Test method and system for acceleration performance of unmanned motorcycle under extreme working condition

Abstract

The present application relates to a test method and system for the acceleration performance of unmanned motorcycles under extreme conditions, belonging to the technical field of unmanned motorcycle testing. The specific process is as follows: step one, generate an overall scenario library according to the open source scenario data set generated by actual automatic driving testing; step two, collect some extreme condition testing scenarios of some unmanned motorcycles in real environment as part of the extreme condition scenario data set, and use the generative adversarial learning method to train to obtain an extreme condition virtual scenario library based on the overall scenario library and containing extreme condition characteristics; step three, for the generated extreme condition virtual scenario library, use unsupervised learning for clustering, and according to the proportion of various scenarios obtained by clustering, randomly generate according to the proportion when performing extreme condition testing; step four, according to the test results of the unmanned motorcycle in the generated extreme condition virtual scenario library, test the acceleration performance of the unmanned motorcycle.

Description

Technical Field

[0001] This invention relates to a method and system for testing the acceleration performance of unmanned motorcycles under extreme operating conditions, and belongs to the field of unmanned motorcycle testing technology. Background Technology

[0002] Compared to dual-rail vehicles, unmanned motorcycles, as a type of monorail unmanned vehicle, possess advantages such as better concealment, greater maneuverability, and stronger environmental adaptability, demonstrating enormous application potential in areas such as material transport in three-dimensional terrain and confined spaces, environmental monitoring, and military reconnaissance. However, dual-wheeled motorcycles are underactuated systems, inherently unstable, especially under extreme conditions. The stability of unmanned motorcycles faces significant challenges, and only through multiple, non-repeating extreme condition tests can their performance be ensured under real-world extreme conditions.

[0003] However, autonomous motorcycles are a tightly coupled system of vehicle, environment, and task. Testing them involves modules and functions such as the perception system, mapping and localization, decision-making and planning system, and control execution system. This requires testing not only the software functionality of these modules but also their overall performance in real-world road environments. Therefore, highly complex scenarios and tasks need to be designed to demonstrate the reliability and effectiveness of the autonomous motorcycle. Relying solely on road testing is not only inefficient but also poses road safety risks. Therefore, scenario-based virtual simulation testing has become a crucial approach to solving the performance verification challenges of autonomous vehicles. Currently, mainstream virtual simulation testing methods include the test matrix method, worst-case method, Monte Carlo method, and accelerated testing methods.

[0004] The test matrix method predefines test scenarios, and vehicles undergo testing, followed by subjective and objective evaluations. However, it's difficult to exhaustively list all test scenarios; good performance in pre-generated scenarios doesn't guarantee continued good performance in untested environments. The worst-case method evaluates vehicle driving performance under extreme conditions by selecting the most extreme or severe scenarios for analysis. However, it doesn't consider the probability of extreme situations, thus ignoring other scenarios. The Monte Carlo method uses statistical models to fit natural driving data and employs Monte Carlo simulations to test and evaluate vehicles. However, due to the limited number of dangerous scenarios and extreme conditions in natural driving data, its efficiency is low.

[0005] Each of the three methods described above has its own drawbacks. Therefore, we consider using an accelerated testing method. By constructing a virtual testing environment, autonomous vehicles can be tested in this environment as if they were driving on real public roads, but with higher efficiency. At the same time, this addresses the problem of insufficient effective sample size caused by the curse of sparsity. Summary of the Invention

[0006] In view of this, the present invention proposes a method and system for testing the acceleration performance of unmanned motorcycles under extreme conditions, which focuses on three aspects: the construction of an overall scenario library, the enhanced generation of a specific test scenario library for extreme conditions, and the multi-dimensional evaluation of the test results.

[0007] The technical solution for implementing the present invention is as follows:

[0008] Firstly, the present invention provides a method for testing the acceleration performance of an unmanned motorcycle under extreme operating conditions, the specific process of which is as follows:

[0009] Step 1: Generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests;

[0010] Step 2: Collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real environments as a partial extreme working condition scenario dataset. Use the generative adversarial learning method, take the overall scenario library as the original input and the partial extreme working condition dataset as the target, and train them against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features.

[0011] Step 3: For the generated virtual scenario library of extreme working conditions, unsupervised learning is used to perform clustering. Based on the proportion of various scenarios obtained from the clustering, the extreme working conditions are randomly generated according to the proportion.

[0012] Step 4: Based on the test results of the unmanned motorcycle in the generated extreme working condition virtual scenario library, conduct acceleration performance tests on the unmanned motorcycle.

[0013] Furthermore, the specific process of step one in this invention is as follows:

[0014] (1) Merge the typical autonomous driving scenario dataset and edge scenario dataset into a whole scenario dataset;

[0015] (2) Based on the similarity to the extreme working condition scenario, fuzzy rules are set, a membership function is designed, the membership degree of each scenario in the overall data is calculated, and scenarios with high membership degrees are extracted as the initial overall scenario set; wherein, the membership function is:

[0016] S=αF(e wather )+βG(e road )+γH(e object )

[0017] Wherein, F(e) weather ), G(e road ), H(e object ) are membership functions based on weather feature deviation, road feature deviation, and surrounding object feature deviation, respectively, with α, β, and γ being weighting coefficients, and α+β+γ=1.

[0018] Furthermore, the present invention ultimately selects scenes with a membership degree greater than 50% as the overall scene dataset.

[0019] Furthermore, in step two of this invention, after obtaining the virtual scenario library of extreme working conditions, the specific scenarios in the overall scenario library are transformed and combined according to the characteristics of the actual extreme working condition scenarios to generate different extreme working condition test samples.

[0020] Furthermore, the specific process of using the generative adversarial learning method in step two of this invention is as follows:

[0021] Using a subset of unmanned motorcycles under real-world extreme conditions as real data and thus the learning object, a generative adversarial network (GAN) training method is employed. The initial generator is initialized with the overall scene set, and random noise is added as the generated data. During training, firstly, the generator is fixed, and the discriminator is trained to accurately distinguish between real and fake samples. Secondly, the discriminator is fixed, and the generator is trained to generate samples sufficient to deceive the discriminator.

[0022] Furthermore, the value function of the adversarial network using the generative adversarial learning method in step two of this invention is as follows:

[0023]

[0024] Among them, P data (x) represents the true data distribution, P z (z) represents the distribution of generated data, z represents noise, D(x) represents the discriminator's judgment result on the real data, and D(G(z))) represents the discriminator's judgment result on the generated data.

[0025] Furthermore, the cost functions of the discriminator and the generator described in this invention are as follows:

[0026]

[0027] For the discriminator, the larger V(G,D) is, the better; for the generator, the smaller V(G,D) is, the better.

[0028] Furthermore, in step three of this invention, the unsupervised clustering of scene data adopts the K-means clustering method. For a given scene set, the samples are divided into K clusters according to the distance between the samples.

[0029] Furthermore, this invention calculates the distance between numerical objects using Euclidean distance.

[0030]

[0031] Where k is the number of clusters, C iLet x be the point set of the i-th cluster, and x be the points belonging to C. i Data points, μ i It is the center of the i-th cluster.

[0032] The choice should satisfy E>E. min The test samples, E is the distance between samples obtained from clustering, E min This is the minimum sample distance set.

[0033] Secondly, the present invention provides an acceleration performance testing system for unmanned motorcycles under extreme operating conditions, comprising: a data acquisition module, a data generation module, a clustering module, and an acceleration performance testing module.

[0034] The data acquisition module is used to collect and generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests;

[0035] The data generation module is used to collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real environments. As a partial extreme working condition scenario dataset, the generative adversarial learning method is used to take the overall scenario library as the original input and the partial extreme working condition set as the target. The two are trained against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features.

[0036] The data clustering module is used to cluster the generated virtual scenario library of extreme working conditions using unsupervised learning. Based on the proportion of various scenarios obtained from the clustering, the data is randomly generated according to the proportion during extreme working condition testing.

[0037] The acceleration performance testing module is used to perform acceleration tests on unmanned motorcycles based on the test results of the unmanned motorcycles in the generated extreme working condition virtual scenario library.

[0038] Beneficial effects:

[0039] This invention uses adversarial learning as the main method. It extracts traditional scenarios and trains a generator based on adversarial learning to generate the required extreme working conditions. Then, it places the unmanned motorcycle in the extreme working conditions that appear in proportion according to clustering and occurrence probability, and limits the distance between test scenarios to avoid repeated testing as much as possible. Finally, it evaluates the key indicators of extreme working conditions based on various categories and multi-dimensional methods. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments 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.

[0041] Figure 1 This is a flowchart of the performance testing method for unmanned motorcycles under extreme operating conditions according to the present invention.

[0042] Figure 2 This is a flowchart of the adversarial training method of the present invention;

[0043] Figure 3 This is the clustering result of the present invention. Detailed Implementation

[0044] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0045] It should be noted that, in the absence of conflict, the following embodiments and features can be combined with each other; and, based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0046] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0047] This application provides an embodiment of a method for testing the acceleration performance of an unmanned motorcycle under extreme operating conditions, such as... Figure 1 As shown, the specific process is as follows:

[0048] Step 1: Generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests.

[0049] Before generating the overall scene library, this step also sets fuzzy rules according to the scenarios that may occur under extreme working conditions of unmanned motorcycles, and removes scenes with low membership degree while retaining scenes with high membership degree.

[0050] When implementing this step, follow these steps:

[0051] (1) Overall scene data acquisition: The overall scene dataset is merged from the typical autonomous driving scene dataset and the edge scene dataset.

[0052] (2) Based on the similarity to the extreme working condition scenario, fuzzy rules are set, a membership function is designed, the membership degree of each scenario in the overall data is calculated, and scenarios with high membership degrees are extracted as the initial overall scenario set; wherein, the membership function is:

[0053] S=αF(e weather )+βG(e road )+γH(e object )

[0054] Wherein, F(e) weather ), G(e road ), H( eobject The membership functions are based on weather feature bias, road feature bias, and surrounding object feature bias, respectively, with α, β, and γ being weighting coefficients, and α + β + γ = 1. Scenes with a membership degree greater than 50% are ultimately selected as the overall scene dataset.

[0055] Step 2: Collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real-world environments as a partial extreme working condition scenario dataset. Using a generative adversarial learning method, take the overall scenario library as the original input and the partial extreme working condition dataset as the target, and train them against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features.

[0056] After obtaining the virtual scenario library of extreme working conditions, this step further transforms and combines the specific scenarios in the overall scenario library according to the characteristics of actual extreme working condition scenarios, thereby generating a sufficient number of different extreme working condition test samples.

[0057] In practice, adversarial learning is performed according to the following process: Figure 2 As shown:

[0058] Using a subset of scenarios of autonomous motorcycles operating under real-world extreme conditions as the training data, a generative adversarial network (GAN) training method (containing a generator and a discriminator) is employed. The generator is initially initialized with the overall scene set, with random noise added as generated data. First, with the generator fixed, the goal is for the discriminator to accurately distinguish between real and fake samples. Second, with the discriminator fixed, the goal is for the generator to generate samples sufficient to deceive the discriminator. The overall value function is as follows:

[0059]

[0060] Among them, P data (x) represents the true data distribution, D z (z) represents the distribution of generated data, z represents noise, D(x) represents the discrimination result of the discriminator on the real data, and D(G((z))) represents the discrimination result of the discriminator on the generated data;

[0061] The cost functions for the discriminator and the generator are as follows:

[0062]

[0063] For the discriminator D, a larger V(G,D) is better; while for the generator G, a smaller V(G,D) is better. During training, one is always fixed while the other is trained.

[0064] The ultimate goal is to train a sufficiently good generator that can randomly generate a library of virtual scenarios for extreme working conditions.

[0065] Step 3: For the generated virtual scenario library of extreme working conditions, unsupervised learning is used to perform clustering to generate several different extreme working condition scenarios. Based on the proportion of each scenario obtained from the clustering, the scenarios are randomly generated according to the proportion during extreme working condition testing.

[0066] In this step, a generator is used to randomly generate scene data. Unsupervised clustering is then used to derive scenes with various characteristics. The probability of high-probability extreme conditions occurring is increased proportionally to avoid serious deviations in test results caused by bad or duplicate data, which could affect the direction of control optimization for unmanned motorcycles.

[0067] In this step, the unsupervised clustering of scene data uses the K-means clustering method. For a given scene set, the samples are divided into K clusters based on the distance between them. K-means clustering calculates the distance between numerical objects using Euclidean distance.

[0068]

[0069] Where k is the number of clusters, C i Let x be the point set of the i-th cluster, and x be the points belonging to C. i Data points, μ i Let i be the center of the i-th cluster, and This represents the square of the Euclidean distance between x and the cluster center. K-means clustering aims to maximize the distance between points within a cluster and minimize the distance between clusters. The clustering result is as follows: Figure 3 As shown.

[0070] Using K-means clustering, several virtual limit scenarios with varying sample sizes can be obtained, categorized into high-probability and low-probability scenarios based on sample size. Simultaneously, considering that the generated sample library inevitably contains some duplicate samples, the distance between numerical objects recorded during clustering should not be too small to minimize duplicate results during testing, thus accelerating the testing process. The selection of test samples should meet the following requirements:

[0071] E>E min

[0072] The test samples are provided proportionally by K-means clustering, where E is the distance between samples obtained from the clustering. Where E... min This is the minimum sample distance to set in order to avoid repeated testing as much as possible.

[0073] Step four: Based on the test results of the unmanned motorcycle in the generated virtual scenario library of extreme working conditions, conduct acceleration performance tests and evaluations on the unmanned motorcycle.

[0074] Performance testing and evaluation in this step: Evaluation is based on the response data of the autonomous motorcycle during virtual scenario testing, used as key performance indicators. Key performance indicators under extreme conditions include the stability of the autonomous motorcycle, evaluated using its lateral speed, roll angle, and roll rate during testing; the autonomous motorcycle's passability is evaluated based on the width and height of obstacles it can traverse.

[0075] For example, the stability of an unmanned motorcycle during motion can be measured based on its roll angle, roll rate, and lateral acceleration; the anti-interference capability of the unmanned motorcycle's self-balancing control system can be evaluated based on indicators such as roll rate and lateral acceleration after a horizontal impact.

[0076] This invention uses reinforcement learning to learn from a comprehensive scenario library and then performs adversarial learning against collected data on specific extreme operating conditions, thereby generating a virtual simulated extreme operating condition scenario library. Unmanned motorcycles are tested within this virtual extreme operating condition scenario library generated through adversarial learning. This virtual extreme operating condition scenario library compensates for the deficiencies in corresponding extreme operating condition scenarios in real-world environments. Furthermore, conducting extreme operating condition testing in a virtual environment also achieves the goal of accelerating testing.

[0077] Furthermore, this invention clusters the generated virtual scene library and tests it according to the occurrence ratio, thereby simulating the occurrence of various extreme working conditions of unmanned motorcycles in real environments. This avoids the problems of large-scale testing of low-probability extreme working conditions and low-probability testing of high-probability extreme working conditions, thus achieving accelerated testing.

[0078] This application provides an acceleration performance testing system for unmanned motorcycles under extreme operating conditions, comprising: a data acquisition module, a data generation module, a clustering module, and an acceleration performance testing module.

[0079] The data acquisition module is used to collect and generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests;

[0080] The data generation module is used to collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real environments. As a partial extreme working condition scenario dataset, the generative adversarial learning method is used to take the overall scenario library as the original input and the partial extreme working condition set as the target. The two are trained against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features.

[0081] The data clustering module is used to cluster the generated virtual scenario library of extreme working conditions using unsupervised learning. Based on the proportion of various scenarios obtained from the clustering, the data is randomly generated according to the proportion during extreme working condition testing.

[0082] The acceleration performance testing module is used to perform acceleration tests on unmanned motorcycles based on the test results of the unmanned motorcycles in the generated extreme working condition virtual scenario library.

[0083] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for testing the acceleration performance of an unmanned motorcycle under extreme operating conditions, characterized in that, The specific process is as follows: Step 1: Generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests; Step 2: Collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real environments as a partial extreme working condition scenario dataset. Use the generative adversarial learning method, take the overall scenario library as the original input and the partial extreme working condition dataset as the target, and train them against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features. Step 3: For the generated virtual scenario library of extreme working conditions, unsupervised learning is used to perform clustering. Based on the proportion of various scenarios obtained from the clustering, the extreme working conditions are randomly generated according to the proportion. Step 4: Based on the test results of the unmanned motorcycle in the generated extreme working condition virtual scenario library, conduct acceleration performance tests on the unmanned motorcycle; The specific process of using the generative adversarial learning method in step two is as follows: Using a subset of unmanned motorcycles under real-world extreme conditions as real data and thus the learning object, a generative adversarial network (GAN) training method is employed. The initial generator is initialized with the overall scene set, and random noise is added as the generated data. During training, firstly, the generator is fixed, and the discriminator is trained to accurately distinguish between real and fake samples. Secondly, the discriminator is fixed, and the generator is trained to generate samples sufficient to deceive the discriminator. The value function of the adversarial network using generative adversarial learning in step two is as follows: in, For the true data distribution, To generate the data distribution, z represents noise. The discriminator's judgment result on the real data. The discriminator's judgment result on the generated data; The cost functions for the discriminator and the generator are as follows: For the discriminator, the larger V(G,D) is, the better; for the generator, the smaller V(G,D) is, the better.

2. The acceleration performance testing method for unmanned motorcycles under extreme operating conditions according to claim 1, characterized in that, The specific process of step one is as follows: (1) Merge the typical autonomous driving scenario dataset and the edge scenario dataset into a unified scenario dataset; (2) Based on the similarity with the extreme working condition scenario, fuzzy rules are set, a membership function is designed, the membership degree of each scenario in the overall data is calculated, and scenarios with high membership degree are extracted as the initial overall scenario set; wherein, the membership function is: in, These are membership functions based on weather feature bias, road feature bias, and surrounding object feature bias, respectively. These are weighting coefficients. .

3. The acceleration performance testing method for unmanned motorcycles under extreme operating conditions according to claim 2, characterized in that, Finally, scenes with a membership degree greater than 50% were selected as the overall scene dataset.

4. The acceleration performance testing method for unmanned motorcycles under extreme operating conditions according to claim 1, characterized in that, After obtaining the virtual scenario library of extreme working conditions, step two further transforms and combines the specific scenarios in the overall scenario library according to the characteristics of the actual extreme working conditions to generate test samples for different extreme working conditions.

5. The acceleration performance testing method for unmanned motorcycles under extreme operating conditions according to claim 1, characterized in that, In step three, the unsupervised clustering of the scene data adopts the K-means clustering method. For a given scene set, the samples are divided into K clusters according to the distance between the samples.

6. The acceleration performance testing method for unmanned motorcycles under extreme operating conditions according to claim 5, characterized in that, The distance between samples is calculated using Euclidean distance. Where k is the number of clusters, C i Let x be the point set of the i-th cluster, and x be the points belonging to C. i Data points, It is the center of the i-th cluster; The selection should satisfy... The test samples, where E is the distance between samples obtained from clustering. This is the minimum sample distance set.

7. A testing system based on the acceleration performance testing method for unmanned motorcycles under extreme operating conditions as described in claim 1, characterized in that, include: Data acquisition module, data generation module, clustering module, and performance testing module; The data acquisition module is used to collect and generate an overall scenario library based on the open-source scenario dataset generated from actual autonomous driving tests; The data generation module is used to collect some scenarios of unmanned motorcycles undergoing extreme working condition tests in real environments. As a partial extreme working condition scenario dataset, the generative adversarial learning method is used to take the overall scenario library as the original input and the partial extreme working condition set as the target. The two are trained against each other to obtain a virtual extreme working condition scenario library that is based on the overall scenario library and contains extreme working condition features. The data clustering module is used to cluster the generated virtual scenario library of extreme working conditions using unsupervised learning. Based on the proportion of various scenarios obtained from the clustering, the data is randomly generated according to the proportion during extreme working condition testing. The acceleration performance testing module is used to perform acceleration tests on unmanned motorcycles based on the test results of the unmanned motorcycles in the generated extreme working condition virtual scenario library.

Images