expressed as follows:
[0059] ;
[0060] ;
[0061] in, Describe the objective function. This represents the absolute value of the speed difference between the tested autonomous vehicle and surrounding dynamic traffic participants / static objects. This indicates the distance between the tested autonomous vehicle and surrounding dynamic traffic participants / static objects; Key parameters representing the five categories of related elements that constitute a segment of a traffic simulation scenario; Indicates key parameters related to people, This indicates key parameters related to the vehicle. Indicates key parameters related to the road. Indicates key environmental parameters, This indicates key parameters related to the task; Represent the solution space; This indicates uncertainties in traffic flow; This indicates random and dangerous driving behavior factors in the background vehicle; Represents the expectation operator; Indicates in and In the context of uncertainty in characterization, Traffic simulations are performed on autonomous vehicles, using key traffic scenario parameters as input. This indicates the output of the simulation evaluation results; and Indicates being , and The results of the evaluation of the critical performance of autonomous vehicles under the given input conditions.
[0062] S5: The bi-objective monkey swarm random agent optimization algorithm is used to solve the bi-objective stochastic simulation optimization model, obtaining the Pareto stochastic optimal key parameter solution set. The solution process of the bi-objective monkey swarm random agent optimization algorithm is as follows:
[0063] (1) Initialization:
[0064] Using the monkey's position Characterizing the key parameter set in the solution space Randomly generated The initial position of each monkey, and the position of any monkey can be expressed as: , , in The dimension representing the monkey positions is the dimension of the solution (key parameters). Initial monkey troop location set. Using autonomous driving simulation platforms (such as Carla) to study the location set of monkey groups Conduct a real-world assessment to obtain a set of evaluation values for the two objectives: , Number of updates and iterations .
[0065] based on and Establish a regression kriging model for the absolute value of the speed difference: ,in and Representing the model respectively In a certain solution Given the mean and variance, then in a certain solution... In the case of the first Percentile function ;based on and Establish a regression kriging model for the spacing: ,in and Representing the model respectively In a certain solution Given the mean and variance, then in a certain solution... In the case of the first Percentile function .here, Indicates a normal distribution. This represents the inverse cumulative distribution function of the standard normal distribution.
[0066] (2) Climbing process:
[0067] For each monkey, a single climbing operation includes:
[0068] use Represents the position of the i-th monkey, and a vector is randomly generated. , , Let be the displacement of the i-th monkey in the j-th dimension, where Set with equal probability or , This is called the stride length of a monkey climbing a mountain; [It is used to] determine... and In the Percentile function and Percentile dominance under dual-objective evaluation includes: Case (a), if ,and , Then it is called Percentile dominance in the j-th dimension , represented as ;in, express Evaluation indicators that pertain to the absolute value of the speed difference or the evaluation indicators for the distance. express or ; Case (b), if ,and , Then it is called Percentile dominance in the j-th dimension ; Case (c), if and If none of these conditions are met, then it means... and Between the first There is no percentile dominance relationship in terms of dimension;
[0069] based on and The percentile dominance relationship between them is updated to update the i-th monkey in the th position. Position in dimension: In case (a), In case (b), In case (c), set with a certain probability. (Maintaining the diversity of solutions); among which, For the updated i-th monkey in the i-th position Position in dimensions;
[0070] Perform the above operation on each dimension of the monkey's position to obtain the updated monkey position. ;if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, remain unchanged.
[0071] Repeat the above climbing operation for each monkey until all dimensions of the monkey's position are correct. and The percentile dominance relationship between them exceeds the preset ratio (such as 80%, 85% or 90%, etc., and the preset ratio can also be adjusted to the preset number of dimensions as needed) or all of them are cases (b) or cases (c), or the maximum number of climbing times is reached.
[0072] (3) The look-jump process:
[0073] For each monkey, in each dimension of its position, within the interval , Generate random numbers respectively Gain a new position to observe. ; This is characterized by the monkey's field of vision, that is, the maximum distance a monkey can observe.
[0074] judge and In the Percentile surrogate function and Percentile dominance under dual-objective evaluation, if percentile dominance and In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat the above steps until a suitable one is found. Or after a certain number of observations. The update / replacement condition here is set to: only percentile dominance , It was only replaced with To ensure better exploratory nature, the replacement condition can also be set as follows: percentile dominance or and There is no percentile dominance relationship between them.
[0075] For each monkey, repeat the crawling process starting from the updated position until the crawling process is completed. Obtain the latest set of monkey troop locations. And conduct an assessment, and then... and Update.
[0076] Specifically, the latest set of monkey troop locations is obtained and evaluated, and then... and The update includes:
[0077] Get the latest set of monkey troop locations Then update using any of the following strategies. and :
[0078] Strategy 1: Utilize an autonomous driving simulation platform to obtain the latest monkey troop location data. A realistic assessment was conducted to obtain a set of evaluation values for both the absolute value of the velocity difference and the distance between the two targets. and ;use , and renew and ;
[0079] Strategy 2: Adopt the approach from the previous iteration and Determine the latest set of monkey troop locations The percentile dominance relationship between each pair of solutions is used to select the solution that has the largest number of percentile dominance relationships with other solutions. The optimal monkey position is determined, and a realistic evaluation is performed using an autonomous driving simulation platform. The optimal monkey position and the corresponding realistic evaluation results are then used to update the system. and .
[0080] (4) Somersault process:
[0081] To explore new search areas, based on the current positions of M monkeys: , Each monkey performs the following somersault:
[0082] Step 1: Based on the monkey's location, within the interval This interval This is called the monkey's somersault distance, where c and d are the lower and upper limits of the monkey's somersault distance, respectively.
[0083] Step 2: Calculate the position of the i-th monkey after performing a somersault in the j-th dimension: , Indicates the monkey troop in the th... Geometric center in a dimension; The fulcrum used by the monkey troop to perform a somersault. Represents monkey In dimensions The direction of the somersault;
[0084] Step 3: After the somersault, if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat steps 1 and 2 until a feasible solution is found. .
[0085] (5) Output of the random Pareto optimal solution set:
[0086] Repeat steps (2) through (4) until the termination condition is met, outputting the set of all monkey troop locations that were actually evaluated and the final updated set. and and utilize the final update and The sets of monkey troop locations that were actually evaluated were re-evaluated to obtain the Pareto-random optimal set of key parameters.
[0087] S6: Based on the Pareto random optimal key parameter solution set, the critical scenarios that autonomous vehicles can cope with are determined. The performance domain of autonomous vehicles can be divided into the safety domain and the accident domain, thereby uncovering the performance boundary of the autonomous vehicle under test.
[0088] The above embodiments provide a method for mining the performance boundary of autonomous vehicles. This method uses key parameters to characterize the performance boundary of autonomous vehicles and combines virtual simulation testing with a bi-objective monkey swarm random agent optimization algorithm. This reduces computational costs while quickly finding the global optimum, thus accurately mining the performance boundary of autonomous vehicles.
[0089] This invention also provides a system for mining the performance boundaries of autonomous vehicles, comprising:
[0090] The element decoupling module is used to decouple five types of elements that characterize autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks.
[0091] The parameterization module is used to parameterize five types of elements: people, vehicles, roads, environment, and tasks, so that they can be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios.
[0092] The key parameter screening module is used to screen out key parameter sets of five related elements: people, vehicles, roads, environment, and tasks through sensitivity analysis.
[0093] The optimization model building module is used to establish a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining. The selected key parameter set is used as decision variables, traffic flow uncertainty and background vehicle random dangerous driving behavior are used as disturbance factors, autonomous driving simulation platform is used as evaluation tool, and the absolute value of the speed difference and distance between the tested autonomous vehicle and the surrounding dynamic traffic participants and static objects is minimized as the optimization objective.
[0094] The model solving module is used to solve the bi-objective stochastic simulation optimization model using the bi-objective monkey swarm random agent optimization algorithm, and obtain the Pareto stochastic optimal key parameter solution set.
[0095] The performance boundary output module is used to determine the critical scenarios that the autonomous vehicle can cope with based on the Pareto random optimal key parameter solution set, and to derive the performance boundary of the autonomous vehicle under test.
[0096] It should be understood that the functional unit modules in the various embodiments of the present invention can be concentrated in one processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated into one unit module, and can be implemented in hardware or software.
[0097] Furthermore, embodiments of the present invention also provide an electronic device, comprising:
[0098] A memory on which computer programs are stored;
[0099] A processor is used to load and execute the computer program to implement the previously described method for mining the performance boundaries of autonomous vehicles.
[0100] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned method for mining the performance boundaries of autonomous vehicles.
[0101] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.
[0102] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0103] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0106] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for mining the performance boundary of an autonomous vehicle, characterized in that, Includes the following steps: Decouple the five elements representing autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks; The five elements of people, vehicles, roads, environment and tasks are parameterized to be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios. Sensitivity analysis was used to identify key parameter sets for five categories of related elements: people, vehicles, roads, environment, and tasks. Using the selected set of key parameters as decision variables, traffic flow uncertainty and random dangerous driving behavior of background vehicles as disturbance factors, an autonomous driving simulation platform as an evaluation tool, and minimizing the absolute value of the speed difference and distance between the tested autonomous vehicle and surrounding dynamic traffic participants and static objects as optimization objectives, a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining is established. A bi-objective monkey swarm random agent optimization algorithm is used to solve the bi-objective stochastic simulation optimization model, and the Pareto stochastic optimal key parameter solution set is obtained. Based on the Pareto random optimal key parameter solution set, the critical scenarios that the autonomous vehicle can cope with are determined, and the performance boundary of the tested autonomous vehicle is obtained. The process of solving the bi-objective stochastic simulation optimization model using the bi-objective monkey swarm random agent optimization algorithm includes: Using the monkey's position Characterize the key parameter set and randomly generate the initial monkey troop location set. And use an autonomous driving simulation platform to set the initial monkey troop location set. A realistic assessment was conducted to obtain a set of evaluation values for both the absolute value of the velocity difference and the distance between the two targets. and ;based on and Establish a regression kriging model for the absolute value of the speed difference. ,based on and Establish a regression kriging model for the spacing. ,use and The location of the monkeys is evaluated; then, the monkey troop is iteratively subjected to climbing, look-jump, and somersault processes to refine the troop's location set. and Perform updates; when the termination condition is met, output the set of all actually evaluated monkey troop locations and the final updated set. and and utilize the final update and The sets of monkey troop locations that were actually evaluated were re-evaluated to obtain the Pareto-random optimal set of key parameters.
2. The method for mining the performance boundary of autonomous vehicles according to claim 1, characterized in that, The key parameter set for the five related elements—people, vehicles, roads, environment, and tasks—includes: Key parameters related to people include travel route, departure time, and motion attributes; key parameters related to vehicles include initial relative position, initial relative speed, and start time; key parameters related to roads include road type, road channelization conditions, and intersection type; key parameters related to the environment include weather, light, and occlusion; and key parameters related to tasks include route selection, expected speed, and arrival time.
3. The method for mining the performance boundary of autonomous vehicles according to claim 1, characterized in that, The crawling process includes: For each monkey, a single climbing operation includes: use Represents the position of the i-th monkey, and a vector is randomly generated. , , Let n be the displacement of the i-th monkey in the j-th dimension, where n is the dimension number of the monkey's position; combined with and judge and Percentile dominance relationships include: Case (a). Percentile dominance in the j-th dimension ; Situation (b) Percentile dominance in the j-th dimension ; Case (c), and Between the second There is no percentile dominance relationship in terms of dimension; based on and The percentile dominance relationship between them is updated to update the i-th monkey in the th position. Position in dimension: In case (a), In case (b), In case (c), set with a certain probability. ;in, For the updated i-th monkey in the i-th position Position in dimensions; Perform the above operation on each dimension of the monkey's position to obtain the updated monkey position. ;if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, remain unchanged; Repeat the above climbing operation for each monkey until all dimensions of the monkey's position are correct. and The percentile dominance relationship between them exceeds the preset proportion or is all of case (b) or case (c), or the maximum number of climbs allowed has been reached.
4. The method for mining the performance boundary of autonomous vehicles according to claim 3, characterized in that, The view-jump process includes: For each monkey, in each dimension of its position, within the interval Generate random numbers respectively Gain a new position to observe. , Characterized by the monkey's field of vision; judge and Percentile dominance, if percentile dominance or and There is no percentile dominance relationship between them, and In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat the above steps until a suitable one is found. Or reach a certain number of observations; For each monkey, repeat the crawling process starting from the updated position until the crawling process is completed; Obtain and evaluate the latest set of monkey troop locations, and then... and Update.
5. The method for mining the performance boundary of autonomous vehicles according to claim 4, characterized in that, Obtain and evaluate the latest set of monkey troop locations, and then... and The update includes: Get the latest set of monkey troop locations Then update using any of the following strategies. and : Strategy 1: Utilize an autonomous driving simulation platform to obtain the latest monkey troop location set. A realistic assessment was conducted to obtain a set of evaluation values for both the absolute value of the velocity difference and the distance between the two targets. and ;use , and renew and ; Strategy 2: Adopt the approach from the previous iteration and Determine the latest set of monkey troop locations The percentile dominance relationship between each pair of solutions is used to select the solution that has the largest number of percentile dominance relationships with other solutions. The optimal monkey position is determined, and a realistic evaluation is performed using an autonomous driving simulation platform. The optimal monkey position and the corresponding realistic evaluation results are then used to update the system. and .
6. The method for mining the performance boundary of autonomous vehicles according to claim 1, characterized in that, The somersault process includes: Based on the current positions of M monkeys: , Each monkey performs the following somersault: Step 1: Based on the monkey's location, within the interval This interval This is known as the monkey's somersault distance; Step 2: Calculate the position of the i-th monkey after performing a somersault in the j-th dimension: , This indicates that the monkey troop was in the [number]th [year]. Geometric center in a dimension; Step 3: After the somersault, if In solution space If it is feasible, then update the position of the i-th monkey: Otherwise, repeat steps 1 and 2 until a feasible solution is found. .
7. A system for uncovering the performance boundaries of autonomous vehicles, characterized in that, The method for performing the autonomous vehicle performance boundary mining method as described in any one of claims 1 to 6 includes: The element decoupling module is used to decouple five types of elements that characterize autonomous driving traffic simulation test scenarios: people, vehicles, roads, environment, and tasks. The parameterization module is used to parameterize five types of elements: people, vehicles, roads, environment, and tasks, so that they can be imported into the autonomous driving simulation platform to generate dynamic traffic simulation scenarios. The key parameter screening module is used to screen out key parameter sets of five related elements: people, vehicles, roads, environment, and tasks through sensitivity analysis. The optimization model building module is used to establish a dual-objective stochastic simulation optimization model for autonomous driving performance boundary mining. The selected key parameter set is used as decision variables, traffic flow uncertainty and background vehicle random dangerous driving behavior are used as disturbance factors, autonomous driving simulation platform is used as evaluation tool, and the absolute value of the speed difference and distance between the tested autonomous vehicle and the surrounding dynamic traffic participants and static objects is minimized as the optimization objective. The model solving module is used to solve the bi-objective stochastic simulation optimization model using the bi-objective monkey swarm random agent optimization algorithm, and obtain the Pareto stochastic optimal key parameter solution set. The performance boundary output module is used to determine the critical scenarios that the autonomous vehicle can cope with based on the Pareto random optimal key parameter solution set, and to derive the performance boundary of the autonomous vehicle under test.
8. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for loading and executing the computer program to implement the autonomous vehicle performance boundary mining method as described in any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the autonomous vehicle performance boundary mining method as described in any one of claims 1 to 6.