Automatic driving vehicle open road test route planning method

Abstract

The invention relates to the technical field of automatic driving test, and discloses an automatic driving vehicle open road test route planning method comprising the following steps: 1, collecting target historical data; 2, constructing a traffic flow characteristic prediction AI model based on deep learning and a test road risk level prediction AI model based on deep learning; 3, taking the road characteristics and future traffic flow characteristics output by the traffic flow characteristic prediction AI model as input, and outputting future test road risk levels of the roads in the urban road network by the test road risk level prediction AI model; and step 4, based on the future test road risk level, according to the test demand index, screening out target test roads, and combining the target test roads to obtain a test route planning scheme. According to the invention, a test route planning scheme meeting requirements can be provided for the open road test of the automatic driving vehicle, the test efficiency can be improved, and the test effect can be optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving testing, and in particular to a method for planning an open road test route for an autonomous driving vehicle. Background Art

[0002] In order to verify the safety of vehicles equipped with autonomous driving functions, large-scale road tests with a length of hundreds of millions of kilometers are required. However, a common problem in current open road tests is low test efficiency. This is because the test routes often contain a large number of low-risk sections with high scene repetition, and the tests on such sections cannot effectively cover the operational design domain (ODD) of the autonomous driving system. This means that longer road test mileage is required to fully verify the safety of the vehicle, which is a huge challenge to manpower, material resources and financial resources.

[0003] In addition, since open road testing of autonomous vehicles is a relatively new field, the current relevant testing regulations and standards are not yet perfect, resulting in a lack of scientific theoretical guidance when selecting open test roads, and the test road planning is not objective and complete. Secondly, researchers mainly focus on improving the testing efficiency of autonomous driving systems in simulation environment testing and closed field testing, while there is less research on open road testing. This imbalance has led to slow progress in open road testing of autonomous vehicles. Summary of the invention

[0004] The present invention aims to provide a method for planning an open road test route for an autonomous driving vehicle, which can provide a test route planning solution that meets the requirements for open road testing of autonomous driving vehicles, help improve its test efficiency, and optimize the test effect.

[0005] The basic solution provided by the present invention is: a method for planning a route for an open road test of an autonomous driving vehicle, comprising the following steps:

[0006] Step 1, collecting target historical data; the target historical data includes road characteristic data, environmental characteristic data, traffic flow characteristic data, accident and violation data; and based on the accident and violation data, constructing a test road risk level library of the urban road network;

[0007] Step 2, constructing a traffic flow characteristics prediction AI model based on deep learning and a test road risk level prediction AI model based on deep learning; the test road risk level prediction AI model uses road characteristics and traffic flow characteristics as independent variables and the test road risk level as a dependent variable for analysis; and divides the training set and the validation set based on the target historical data, and optimizes the traffic flow characteristics prediction AI model and the test road risk level prediction AI model based on the training set and the validation set;

[0008] Step 3, using the road characteristics and the future traffic flow characteristics output by the traffic flow characteristics prediction AI model as input, the test road risk level prediction AI model outputs the future test road risk level of the roads in the urban road network;

[0009] Step 4: Based on the risk level of future test roads and test demand indicators, select target test roads, combine target test roads, and obtain a test route planning solution.

[0010] The working principle and advantages of the present invention are:

[0011] The present invention provides a method for planning an open road test route for an autonomous driving vehicle. The method includes a traffic flow characteristic prediction AI model based on deep learning and a test road risk level prediction AI model. Through the AI ​​model, various historical information of the road itself and the traffic flow are used, combined with historical traffic violations, accidents and other data disclosed on the Internet, to predict the traffic flow conditions of future driving sections and the risk level of the test road, and incorporate the predicted test road risk level into the design considerations of the road test route. Based on different open road test requirements, the proportion of high-risk level sections in the test route can be targeted to increase, making the test scene more complex, thereby increasing the probability of exposing problems during the test process, effectively increasing the ratio of complex scenes during the road test process, and better verifying the safety and reliability of the autonomous driving system, which helps to speed up the test process of autonomous driving vehicles on open roads.

[0012] In particular, this solution incorporates the "test road risk level" into the test route planning, and uses the "test road risk level" as a criterion for judging whether the road is suitable as a test section, which improves the road test efficiency of the selected route and can efficiently verify the safety of autonomous driving vehicles. In addition, compared with the existing open test road planning schemes, most of the existing schemes only focus on the matching relationship between the road environment or mileage of the test road and the required test conditions, which can meet the testing needs of ordinary vehicles. However, for higher-level autonomous driving vehicles (such as L4 and L5 autonomous driving vehicles), their automation level is higher and there are more requirements for test scenarios, which require the test scenarios to be more challenging for the vehicle and the test scenarios covered to be more effective and comprehensive. Based on this, this solution first explores the potential relationship between the road characteristics and traffic flow characteristics and the test road risk, and learns this potential relationship based on a specific test road risk level prediction AI model to obtain the future test road risk level of each road section, support the generation of test route planning schemes, and make the scheme design more scientific. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1This is a schematic diagram of the overall process of an embodiment of a method for planning an open road test route for an autonomous driving vehicle according to the present invention;

[0014] Figure 2 A schematic diagram of an AI model establishment process for an embodiment of a method for planning an open road test route for an autonomous driving vehicle according to the present invention;

[0015] Figure 3 A schematic diagram of a test route planning process of an open road test route planning method for an autonomous driving vehicle according to an embodiment of the present invention;

[0016] Figure 4 A schematic diagram of test route planning generation for an embodiment of a method for planning an open road test route for an autonomous driving vehicle of the present invention. DETAILED DESCRIPTION

[0017] The following is a further detailed description through specific implementation methods:

[0018] The embodiment is basically as shown in the attached Figure 1 A method for planning a route for an open road test of an autonomous driving vehicle is shown, comprising the following steps:

[0019] Step 1, collecting target historical data; the target historical data includes road characteristic data, environmental characteristic data, traffic flow characteristic data, accident and violation data; and based on the accident and violation data, constructing a test road risk level library of the urban road network.

[0020] The target historical data are road characteristic data, environmental characteristic data, traffic flow characteristic data, accident and violation data of the urban road network in the past year.

[0021] Among them, according to the API interface of existing maps such as Baidu / AutoNavi / Tencent or the urban road network data set published on government websites, the inherent characteristics of the road itself that do not change with time, including inherent characteristics such as the number of lanes and lane width, and POI (Point of Interest) are collected as road feature data, and a road feature database containing all roads in the urban road network is established for easy access.

[0022] Based on the API interfaces of existing maps such as Baidu / AutoNavi / Tencent or urban road network climate data sets published on government websites, climate and weather data on roads within one year are collected, including data on rain, fog, snow, dust and other changes over time, as environmental characteristic data, and a climate and environmental characteristic database covering all roads in the urban road network is established for easy access.

[0023] Traffic flow data of all roads in the urban road network within one year are collected through the API interfaces of existing maps such as Baidu / AutoNavi / Tencent, including traffic flow characteristics such as motor vehicle flow, non-motor vehicle flow, and average vehicle speed, as traffic flow feature data, and a traffic flow characteristic database covering all roads in the urban road network is established for easy access.

[0024] According to the API interface of existing maps such as Baidu / AutoNavi / Tencent, the traffic accident / violation road section data of the target city within one year is obtained, and the accident type, longitude and latitude, road information and information on high-incidence areas of violations are sorted out as accident and violation data. Furthermore, the test risk level of the road is mapped based on the weighted average of the number of accidents and violations in the accident and violation data.

[0025] In this step, when constructing the test road risk level library of the urban road network, the following sub-steps are included:

[0026] Based on the accident and violation data, the test road risk of each section in the urban road network is calculated; the calculated test road risk is in the range of [0,1], and it is divided into multiple levels according to the range; thus forming a test road risk level library of the urban road network.

[0027] The test road risk is calculated according to the following formula:

[0028] Among them, m i Refers to the number of traffic accidents occurring on road section i each year; n i Refers to the number of traffic violations occurring on road section i each year; M is the upper limit of the maximum number of accidents occurring on a single road section in a specific urban road network area, and N is the upper limit of the number of traffic violations occurring on a single road section in a specific urban road network area; w1 is the impact weight of traffic accidents on the risk of the test road, and w2 is the impact weight of traffic violations on the risk of the test road; the values ​​of w1 and w2 satisfy w1+w2=1, and the size relationship and values ​​of w1 and w2 are determined according to the relationship between traffic violations and accidents within one year in the road section area.

[0029] When the test road risk is within the range of [0,0.2], it is classified as low risk; when the test road risk is within the range of [0.2,0.4], it is classified as medium-low risk; when the test road risk is within the range of [0.4,0.6], it is classified as medium risk; when the test road risk is within the range of [0.6,0.8], it is classified as medium-high risk; when the test road risk is within the range of [0.8,1], it is classified as high risk.

[0030] Step 2: Build a deep learning-based traffic flow characteristics prediction AI model and a deep learning-based test road risk level prediction AI model, such as Figure 2 shown.

[0031] The traffic flow characteristics prediction AI model takes the traffic flow characteristics data in the target historical data as input and the future traffic flow characteristics data as output. The model can capture and characterize the evolution of factors such as motor vehicle flow, average traffic flow speed, climate environment, accidents and violations, and road characteristics in traffic flow characteristics by learning the collected historical traffic flow characteristics data, and combine the above information to predict the changing trend of future traffic flow.

[0032] The test road risk level prediction AI model uses road characteristics and traffic flow characteristics as independent variables and the test road risk level as the dependent variable for analysis; and divides the training set and the verification set based on the target historical data, and optimizes the traffic flow characteristics prediction AI model and the test road risk level prediction AI model based on the training set and the verification set.

[0033] Step 3: Using the road characteristics and the future traffic flow characteristics output by the traffic flow characteristics prediction AI model as input, the test road risk level prediction AI model outputs the future test road risk level of the roads in the urban road network.

[0034] In this embodiment, a test risk level prediction database including all roads in the urban road network can also be constructed based on the output future test road risk level for use in related tests.

[0035] Step 4: Based on the risk level of future test roads and test demand indicators, select target test roads, combine target test roads, and obtain a test route planning solution, such as Figure 3 shown.

[0036] The test requirement indicators include: test time period, test route length, and vehicle design operating range definition (i.e., ODD definition, which defines the drivable conditions of the autonomous driving system).

[0037] In this step, the roads are first sorted from high to low based on the risk level of future test roads, and a specific number of target test roads are selected according to the test demand indicators.

[0038] When combining target test roads, it includes: setting each target test road as a point, and setting the distance between target test roads as an edge; setting the planning goal to shorten the test mileage; generating the shortest path containing all points based on the minimum spanning tree algorithm (Prim algorithm) as a test route planning solution; such as Figure 4 As shown, it is an example diagram of a test route planning scheme obtained by using this method.

[0039] When combining target test roads, it includes: setting each target test road as a point, and setting the distance between the target test roads as an edge; setting a test starting point and a test end point, and setting a planning goal to find the shortest route between the test starting point and the test end point that covers high-risk roads; generating the shortest path between two points based on the shortest path algorithm (Dijkstra algorithm, single-source shortest path algorithm), and obtaining a test route with a short distance between the two points and a high risk level as a test route planning solution.

[0040] The present embodiment provides a method for planning an open road test route for an autonomous driving vehicle, which can provide a test route planning solution that meets the requirements for open road testing of autonomous driving vehicles, thereby helping to improve its test efficiency and optimize test results.

[0041] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement the scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. A method for planning an open road test route for an autonomous driving vehicle, characterized in that: The following steps are involved: Step 1, collecting target historical data; the target historical data includes road characteristic data, environmental characteristic data, traffic flow characteristic data, accident and violation data; and based on the accident and violation data, constructing a test road risk level library of the urban road network; Step 2, constructing a traffic flow characteristics prediction AI model based on deep learning and a test road risk level prediction AI model based on deep learning; the test road risk level prediction AI model uses road characteristics and traffic flow characteristics as independent variables and the test road risk level as a dependent variable for analysis; and divides the training set and the validation set based on the target historical data, and optimizes the traffic flow characteristics prediction AI model and the test road risk level prediction AI model based on the training set and the validation set; Step 3, using the road characteristics and the future traffic flow characteristics output by the traffic flow characteristics prediction AI model as input, the test road risk level prediction AI model outputs the future test road risk level of the roads in the urban road network; Step 4: Based on the risk level of future test roads and test demand indicators, select target test roads, combine target test roads, and obtain a test route planning solution.

2. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: In step 1, when constructing the test road risk level library of the urban road network, the following sub-steps are included: Based on the accident and violation data, the test road risk of each section in the urban road network is calculated; the calculated test road risk is in the range of [0,1], and it is divided into multiple levels according to the range; thus forming a test road risk level library of the urban road network; The test road risk is calculated according to the following formula: Among them, m i Refers to the number of traffic accidents occurring on road section i each year; n i Refers to the number of traffic violations occurring on road section i each year; M is the upper limit of the maximum number of accidents occurring on a single road section in a specific urban road network area, and N is the upper limit of the number of traffic violations occurring on a single road section in a specific urban road network area; w1 is the impact weight of traffic accidents on the risk of the test road, and w2 is the impact weight of traffic violations on the risk of the test road; the values ​​of w1 and w2 satisfy w1+w2=1, and the size relationship and values ​​of w1 and w2 are determined according to the relationship between traffic violations and accidents within one year in the road section area.

3. The method for planning an open road test route for an autonomous driving vehicle according to claim 2, characterized in that: When the test road risk is within the range of [0,0.2], it is classified as low risk; when the test road risk is within the range of [0.2,0.4], it is classified as medium-low risk; when the test road risk is within the range of [0.4,0.6], it is classified as medium risk; when the test road risk is within the range of [0.6,0.8], it is classified as medium-high risk; when the test road risk is within the range of [0.8,1], it is classified as high risk.

4. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: The target historical data are road characteristic data, environmental characteristic data, traffic flow characteristic data, accident and violation data of the urban road network in the past year.

5. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: The test requirement indicators include: test time period, test route length, and vehicle design operating range definition.

6. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: In step 4, the roads are first sorted from high to low based on the risk level of the future test roads, and a specific number of target test roads are screened out according to the test demand indicators.

7. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: When combining target test roads, it includes: setting each target test road as a point, and setting the distance between the target test roads as an edge; setting the planning goal to shorten the test mileage; and generating the shortest path including all points based on the minimum spanning tree algorithm as a test route planning solution.

8. The method for planning an open road test route for an autonomous driving vehicle according to claim 1, characterized in that: When combining target test roads, it includes: setting each target test road as a point, and setting the distance between the target test roads as an edge; setting a test starting point and a test end point, and setting a planning goal to find the shortest route between the test starting point and the test end point that covers high-risk roads; generating the shortest path between two points based on the shortest path algorithm, and obtaining a test route with a short distance between the two points and a high risk level as a test route planning solution.

Images