Method and device for generating a generalized intelligent driving scenario

Abstract

The application discloses a method for generating a generalized intelligent driving scene, comprising: obtaining the requirement of the generalized scene and determining the parameters that need to be generalized; configuring the parameters; performing the generalization processing on the parameters, including the number optimization on the test cases generated based on the parameters and the nested loop optimization on the generation process; saving the parameters after the generalization processing in the form of a parameter list to a parameter database; calling the corresponding parameter list from the parameter database; generating a scene file, calling a scene generation service, generating the scene file according to the parameter list and saving the scene file in a scene library; generating an intelligent driving scene, calling a virtual simulation service, calling and loading the corresponding scene file from the scene library, and analyzing the scene file to generate a virtual intelligent driving scene. The application also discloses a device for generating a generalized intelligent driving scene, comprising: a client, a generalization processing server, a parameter database, a master control server and a scene library.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and more specifically, to the field of generating virtual scenes for intelligent driving. Background Technology

[0002] In the intelligent driving testing system, simulation testing has become a crucial part of intelligent driving vehicle development and testing due to its advantages such as reducing testing cycle and cost, improving test coverage, and avoiding safety issues associated with real-vehicle road testing. Scenario libraries, as the cornerstone of simulation testing, contain rich attribute information, and scenario generalization methods are indispensable for building a high-coverage scenario library. Currently, most autonomous driving companies lack the ability to generalize virtual scenarios, relying solely on manual modification of scenario parameters to generate new scenarios, making it difficult to find long-tail scenarios with significant testing value. Existing scenario generalization capabilities only perform independent generalization of some test parameters, then generate large amounts of data through permutations and combinations, including many invalid scenarios. Furthermore, the generalization algorithms are mainly based on nested loops; each execution of the outer loop causes the inner loop to iterate entirely, resulting in low generalization efficiency. Therefore, building an efficient simulation scenario generalization system to improve testing efficiency is a key focus and challenge in intelligent driving simulation development. Summary of the Invention

[0003] According to an embodiment of the present invention, a method for generating generalized intelligent driving scenarios is proposed, comprising:

[0004] Obtain the requirements of the generalization scenario. The requirements of the generalization scenario correspond to the parameters that need to be generalized. Determine the parameters that need to be generalized based on the requirements of the generalization scenario.

[0005] Configure the parameters, including the number of generalization combinations, the range of parameter values, and the step size of the parameters for the determined parameters that need to be generalized.

[0006] The parameters are generalized, which includes optimizing the number of test cases generated based on the parameters and optimizing the nested loop process of test case generation.

[0007] The generalized parameters are saved to the parameter database. The generalized parameters are saved in the parameter database in the form of a parameter list. The parameter list includes the parameters, the parameter configuration information, and the test cases. The parameter list corresponds to the requirements of the generalization scenario.

[0008] After the parameters have been generalized, the corresponding parameter list is retrieved from the parameter database according to the requirements of the generalization scenario.

[0009] Generate scene files, call the scene generation service, the scene generation service reads the parameter list, generates scene files based on the parameter list, and saves the scene files in the scene library;

[0010] The system generates intelligent driving scenarios by calling virtual simulation services. The virtual simulation services then retrieve and load the corresponding scenario files from the scenario library and parse the scenario files to generate virtual intelligent driving scenarios.

[0011] In one embodiment, when obtaining the requirements of a generalized scenario, a unique task ID is assigned to the requirement, and the user ID of the user who submitted the requirement and the timestamp of the submission are recorded simultaneously. When saving the generalized parameters to the parameter database, the task ID, user ID, and timestamp are stored together with the corresponding parameter list in the parameter database, where a one-to-one correspondence is established between the parameter list and the requirements of the generalized scenario through the task ID.

[0012] In one embodiment, the step of generalizing the parameters, including optimizing the number of test cases generated by the parameters, includes: based on the number n of generalization combinations, for all parameters, listing all possible combinations of selecting n parameters from all parameters to form a generalization combination; for each generalization combination, listing all test cases based on the value range and step size of each parameter; comparing the test cases of all generalization combinations and deleting duplicate test cases; and using the remaining non-duplicate test cases as the optimized test cases.

[0013] In one embodiment, the nested loop optimization of the test case generation process in the parameter generalization step includes: establishing an outer nested table and an inner nested table, wherein the elements contained in the inner nested table are a subset of the elements contained in the outer nested table; creating an index corresponding to the elements contained in the inner nested table, the index enabling the elements to be directly located in both the outer and inner nested tables; executing a loop in the outer nested table, wherein the elements are traversed and assigned values ​​within the loop in the outer nested table; executing a loop in the inner nested table, wherein the elements are directly retrieved from the outer nested table through the index during the loop in the inner nested table, thereby reducing the number of times the nested loop is executed.

[0014] In one embodiment, the scene generation service generates a scene file for each line of the parameter list, and the scene files are saved in XML format.

[0015] According to an embodiment of the present invention, a device for generating generalized intelligent driving scenarios is proposed, comprising: a client, a generalization processing server, a parameter database, a main control server, and a scenario library;

[0016] The client has an interactive interface. The client obtains the requirements of the generalization scenario through the interactive interface. The requirements of the generalization scenario correspond to the parameters that need to be generalized. The parameters that need to be generalized are determined according to the requirements of the generalization scenario.

[0017] The client sends the parameters to be generalized to the generalization server. The generalization server retrieves the parameter configuration information from the parameter database and configures the parameters. The parameter configuration information includes: the number of generalization combinations, the parameter value range, and the parameter step size. The generalization server performs generalization processing on the parameters, which includes optimizing the number of test cases generated based on the parameters and optimizing the nested loop process of test case generation. The generalization server saves the generalized parameters to the parameter database. The generalized parameters are stored in the parameter database in the form of a parameter list, which includes the parameters, parameter configuration information, and test cases. The parameter list corresponds to the requirements of the generalization scenario.

[0018] After the generalization server completes parameter generalization and saves the generalized parameters to the parameter database, it sends a completion command to the master control server. The master control server parses the completion command and retrieves the corresponding parameter list from the parameter database. The master control server then calls the scene generation service, which reads the parameter list, generates a scene file based on the parameter list, and saves the scene file in the scene library.

[0019] The client calls the virtual simulation service, which retrieves and loads the corresponding scene file from the scene library, parses the scene file, and generates a virtual intelligent driving scene.

[0020] In one embodiment, when a client obtains a requirement for a generalized scenario, it assigns a unique task ID to that requirement and simultaneously records the user ID of the user who submitted the requirement and the timestamp of the submission. When the generalization processing server saves the generalized parameters to the parameter database, the task ID, user ID, and timestamp are stored together with the corresponding parameter list in the parameter database, where a one-to-one correspondence is established between the parameter list and the requirements of the generalized scenario through the task ID.

[0021] In one embodiment, the completion instruction sent by the generalization processing server to the master control server includes a task ID. The master control server parses the completion instruction to obtain the task ID and retrieves the corresponding parameter list from the parameter database based on the task ID.

[0022] In one embodiment, the scene file generated by the scene generation service includes a task ID. The virtual simulation service called by the client calls and loads the corresponding scene file from the scene library based on the task ID, and parses the scene file to generate a virtual intelligent driving scene.

[0023] In one embodiment, the generalization processing server optimizes the number of test cases generated by the parameters. Specifically, based on the number n of generalization combinations, the server lists all possible combinations of n parameters selected from all parameters to form a generalization combination. For each generalization combination, it lists all test cases based on the value range and step size of each parameter. The server compares the test cases of all generalization combinations and removes duplicate test cases. The remaining non-duplicate test cases are then used as the optimized test cases.

[0024] In one embodiment, the generalization server optimizes the test case generation process using nested loops. The generalization server establishes an outer nested table and an inner nested table, where the elements in the inner nested table are a subset of the elements in the outer nested table. An index is created, corresponding to an element in the inner nested table, allowing the element to be directly located within both the outer and inner nested tables. A loop is executed within the outer nested table, iterating and assigning values ​​to the elements. Similarly, a loop is executed within the inner nested table, where the elements are directly retrieved from the outer nested table using the index during the loop, thus reducing the number of nested loop executions.

[0025] In one embodiment, the scene generation service invoked by the master server generates a scene file for each line of the parameter list, and the scene file is saved in XML format.

[0026] In one embodiment, the client, generalization processing server, parameter database, master control server, and scene library are deployed in a distributed manner and communicate with each other through a communication network.

[0027] In one embodiment, the virtual simulation service is deployed on the client or on the network; the scene generation service is deployed on the main control server or on the network.

[0028] In one embodiment, the master server is located in the cloud.

[0029] The generalized intelligent driving scenario generation method and apparatus of the present invention solve the following problems: addressing the current situation where most intelligent driving companies lack the ability to automatically generalize intelligent driving scenarios, thus improving the test coverage of the scenario library; and solving the problems of large data volume, slow speed, and low efficiency of existing intelligent driving scenario generalization methods. The generalized intelligent driving scenario generation method and apparatus of the present invention have the following beneficial effects: based on the T-way generalization method, improving scenario test coverage; employing index nesting to build a generalization logic algorithm, improving the efficiency of simulation scenario generalization; and adopting an end-to-end full-process technical architecture to automatically generalize intelligent driving scenarios, replacing manual parameter modification and saving significant manpower and time costs required for scenario generalization. Attached Figure Description

[0030] Figure 1 A flowchart of a method for generating generalized intelligent driving scenarios according to an embodiment of the present invention is disclosed.

[0031] Figure 2 A schematic diagram illustrating nested loop optimization in a method for generating generalized intelligent driving scenarios according to an embodiment of the present invention is disclosed.

[0032] Figure 3 A schematic diagram of the structure of a generalized intelligent driving scenario generation device according to an embodiment of the present invention is disclosed. Detailed Implementation

[0033] refer to Figure 1 As shown, Figure 1 A flowchart of a method for generating a generalized intelligent driving scenario according to an embodiment of the present invention is disclosed. The method for generating a generalized intelligent driving scenario is characterized by comprising:

[0034] S101. Obtain the requirements of the generalization scenario. The requirements of the generalization scenario correspond to the parameters that need to be generalized. Based on the requirements of the generalization scenario, the parameters that need to be generalized are determined. In one embodiment, when obtaining the requirements of the generalization scenario, a unique task ID is assigned to the requirements of the generalization scenario, and the user ID of the user who made the requirement and the timestamp of the requirement are recorded at the same time. The task ID, as the identifier of the requirements of the generalization scenario, will be continuously used in subsequent processing. All data associated with the requirements of the generalization scenario will use the task ID. Obtaining the user ID and timestamp is for data integrity considerations. All tests of intelligent driving contain rich attribute information. However, for each specific simulation scenario, only some relevant attributes are involved, and the remaining attributes are not closely related to the specific simulation scenario. The number of attributes (or parameters) will directly affect the number of subsequent test cases and the number of nested loop calculations, which will greatly affect the computational load and time consumption of simulation tests. Since specific test tasks are only related to some attributes in the scenario library, the attributes (parameters) corresponding to these requirements are determined by determining the requirements of the generalization scenario, thereby determining the parameters that need to be generalized to form a generalization combination. The goal of generalization composition is to enable the model to infer the meaning of combinations of unknown objects based on the properties of known objects. Common parameters used in simulation scenarios during testing include environmental visibility, vehicle speed, obstacle vehicle speed, road adhesion coefficient, and following distance. By obtaining the requirements of the generalization scenario, the parameters to be generalized can be determined from the above parameters. In one embodiment, based on the requirements of the selected generalization scenario, the parameters to be generalized are determined to be vehicle speed and following distance. In another embodiment, initial or expected values ​​can be assigned to the selected parameters. For example, for the selected parameters vehicle speed and following distance, their initial or expected values ​​can be set as: vehicle speed 60 km / h and following distance 200 m.

[0035] S102. Configure the parameters. For the determined parameters that need to be generalized, configure the number of generalization combinations, the value range of the parameters, and the step size of the parameters. After determining the parameters that need to be generalized in step S101, these parameters need to be configured. In one embodiment, the configuration includes the number of generalization combinations, the value range of the parameters, and the step size of the parameters. For example, if the parameters that need to be generalized in step S101 are vehicle speed and following distance, the vehicle speed and following distance are configured as follows:

[0036] The parameter value ranges are: vehicle speed (0–120 km / h) and following distance (0 m or greater). If other parameters are selected for generalization, their value ranges can be obtained from the parameter database.

[0037] Parameter step size: The step size for vehicle speed can be set to 12 km / h, and the step size for following distance can be set to 10 m. Similarly, if the selected parameter to be generalized is another parameter, the corresponding step size can be obtained from the parameter database.

[0038] Number of generalized components: The number of generalized components is determined by the user according to their needs, and the default value is usually 3. The commonly used number of generalized components is 2 to 4.

[0039] S103. Perform parameter generalization. Generalization includes optimizing the number of test cases generated based on the parameters and optimizing the nested loop process of test case generation.

[0040] In one embodiment, the step of generalizing the parameters, including optimizing the number of test cases generated from the parameters, includes:

[0041] Given the number of generalization combinations n, for all parameters, list all possible combinations of selecting n parameters from all parameters to form a generalization combination;

[0042] For each generalization combination, list all test cases based on the value range and step size of each parameter;

[0043] Compare the test cases for all generalization combinations and remove duplicate test cases;

[0044] The remaining non-repeating test cases are used as quantity-optimized test cases.

[0045] The above optimization process is based on T-way's generalization logic. A concrete example will be used below to help understand this optimization process.

[0046] As mentioned earlier, simulation scenarios contain rich attribute information. Most test tasks are only related to some attributes in the scenario library. The goal of combinatorial generalization is to enable the model to infer the meaning of unknown object combinations based on the properties of known objects. The generalization methods used in existing technologies adopt the theory of full permutation and combination generalization, that is, all combinations of test cases must be tested at least once, resulting in a large total number of tests required for full permutation and combination, which will include a large number of invalid data combinations. For example, for an instance containing four parameters A, B, C, and D, each parameter can have two values. Table 1 lists the test cases required for full permutation and combination. The upper part of Table 1 lists the basic values: the basic values ​​contain a total of 4 parameters, namely A, B, C, and D. Each parameter has 2 values, for example, parameter A has a1 and a2. The lower part of Table 1 lists the test cases required for full permutation and combination: using the traditional full permutation and combination method, different values ​​of each parameter will generate different test cases, resulting in a total of 2 4= 16 test cases, as shown in the table, with each row representing one test case. When using full permutations and combinations, if the number of parameters changes from 4 to 10 while the value of each parameter remains 2, the total number of test cases changes from 16 to 2. 10 =1024. If the parameter has more possible values, the total number of test cases will increase exponentially, which will pose a huge challenge to computing and storage resources.

[0047] Table 1. Base values ​​and permutations / combinations

[0048]

[0049] This invention proposes a test case quantity optimization method (T-way generalization method) to address the problem of massive generalization data. The T-way generalization method creates test cases by combining several parameters t (t typically takes the value 2 or 3). For example, for a system with n parameters (assuming each parameter has d values), all combinations of these parameters are d. n Due to limitations in computing and storage resources, it is impossible to generate test cases for all combinations. The T-way combination generalization method can significantly reduce the number of test cases while achieving near-ideal test results.

[0050] Continuing with the basic scenario of the example above: four parameters A, B, C, and D, each of which can take two values, the following example, t=3 (the number of generalization combinations is 3), illustrates the effectiveness of the T-way method in reducing the number of test cases:

[0051] The first step is to list all possible combinations of the three parameters: ABC, ACD, and BCD.

[0052] The second step is to calculate the number of test cases when A, B, and C are combined (see Table 2). Here, there are three types of parameters, A, B, and C. Different values ​​for each type of parameter will generate different test cases, so there are a total of 2... 3 =8 test cases, which are the lower half of Table 2 (each row represents one test case). The value of parameter D can be arbitrarily chosen (d1 or d2).

[0053] Table 2 shows the number of test cases for combinations (ABC).

[0054]

[0055] Third, using the same method, calculate the number of test cases when ABD is a combination (see Table 3). In this case, any change in the value of any one of the three parameters (A, B, and D) will generate different test cases. Therefore, there are a total of 2... 3=8 test cases, where the value of parameter C can be arbitrarily chosen (c1 or c2). Note that the test cases {a1,b1,c1,d1}, {a2,b1,c1,d1}, and {a2,b2,c2,d1} have already appeared in step two. Therefore, delete these three duplicate test cases, leaving only the remaining 5 valid test cases.

[0056] Table 3ABD shows the number of test cases when combined.

[0057]

[0058] Fourth, using the same method, calculate the number of test cases when ACD is a combination (see Table 4). In this case, any change in the value of any one of the three parameters ACD will generate different test cases. Therefore, there are a total of 2... 3 =8 test cases, where the value of parameter B can be any choice (b1 or b2). Upon closer inspection, it is noticed that all 8 test cases generated by ACD have been generated in steps 2 and 3. Therefore, all 8 duplicate test cases are deleted.

[0059] Table 4 shows the number of test cases when ACD is combined.

[0060]

[0061] The fifth step is to calculate the number of test cases when BCD is a combination (see Table 5). In this case, any change in the value of any one of the three parameters (B, C, D) will generate different test cases. Therefore, there are a total of 2... 3 =8 test cases, where the value of parameter A can be any choice (a1 or a2). Upon closer inspection, it is noticed that all 8 test cases generated by BCD have been generated in steps 2 and 3. Therefore, all 8 duplicate test cases are deleted.

[0062] Table 5BCD shows the number of test cases when combined.

[0063]

[0064] Finally, by combining the test cases generated in the above steps (Tables 2 to 5), we obtain a total of 8 + 5 = 13 valid test cases, as shown in the lower half of Table 6, which consists of 13 rows, with each row representing a unique test case.

[0065] Table 6 Final Valid Test Cases

[0066]

[0067] Summary: When using the full permutation and combination method, a total of 2... 4Initially, there were 16 test cases. However, using the T-way combination generalization method, when the number of generalization combinations t=3, the number of test cases was reduced to 13, saving 18.75% of test cases. Applying the T-way combination generalization method to scenario generalization can similarly reduce the number of test cases, thereby avoiding a large number of duplicate and invalid scenario test cases.

[0068] In one embodiment, the nested loop optimization of the test case generation process in the step of generalizing parameters includes:

[0069] Create an outer nested table and an inner nested table, where the elements in the inner nested table are a subset of the elements in the outer nested table;

[0070] Create an index that corresponds to an element contained in the inner nested table, and the index enables the element to be located directly in both the outer and inner nested tables;

[0071] The outer nested table executes a loop, and the elements are traversed and assigned values ​​within the loop of the outer nested table;

[0072] When the inner nested table executes a loop, the element is directly retrieved from the outer nested table through the index during the loop, thereby reducing the number of times the nested loop is executed.

[0073] Generalization algorithms are primarily based on nested loops. In existing technologies, the inner loop iterates through itself every time the outer loop executes, resulting in a large amount of repetitive computation and consequently, low overall efficiency in generalization. The following section will combine... Figure 2 This section introduces an example of nested loop optimization. Figure 2 This illustration reveals a schematic diagram of nested loop optimization in a method for generating generalized intelligent driving scenarios according to an embodiment of the present invention. This example provides a better understanding of the nested loop optimization process of the present invention. Overall, the present invention employs index search to optimize nested loops to achieve generalization; by increasing the index, generalization efficiency can be improved and time consumption reduced. Figure 2 In the example shown, the implementation of the nested index loop is as follows:

[0074] Optimization can be achieved by reducing the number of iterations, thereby improving the performance of nested loops. For example... Figure 2 As shown, prepare two tables, t1 and t2. t1 is the outer nested table, corresponding to the outer nested body, containing three elements: a, b, and c. t2 is the inner nested table, corresponding to the inner nested body, containing two elements: a and b. The elements a and b in the inner nested table are subsets of the elements a, b, and c contained in the outer nested table.

[0075] According to the present invention, an index is introduced. An index is a separate, physical storage structure that sorts the values ​​of one or more columns in a database table. It is a collection of values ​​from one or more columns in a table and a corresponding list of logical pointers to the data pages in the table that physically identify these values. The index acts like a table of contents in a book, allowing you to quickly find the desired content based on page numbers. The index provides pointers to data values ​​stored in specified columns of the table, and these pointers are then sorted according to a specified sorting order. Using the index to find a specific value, you can then follow the pointers to find the row containing that value. This allows for rapid access to specific information in the data table.

[0076] Therefore, in Figure 2 The nested loop shown incorporates an index targeting elements a and b in the inner nested table t2. When the loop iterates through the outer nested table t1, elements a and b are iterated and assigned values ​​within the outer nested table. Then, when the loop iterates through the inner nested table t2, elements a and b in t2 are fields that have already been iterated through in the outer nested table t1. At this point, the values ​​of these fields can be directly obtained through the index and used to retrieve values ​​from the inner nested table t2. In this process, the index plays a direct locating role. Through this direct locating via the index, the number of nested loop calculations is reduced by a significant margin due to the indexing technique, thereby greatly reducing the time required for generalization calculations.

[0077] S104. Save the generalized parameters to the parameter database. The generalized parameters are saved in the parameter database as a parameter list, which includes the parameters, their configuration information, and test cases. The parameter list corresponds to the requirements of the generalization scenario. In one embodiment, when saving the generalized parameters to the parameter database, the task ID, user ID, and timestamp are saved together with the corresponding parameter list in the parameter database. The parameter list and the requirements of the generalization scenario are established in a one-to-one correspondence through the task ID.

[0078] S105. Generalized Parameter Invocation. After the parameter generalization process is completed, the corresponding parameter list is retrieved from the parameter database according to the requirements of the generalization scenario. In one embodiment, after the parameter generalization is completed, a completion instruction is generated, which contains a task ID. Upon receiving this completion instruction, it indicates that the parameter generalization has been completed. At this time, the completion instruction can be parsed to obtain the task ID, and the corresponding generalized parameters and parameter list can be invoked based on the task ID.

[0079] S106. Generate scene files. The scene generation service is invoked. It reads the parameter list, generates scene files based on the parameter list, and saves the scene files in the scene library. In one embodiment, if the parameter list contains multiple lines, the scene generation service generates a scene file for each line of the parameter list. The scene files need to be in a format that the scene library can recognize. In one embodiment, the scene files are saved in XML format.

[0080] S107. Generate an intelligent driving scenario by calling the virtual simulation service. The virtual simulation service retrieves and loads the corresponding scenario file from the scenario library, parses the scenario file, and generates a virtual intelligent driving scenario. The XML format scenario file is parsed and loaded into the virtual simulation service to generate the required scenario.

[0081] According to one embodiment of the present invention, a device for generating generalized intelligent driving scenarios is also proposed. Figure 3 A schematic diagram of a generalized intelligent driving scenario generation device according to an embodiment of the present invention is shown. (Reference) Figure 3 As shown, the device for generating the generalized intelligent driving scenario includes: a client 101, a generalization processing server 102, a parameter database 103, a main control server 104, and a scenario library 105.

[0082] Client 101 can be a personal terminal such as a personal computer, a local server, a mobile phone, or other handheld devices. Client 101 has an interactive interface, which can be an application (such as an App), a terminal program, or a web-based UI. Client 101 obtains the requirements of the generalized scenario through the interactive interface. The requirements of the generalized scenario correspond to the parameters that need to be generalized, and the parameters to be generalized are determined based on the requirements of the generalized scenario. In one embodiment, when obtaining the requirements of the generalized scenario, a unique task ID is assigned to the requirement, and the user ID of the user who made the requirement and the timestamp of the requirement are recorded simultaneously. The process by which client 101 obtains the requirements of the generalized scenario is the same as the process described in step S101 above, and will not be repeated here.

[0083] Client 101 sends the parameters to be generalized to generalization server 102. Generalization server 102 retrieves parameter configuration information from parameter database 103 and configures the parameters. Parameter configuration information includes: the number of generalization combinations, the parameter value range, and the parameter step size. Generalization server 102 performs generalization processing on the parameters, including optimizing the number of test cases generated based on the parameters and optimizing the nested loop process of test case generation. Generalization server 102 saves the generalized parameters to parameter database 103 in the form of a parameter list. The parameter list includes parameters, parameter configuration information, and test cases, and the parameter list corresponds to the requirements of the generalization scenario. The process of generalization server 102 retrieving parameter configuration information from parameter database 103 and configuring the parameters is the same as the parameter configuration process in step S102 mentioned above. The generalization processing performed by the generalization server 102, including the optimization of the number of parameter-generated test cases and the nested loop optimization of the test case generation process, is the same as the process of optimizing the number of parameter-generated test cases (T-way generalization method) and the process of nested loop optimization of the test case generation process (index nested loop) described in step S103 above, and will not be repeated here. In one embodiment, when the generalization server 102 saves the generalized parameters to the parameter database 103, the task ID, user ID, and timestamp are stored together with the corresponding parameter list in the parameter database, wherein the parameter list and the requirements of the generalization scenario are established in a one-to-one correspondence through the task ID.

[0084] After generalizing the parameters and saving the generalized parameters to the parameter database 103, the generalization processing server 102 sends a completion command to the master control server 104. The master control server 104 parses the completion command and retrieves the corresponding parameter list from the parameter database 103. The master control server 104 then calls the scene generation service, which reads the parameter list, generates a scene file based on the parameter list, and saves the scene file in the scene library 105. In one embodiment, the completion command sent by the generalization processing server 102 to the master control server 104 includes a task ID. The master control server 104 parses the completion command to obtain the task ID and retrieves the parameter list corresponding to the task ID from the parameter database 103. In another embodiment, the scene file generated by the scene generation service also includes a task ID. In one embodiment, the scene generation service called by the master control server 104 generates one scene file for each line of the parameter list. If the parameter list has multiple lines, multiple scene files are generated accordingly. The scene file needs to be in a format that the scene library 105 can recognize. In one embodiment, the scene file is saved in XML format. The scene generation service can be deployed directly on the main control server 104, or it can be deployed on a network that communicates with the main control server 104.

[0085] Client 101 invokes a virtual simulation service, which retrieves and loads the corresponding scene file from scene library 105, parses the scene file, and generates a virtual intelligent driving scenario. The XML-formatted scene file is parsed and loaded into the virtual simulation service to generate the required scenario. In one embodiment, the scene file also includes a task ID. The virtual simulation service invoked by client 101 retrieves and loads the corresponding scene file from scene library 105 based on the task ID, parses the scene file, and generates a virtual intelligent driving scenario. The virtual simulation service can be deployed on client 101 or on a network communicating with client 101.

[0086] Client 101, generalization processing server 102, parameter database 103, master control server 104, and scene library 105 are deployed on an interconnected network, such as the Internet. In one embodiment, client 101, generalization processing server 102, parameter database 103, master control server 104, and scene library 105 are deployed in a distributed manner and communicate with each other through a communication network. The virtual simulation service can be deployed on client 101 or deployed separately on the network. The scene generation service can be deployed on master control server 104 or deployed separately on the network. In one embodiment, master control server 104 is hosted in the cloud, becoming a cloud server.

[0087] The generalized intelligent driving scenario generation method and apparatus of the present invention solve the following problems: addressing the current situation where most intelligent driving companies lack the ability to automatically generalize intelligent driving scenarios, thus improving the test coverage of the scenario library; and solving the problems of large data volume, slow speed, and low efficiency of existing intelligent driving scenario generalization methods. The generalized intelligent driving scenario generation method and apparatus of the present invention have the following beneficial effects: based on the T-way generalization method, improving scenario test coverage; employing index nesting to build a generalization logic algorithm, improving the efficiency of simulation scenario generalization; and adopting an end-to-end full-process technical architecture to automatically generalize intelligent driving scenarios, replacing manual parameter modification and saving significant manpower and time costs required for scenario generalization.

[0088] It should also be noted that the embodiments listed above are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or modifications made thereto are those that can be directly derived or easily conceived by those skilled in the art from the content disclosed in the present invention, and should all fall within the protection scope of the present invention. The above embodiments are provided for those skilled in the art to implement or use the present invention. Those skilled in the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited to the above embodiments, but should be the maximum scope conforming to the innovative features mentioned in the claims.

Claims

1. A method for generating generalized intelligent driving scenarios, characterized in that, include: Obtain the requirements of the generalization scenario, which correspond to the parameters that need to be generalized, and determine the parameters that need to be generalized based on the requirements of the generalization scenario; Configure the parameters. For the determined parameters that need to be generalized, configure the number of generalization combinations, the range of parameter values, and the step size of the parameters. The parameters that need to be generalized include environmental visibility, vehicle speed, obstacle vehicle speed, road surface adhesion coefficient, and following distance. The number of generalization combinations ranges from 2 to 4. The parameters are generalized, and the generalization process includes optimizing the number of test cases generated based on the parameters and optimizing the test case generation process using nested loops. The generalized parameters are saved to a parameter database. The generalized parameters are saved in the parameter database in the form of a parameter list. The parameter list includes parameters, parameter configuration information and test cases. The parameter list corresponds to the requirements of the generalization scenario. After the parameters have been generalized, the corresponding parameter list is retrieved from the parameter database according to the requirements of the generalization scenario. Generate a scene file, call the scene generation service, the scene generation service reads the parameter list, generates a scene file based on the parameter list, and saves the scene file in the scene library; Generate intelligent driving scenarios, call virtual simulation services, the virtual simulation services call and load the corresponding scenario files from the scenario library, parse the scenario files to generate virtual intelligent driving scenarios; Among the steps of generalizing the parameters, optimizing the number of test cases generated from the parameters includes: Given the number of generalization combinations n, for all parameters, list all possible combinations of selecting n parameters from all parameters to form a generalization combination; For each generalization combination, list all test cases based on the value range and step size of each parameter; Compare the test cases for all generalization combinations and remove duplicate test cases; The remaining non-repeating test cases are used as quantity-optimized test cases; The step of generalizing parameters includes nested loop optimization of the test case generation process, which includes: Create an outer nested table and an inner nested table, where the elements in the inner nested table are a subset of the elements in the outer nested table; Create an index that corresponds to an element contained in the inner nested table, and the index enables the element to be located directly in both the outer and inner nested tables; The outer nested table executes a loop, and the elements are traversed and assigned values ​​within the loop of the outer nested table; When the inner nested table executes a loop, the element is directly retrieved from the outer nested table through the index during the loop, thereby reducing the number of times the nested loop is executed.

2. The method for generating generalized intelligent driving scenarios as described in claim 1, characterized in that, When obtaining the requirements of a generalized scenario, assign a unique task ID to the requirement of that generalized scenario, and at the same time record the user ID of the user who made the requirement and the timestamp of the requirement. When saving the generalized parameters to the parameter database, the task ID, user ID, and timestamp are stored together with the corresponding parameter list in the parameter database. The parameter list and the requirements of the generalized scenario are established in a one-to-one correspondence through the task ID.

3. The method for generating generalized intelligent driving scenarios as described in claim 1, characterized in that, The scene generation service generates a scene file for each line of the parameter list, and the scene files are saved in XML format.

4. A device for generating generalized intelligent driving scenarios, characterized in that, include: Client, generalization processing server, parameter database, master control server, and scene library; The client has an interactive interface. The client obtains the requirements of the generalization scenario through the interactive interface. The requirements of the generalization scenario correspond to the parameters that need to be generalized. The parameters that need to be generalized are determined according to the requirements of the generalization scenario. The client sends the parameters to be generalized to the generalization processing server. The generalization processing server retrieves parameter configuration information from the parameter database and configures the parameters. The parameter configuration information includes: the number of generalization combinations, the value range of the parameters, and the step size of the parameters. The parameters to be generalized include environmental visibility, vehicle speed, obstacle vehicle speed, road surface adhesion coefficient, and following distance. The number of generalization combinations ranges from 2 to 4. The generalization processing server performs generalization processing on the parameters. This generalization processing includes optimizing the number of test cases generated based on the parameters and optimizing the test case generation process using nested loops. The generalization processing server saves the generalized parameters to the parameter database. The generalized parameters are stored in the parameter database as a parameter list, which includes the parameters, parameter configuration information, and test cases. The parameter list corresponds to the requirements of the generalization scenario. After completing parameter generalization and saving the generalized parameters to the parameter database, the generalization server sends a completion command to the master control server. The master control server parses the completion command and retrieves the corresponding parameter list from the parameter database. The master control server then calls the scene generation service, which reads the parameter list, generates a scene file based on the parameter list, and saves the scene file in the scene library. The generalization server optimizes the number of test cases generated by the parameters. Specifically, based on the number of generalization combinations n, the generalization server lists all possible combinations of n parameters selected from all parameters to form a generalization combination. For each generalization combination, it lists all test cases based on the value range and step size of each parameter. Finally, it compares all generalization combinations. Test cases are generated by removing duplicate test cases; the remaining unique test cases are then used as a quantity-optimized test case set. The generalization server optimizes the test case generation process using nested loops. Specifically, the server establishes an outer nested table and an inner nested table, where the elements in the inner nested table are a subset of the elements in the outer nested table. An index is created corresponding to the elements in the inner nested table, allowing direct location of the elements within both the outer and inner nested tables. The outer nested table iterates through and assigns values ​​to the elements within it. The inner nested table also iterates through and directly retrieves values ​​from the outer nested table using the index, reducing the number of nested loop executions. The client calls the virtual simulation service, which retrieves and loads the corresponding scene file from the scene library, parses the scene file, and generates a virtual intelligent driving scene.

5. The apparatus for generating generalized intelligent driving scenarios as described in claim 4, characterized in that, When the client obtains the requirements of the generalized scenario, it assigns a unique task ID to the requirements of the generalized scenario, and at the same time records the user ID of the user who made the requirement and the timestamp of the requirement. When the generalization processing server saves the generalized parameters to the parameter database, the task ID, user ID, and timestamp are stored together with the corresponding parameter list in the parameter database. The parameter list and the requirements of the generalization scenario are established in a one-to-one correspondence through the task ID.

6. The apparatus for generating generalized intelligent driving scenarios as described in claim 5, characterized in that, The completion instruction sent by the generalization processing server to the main control server includes the task ID. The main control server parses the completion instruction to obtain the task ID and retrieves the corresponding parameter list from the parameter database based on the task ID.

7. The apparatus for generating generalized intelligent driving scenarios as described in claim 5, characterized in that, The scene file generated by the scene generation service includes a task ID. The virtual simulation service called by the client calls and loads the corresponding scene file from the scene library based on the task ID, and parses the scene file to generate a virtual intelligent driving scene.

8. The apparatus for generating generalized intelligent driving scenarios as described in claim 4, characterized in that, The scene generation service called by the master server generates a scene file for each line of the parameter list, and the scene files are saved in XML format.

9. The apparatus for generating generalized intelligent driving scenarios as described in claim 4, characterized in that, The client, generalization processing server, parameter database, master control server, and scene library are deployed in a distributed manner and communicate with each other through a communication network.

10. The apparatus for generating generalized intelligent driving scenarios as described in claim 9, characterized in that, The virtual simulation service is deployed on the client or on the network; the scene generation service is deployed on the main control server or on the network.

11. The apparatus for generating generalized intelligent driving scenarios as described in claim 10, characterized in that, The main control server is located in the cloud.

Images