Test methods, devices and electronic equipment for intelligent driving systems

Abstract

This application provides a testing method, apparatus, and electronic device for an intelligent driving system. The method includes: acquiring high-precision map data and multiple types of first sensor data for a test scenario, wherein the multiple types of first sensor data are data collected by various types of sensors in a vehicle in the test scenario; acquiring event text related to a road event to be tested; predicting, based on the high-precision map data, the multiple types of first sensor data, and the event text, multiple types of second sensor data collected by the various types of sensors in the vehicle when the road event to be tested occurs in the test scenario; processing the multiple types of second sensor data according to the intelligent driving system to obtain driving commands output by the intelligent driving system; and testing the intelligent driving system according to the driving commands. This improves the accuracy of intelligent driving system testing.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a testing method, apparatus and electronic device for an intelligent driving system. Background Technology

[0002] With the development of the field of intelligent driving, intelligent driving assistance technology is increasingly involved in the daily driving of vehicles. Therefore, the safety testing of intelligent driving systems is particularly important.

[0003] In related technologies, the safety of intelligent driving systems can be tested through offline driving scenarios. However, offline driving scenarios can only reproduce conventional driving scenarios and cannot reproduce some extreme driving scenarios, resulting in a limited range of driving scenarios for intelligent driving system testing and consequently lower accuracy in intelligent driving system testing. Summary of the Invention

[0004] This application provides a testing method, apparatus, and electronic device for intelligent driving systems, which addresses the technical problem that the limited types of driving scenarios tested in existing intelligent driving system tests result in low accuracy in intelligent driving system evaluations.

[0005] Firstly, this application provides a testing method for an intelligent driving system, the method comprising:

[0006] Acquire high-precision map data and multiple types of first sensor data for the test scenario. The multiple types of first sensor data are data collected by various types of sensors in the vehicle in the test scenario.

[0007] Obtain the event text related to the road event to be tested;

[0008] Based on high-precision map data, the multiple types of first sensor data, and the event text, predict the multiple types of second sensor data collected by the various types of sensors in the vehicle when the road event to be tested occurs in the test scenario.

[0009] The intelligent driving system processes the data from the various second sensors to obtain driving commands output by the intelligent driving system, and then tests the intelligent driving system based on the driving commands.

[0010] In one possible implementation, based on high-precision map data, the multiple types of first sensor data, and the event text, the system predicts the types of second sensor data collected by various types of sensors in the vehicle when the road event to be tested occurs in the test scenario. These include:

[0011] Based on the high-precision map data and the multi-type first sensor data, the scene characteristics corresponding to the test scene are determined;

[0012] Based on the event text, determine the characteristics of the road event to be tested corresponding to the event text;

[0013] Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, predict the multiple types of second sensor data.

[0014] In one possible implementation, the scene features corresponding to the test scene are determined based on the high-precision map data and the multiple types of first sensor data, including:

[0015] Determine the map features corresponding to the high-precision map data, and the sensor features corresponding to each type of the first sensor data, wherein the sensor features and the map features include time-series information;

[0016] The scene features are obtained by fusing the map features and multiple sensor features.

[0017] In one possible implementation, predicting the multiple types of second sensor data based on the scene features, the road event features to be measured, and the multiple types of first sensor data includes:

[0018] The scene features and the road event features to be tested are fused with the sensor features corresponding to each type of the first sensor data to obtain the fused features corresponding to each type of the first sensor data.

[0019] For any type of first sensor data, predict the second sensor data corresponding to the first sensor data based on the fusion features corresponding to the first sensor data.

[0020] In one possible implementation, the intelligent driving system processes the data from the multiple types of second sensors to obtain driving commands output by the intelligent driving system, including:

[0021] The various types of second sensor data are input to multiple sensors connected to the intelligent driving system;

[0022] The system acquires driving instructions generated by the intelligent driving system, which are determined by the intelligent driving system based on multiple types of second sensor data fed back from multiple connected sensors.

[0023] In one possible implementation, the various types of second sensor data are input to multiple sensors connected to the intelligent driving system, including:

[0024] Determine the sensor type for each of the aforementioned sensors;

[0025] According to the sensor type, input the corresponding second sensor data to each sensor.

[0026] In one possible implementation, testing the intelligent driving system according to the driving command includes:

[0027] Acquire at least one set of target driving behaviors when the road event to be tested occurs in the test scenario;

[0028] If the driving behavior indicated by the driving command matches the at least one set of target driving behaviors, then the intelligent driving system passes the test when the test road event occurs in the test scenario.

[0029] Secondly, this application provides a testing apparatus for an intelligent driving system, which includes a first acquisition module, a second acquisition module, a prediction module, and a processing module, wherein:

[0030] The first acquisition module is used to acquire high-precision map data of the test scene and multiple types of first sensor data, wherein the multiple types of first sensor data are data collected by various types of sensors in the vehicle in the test scene;

[0031] The second acquisition module is used to acquire event text related to the road event to be tested;

[0032] The prediction module is used to predict, based on high-precision map data, the multiple types of first sensor data, and the event text, the multiple types of second sensor data collected by the various types of sensors in the vehicle when the road event to be tested occurs in the test scenario.

[0033] The processing module is used to process the data from the multiple types of second sensors according to the intelligent driving system to obtain the driving instructions output by the intelligent driving system, and to test the intelligent driving system according to the driving instructions.

[0034] In one possible implementation, the prediction module is specifically used for:

[0035] Based on the high-precision map data and the multi-type first sensor data, the scene characteristics corresponding to the test scene are determined;

[0036] Based on the event text, determine the characteristics of the road event to be tested corresponding to the event text;

[0037] Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, predict the multiple types of second sensor data.

[0038] In one possible implementation, the prediction module is specifically used for:

[0039] Determine the map features corresponding to the high-precision map data, and the sensor features corresponding to each type of the first sensor data, wherein the sensor features and the map features include time-series information;

[0040] The scene features are obtained by fusing the map features and multiple sensor features.

[0041] In one possible implementation, the prediction module is specifically used for:

[0042] The scene features and the road event features to be tested are fused with the sensor features corresponding to each type of the first sensor data to obtain the fused features corresponding to each type of the first sensor data.

[0043] For any type of first sensor data, predict the second sensor data corresponding to the first sensor data based on the fusion features corresponding to the first sensor data.

[0044] In one possible implementation, the processing module is specifically used for:

[0045] The various types of second sensor data are input to multiple sensors connected to the intelligent driving system;

[0046] The system acquires driving instructions generated by the intelligent driving system, which are determined by the intelligent driving system based on multiple types of second sensor data fed back from multiple connected sensors.

[0047] In one possible implementation, the processing module is specifically used for:

[0048] Determine the sensor type for each of the aforementioned sensors;

[0049] According to the sensor type, input the corresponding second sensor data to each sensor.

[0050] In one possible implementation, the processing module is specifically used for:

[0051] Acquire at least one set of target driving behaviors when the road event to be tested occurs in the test scenario;

[0052] If the driving behavior indicated by the driving command matches the at least one set of target driving behaviors, then the intelligent driving system passes the test when the test road event occurs in the test scenario.

[0053] Thirdly, this application provides an electronic device, including: a processor and a memory;

[0054] The memory stores computer-executed instructions;

[0055] The processor executes computer execution instructions stored in the memory, causing the processor to perform test methods for intelligent driving systems, as described in the first aspect and various possible aspects of the first aspect.

[0056] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used in the first aspect and various possible testing methods for intelligent driving systems involved in the first aspect.

[0057] This application provides a testing method, apparatus, and electronic device for an intelligent driving system. The electronic device can acquire high-precision map data and multiple types of first sensor data for a test scenario. The multiple types of first sensor data refer to data collected by various types of sensors in the vehicle within the test scenario. Event text related to a road event under test is obtained. Based on the high-precision map data, the multiple types of first sensor data, and the event text, multiple types of second sensor data collected by various types of sensors in the vehicle when the road event under test occurs in the test scenario is predicted. The intelligent driving system processes the multiple types of second sensor data to obtain driving commands output by the intelligent driving system, and then tests the intelligent driving system based on these driving commands. In this method, because the electronic device can predict the multiple types of second sensor data when the road event under test occurs in the current scenario (test scenario) based on the multiple types of first sensor data and the event text related to the road event under test, and can obtain driving commands from the intelligent driving system through the multiple types of second sensor data, the electronic device can simulate sensor data of the vehicle in different driving scenarios by adjusting the event text. This improves the variety of driving scenarios tested by the intelligent driving system and enhances the accuracy of the intelligent driving system test. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of the present disclosure;

[0059] Figure 2 A flowchart illustrating a testing method for an intelligent driving system provided in an embodiment of this application;

[0060] Figure 3 This application provides a schematic diagram illustrating the relationship between a road event to be tested and the event text.

[0061] Figure 4 A schematic diagram illustrating a process for predicting second sensor data provided in an embodiment of this application;

[0062] Figure 5 This application provides a schematic diagram of a process for generating driving instructions.

[0063] Figure 6 This is a schematic diagram illustrating a method for predicting multiple types of second sensor data provided in an embodiment of this application;

[0064] Figure 7 This is a schematic diagram illustrating a process for acquiring scene features, provided in an embodiment of this application.

[0065] Figure 8 A schematic diagram illustrating a process for predicting second sensor data provided in an embodiment of this application;

[0066] Figure 9 A schematic diagram illustrating the training process of a driving scene generation model provided in an embodiment of this application;

[0067] Figure 10 A schematic diagram of the structure of a testing device for an intelligent driving system provided in an embodiment of this application;

[0068] Figure 11 A schematic diagram of the hardware structure of the electronic device provided in this application. Detailed Implementation

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0070] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0071] For ease of understanding, the concepts involved in the embodiments of this application will be explained below.

[0072] High-precision maps: High-precision maps can be used for autonomous driving assistance. The relative accuracy of high-precision maps is at the centimeter level. In addition, high-precision maps include rich real-time traffic dynamic information such as vehicle lines, road signs, traffic signs, traffic lights, lane curvature, slope and lane level. High-precision maps mainly serve the intelligent driving system to judge, make decisions and control the driving environment.

[0073] Test Scenario: The test scenario can be an initial scenario associated with high-precision map data. For example, the test scenario can be a single-lane scenario, a two-lane scenario, or a three-lane scenario. After the high-precision map data is determined, the main body of the test scenario will not change (it will not suddenly change from three lanes to single lanes). In practical applications, multiple test scenarios can be selected as initial scenarios to test the same road event to be tested.

[0074] Event text: The event text can be text related to the road event to be tested. For example, the road event to be tested can be an overtaking event, a lane changing event, an emergency stop event upon encountering an obstacle, etc. This application embodiment does not limit this, and the event text can be edited based on the road event to be tested. For example, the event text can be: an obstacle is added in the middle of the lane, the traffic light turns red, changing lanes to the left or overtaking the vehicle in front, etc. The event text can also be a combination of multiple events to be tested, such as: an obstacle is added in the middle of the lane, swerving to the right to avoid danger, etc. This application embodiment does not limit this, and the electronic device can build different driving scenarios based on the event text.

[0075] Multi-sensor data: Multi-sensor data can be data collected by various types of sensors in the vehicle. For example, multi-sensor data can include: video data captured by cameras, inertial data collected by inertial measurement units (IMUs), lidar data collected by lidar sensors, millimeter-wave data collected by millimeter-wave radar sensors, and position data collected by positioning devices, etc. It should be noted that the above content is only an example of multi-sensor data and is not intended to limit the types of multi-sensor data. In practical applications, multi-sensor data can be any data collected by the vehicle, and this application embodiment does not limit it.

[0076] Driving commands: Driving commands are driving operation instructions output by the intelligent driving system. Each driving command corresponds to a set of driving actions. For example, if the driving command is a brake command, the corresponding driving action is braking; if the driving command is a command to turn on the left turn signal or a command to rotate the steering wheel counter-clockwise (turn the steering wheel to the left), the corresponding set of driving actions is changing lanes to the left.

[0077] In the field of intelligent driving, with the increasing integration of intelligent driving assistance technologies into daily vehicle operation, testing the safety of intelligent driving systems is particularly crucial. Currently, the safety of intelligent driving systems can be tested through offline driving scenarios. For example, different types of driving scenarios can be set up offline, and the intelligent driving system can control the vehicle to drive within these scenarios to assess its safety. However, offline driving scenarios are typically relatively conventional, such as pedestrian detection, reversing into parking spaces, and lane changing / overtaking scenarios. These offline scenarios cannot replicate some extreme driving situations, resulting in a limited range of driving scenario types for intelligent driving system testing, and consequently, lower accuracy in testing.

[0078] To address the technical problems in related technologies, this application provides a testing method for an intelligent driving system. An electronic device can acquire a high-precision map of the test scenario and various types of first sensor data actually collected by the vehicle in the test scenario. It can also acquire event text related to a road event under test. Based on the high-precision map data, the various types of first sensor data, and the event text, it predicts various types of second sensor data collected by the vehicle's sensors when the road event under test occurs in the test scenario. This second sensor data is then input to multiple sensors connected to the intelligent driving system, and driving commands generated by the intelligent driving system based on the second sensor data are obtained. The intelligent driving system is then tested according to these driving commands. Since the second sensor data represents the data collected by the vehicle's sensors when the road event under test occurs, as predicted by the electronic device, the electronic device can determine the driving commands output by the intelligent driving system when the road event occurs. This allows for online simulation testing of the intelligent driving system, improving the safety of the testing. Furthermore, the electronic device can adjust the event text to simulate sensor data from different driving scenarios, thereby increasing the types of driving scenarios tested and improving the accuracy of the intelligent driving system test.

[0079] Based on the road information provided by the collected high-precision map data and combined with multimodal vehicle perception data, this application proposes to extract features from the map and vehicle perception data through a deep generative model and perform feature fusion. At the same time, according to the prior road scene settings, corresponding dynamic scene driving data is generated. The generated scene data can be used as the perception information input of the intelligent assisted driving system, thereby evaluating the assisted driving system's ability to respond to real-time changing scenes and constructing an assisted driving system evaluation policy platform.

[0080] Specifically, this simulation platform first collects historical driving road data and high-precision map data from users. Then, it manually annotates scene description text. Subsequently, the collected visual lane driving data, high-precision map data, multimodal sensor perception data, and manually annotated road scene description text are used as input to train a deep learning prediction model. After training, the model can generate road driving scene data consistent with the human-generated prior scene description, based on initial short-term visual driving data, high-precision map data, multimodal sensor perception data, and human-generated prior scene settings (text format). Finally, the dynamic scene data generated by the model is used as input to the intelligent assisted driving system to test the system's responsiveness and outputs intelligent quantitative indicators based on the system's response results. This platform can automatically, efficiently, and in a multi-scenario variable manner comprehensively determine the capabilities of intelligent driving assistance systems, providing reliable quantitative indicators for related research in the autonomous driving / assisted driving industry and greatly promoting the development of the intelligent assisted driving industry.

[0081] Below, in conjunction with Figure 1 The application scenarios of the embodiments of this disclosure will be described.

[0082] Figure 1 This is a schematic diagram illustrating an application scenario provided by an embodiment of this disclosure. Please refer to [link / reference]. Figure 1 This includes electronic devices and intelligent driving systems. The electronic devices can predict vehicle sensor data in simulated driving scenarios and input this data into the intelligent driving system. After receiving the sensor data, the intelligent driving system can determine the current driving scenario and make driving decisions accordingly. The electronic devices can test these driving decisions to determine whether the intelligent driving system performs well in that scenario. In this way, the electronic devices can predict sensor data in different driving scenarios, allowing for testing of the intelligent driving system's driving decisions across various scenarios. This wide range of driving scenarios tested improves the accuracy of the intelligent driving system's testing.

[0083] It should be noted that, Figure 1 The examples provided are merely illustrations of application scenarios for the embodiments of this application and are not intended to limit the application scenarios of the embodiments of this application.

[0084] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0085] Figure 2This is a flowchart illustrating a testing method for an intelligent driving system provided in an embodiment of this application. Please refer to [link / reference]. Figure 2 The method may include:

[0086] S201. Acquire high-precision map data and multiple types of primary sensor data for the test scenario.

[0087] The execution subject of this application embodiment can be an electronic device or a test device for an intelligent driving system installed in an electronic device. The test device for the intelligent driving system can be implemented by software or by a combination of software and hardware. This application embodiment does not limit this.

[0088] Among them, the multiple types of first sensor data can be data collected by various types of sensors in the vehicle in the test scenario. For example, if the test scenario is that the vehicle is driving at a constant speed in the middle lane of a three-lane road, then the multiple types of first sensors can be video data collected by the vehicle's camera device, inertial data collected by the inertial sensor, lidar data collected by the lidar sensor, etc. in this driving scenario.

[0089] It should be noted that the data from multiple first sensors are related to the test scenario. If the test scenario changes, the data from multiple first sensors will also change.

[0090] Optionally, the electronic device can receive high-precision map data and multiple types of first sensor data of the test scenario sent by the server. For example, during the map data acquisition phase, the server can acquire multiple types of high-precision map data, as well as initial sensor data (multiple types of first sensor data) for each type of high-precision map data. After the electronic device determines the test scenario, it can receive the high-precision map data and multiple types of first sensor data of the test scenario sent by the server.

[0091] It should be noted that electronic devices can also acquire high-precision map data and multiple types of first sensor data based on any feasible implementation method, and the embodiments of this application do not limit this.

[0092] It should be noted that electronic devices can collect dynamic road data in various scenarios, including high-precision map data and vehicle perception data of various modalities (such as IMU data, LiDAR data, visual data, millimeter-wave radar data, GPS data, etc.). Then, the dynamic road data is labeled with text descriptions at 5-second intervals, which are used as input for subsequent deep model training and as ground truth data.

[0093] S202. Obtain the event text related to the road event to be tested.

[0094] Optionally, the road events to be tested may include any events that can test the intelligent driving system, such as vehicle lane changing events, vehicle overtaking events, and vehicle reversing into a parking space events. This embodiment of the disclosure does not limit this. Furthermore, since the road events to be tested are online tests, they may be some extreme driving events.

[0095] Optionally, the electronic device can receive event text related to the road event under test sent by other devices. For example, the electronic device can receive event text related to the road event under test sent by a server or other electronic devices. For example, when testing an intelligent driving system, multiple event texts related to the road event under test can be preset. Therefore, the electronic device can randomly select event text related to the road event under test, or it can select event text related to the road event under test that is relevant to the test target. This embodiment of the present disclosure does not limit this.

[0096] It should be noted that the electronic device can also obtain event text related to the road event to be tested according to any feasible implementation method. For example, the electronic device can receive event text related to the road event to be tested input by the user. This disclosure does not limit this.

[0097] It should be noted that a road event to be tested can correspond to one event text or multiple event texts, and this application embodiment does not limit this.

[0098] Below, in conjunction with Figure 3 This section explains the relationship between the road events to be tested and the event text.

[0099] Figure 3 This is a schematic diagram illustrating the relationship between a road event to be tested and its text, provided as an embodiment of this application. Please refer to... Figure 3 This includes vehicle emergency avoidance events. The event text related to these events can include: obstacles added to the driving road, obstacles added to the driving road and the road to the left, and obstacles added to all roads ahead. Each event text describes a different driving scenario; therefore, each vehicle emergency avoidance event can correspond to at least three driving scenarios, and each scenario can be tested.

[0100] It should be noted that electronic devices can acquire one or more event texts. When testing an intelligent driving system, an electronic device can test the driving scenario corresponding to one event text at a time.

[0101] S203. Based on high-precision map data, multiple types of first sensor data, and event text, predict the multiple types of second sensor data collected by various types of sensors in the vehicle when the test road event occurs in the test scenario.

[0102] The second sensor data can be data collected by various types of sensors on the vehicle when a test road event occurs in the test scenario. For example, the electronic device can change the test scenario (i.e., change the driving scenario) based on event text related to the test road event. When the test scenario changes, the sensor data collected by various types of sensors on the vehicle will also change. The electronic device can predict the data that the vehicle's sensors can collect when the test scenario changes, based on the sensor data before the test scenario changes, the event text describing the change in the test scenario, and the current high-precision map data.

[0103] Specifically, electronic devices can determine multiple types of second sensor data according to the following feasible implementation methods: determine the scene features corresponding to the test scene based on high-precision map data and multiple types of first sensor data; determine the event features of the road to be tested corresponding to the event text based on the event text; and predict multiple types of second sensor data based on the scene features, the event features of the road to be tested, and multiple types of first sensor data.

[0104] Scene features can indicate information about the current test scene. For example, scene features can be fused from information such as high-precision map data, information collected by inertial sensors, video information collected by camera devices, radar information collected by lidar sensors, and radar information collected by millimeter-wave radar sensors.

[0105] Optionally, the electronic device can acquire scene features based on a feature extraction network. For example, the electronic device can acquire map features from electronic map data and sensor features from multiple types of first sensor data based on a multimodal data feature extraction network (CNN), and then fuse these features to obtain scene features.

[0106] The road event features to be tested can indicate information about the road event. For example, the road event features to be tested can be text features. For example, the road event features to be tested can be word vectors, sentence vectors, etc., and this application embodiment does not limit this.

[0107] Electronic devices predict multiple types of second sensor data based on scene features, road event features, and multiple types of first sensor data. Each type of first sensor data can correspond to one type of second sensor data. For example, an electronic device can predict inertial sensor data in the second sensor data based on inertial sensor data from the first sensor data; an electronic device can predict video data in the second sensor data based on video data from the first sensor data; and an electronic device can predict lidar data in the second sensor data based on lidar data from the first sensor data.

[0108] It should be noted that the process of predicting the second sensor data described above can be implemented based on a driving scenario generation model. For example, an electronic device can input high-precision map data, event text related to the road event to be tested, and multiple types of first sensor data into a pre-trained driving scenario generation model. The driving scenario generation model can then output multiple types of second sensor data when the road event to be tested occurs in the test scenario.

[0109] Below, in conjunction with Figure 4 The process of predicting data from the second sensor is explained.

[0110] Figure 4 This is a schematic diagram illustrating a process for predicting second sensor data, provided as an embodiment of this application. Please refer to... Figure 4 This includes a pre-trained driving scene generation model. High-precision map data, first sensor data, and event text are input to this model. The first sensor data includes first video data, first inertial data, and first LiDAR data. The driving scene generation model can predict second sensor data, which includes second video data, second inertial data, and second LiDAR data.

[0111] In this way, the driving scenario generation model can determine the multiple sensor data collected by the vehicle when the test road event occurs in the test scenario. This can increase the types of driving scenarios tested by the intelligent driving system and improve the testing accuracy of the intelligent driving system. It should be noted that the above embodiments are only illustrative of the types of sensor data and are not intended to limit the types of sensor data.

[0112] It should be noted that after the driving scene generation model is trained, it is used to infer and generate various types of road driving scene data. In this application, the inference input of the driving scene generation model uses 5 seconds of dynamic video data, high-precision map data, manual prior control text, and other perceptual data as model input. After the controllable generation of various types of road driving scene data is achieved through prior text input, the output of the driving scene generation model serves as the sensor input of the intelligent driving assistance system. This allows for real-time dynamic monitoring of the driving decisions made by the assisted driving system based on different road scenarios, evaluating the distance between the driving system and other vehicles in the scenario, its braking response, and its driving state on the road. Based on these parameters, quantitative calculations are performed to obtain the evaluated intelligent driving assistance system's processing capabilities for different road scenarios. This achieves an objective, efficient, and safe evaluation of the driving system, thereby improving developers' and consumers' objective understanding of the intelligent driving assistance system, assisting in system optimization, reducing consumer misuse during use, lowering the accident rate, and ultimately improving the determination of liability in traffic accidents caused by problems with the intelligent driving assistance system.

[0113] S204. The intelligent driving system processes data from multiple second sensors to obtain driving commands output by the intelligent driving system.

[0114] Specifically, electronic devices can obtain driving commands output by the intelligent driving system based on the following feasible implementation: inputting multiple types of second sensor data to multiple sensors connected to the intelligent driving system, and obtaining driving commands generated by the intelligent driving system.

[0115] The driving commands can be determined by the intelligent driving system based on multiple types of second sensor data fed back from connected sensors. For example, when making driving decisions, the intelligent driving system can receive sensor data fed back from multiple connected sensors to determine the current driving scenario of the vehicle, and then determine the driving decision and output driving commands based on the driving scenario. Therefore, after the electronic device obtains the multiple types of second sensor data that the vehicle can collect when the test road event occurs in the test scenario, it can use the multiple types of second sensor data as input to the multiple types of sensors connected to the intelligent driving system, and thus obtain the driving commands generated by the intelligent driving system when the test road event occurs in the test scenario.

[0116] Specifically, electronic devices input various types of second sensor data to multiple sensors connected to the intelligent driving system. This can be achieved by determining the sensor type of each sensor and, based on that type, inputting corresponding second sensor data to each sensor. For example, a vehicle may include multiple different types of sensors, and the sensor data input by these different types of sensors will also differ. For instance, the input data for a millimeter-wave radar sensor may be millimeter-wave radar data, the input data for a camera device may be video data, and the input data for a lidar sensor may be lidar data, etc. For example, if the electronic device determines that the sensor type is an inertial sensor, it can input inertial sensor data from the second sensor data set to that sensor; if the electronic device determines that the sensor type is a lidar sensor, it can input lidar data from the second sensor data set to that sensor.

[0117] Below, in conjunction with Figure 5 The process of generating driving instructions is explained.

[0118] Figure 5 This is a schematic diagram illustrating a process for generating driving instructions, provided as an embodiment of this application. Please refer to... Figure 5This includes: second sensor data and an intelligent driving system. The second sensor data includes video data, inertial data, and LiDAR data. The intelligent driving system is connected to the camera device, inertial sensors, and LiDAR. Video data is input to the camera device, inertial data to the inertial sensors, and LiDAR data to the LiDAR. Based on this data, the intelligent driving system can determine the current driving scenario and then output driving commands. Because the electronic equipment can predict second sensor data for any driving scenario, the intelligent driving system can test a wider variety of driving scenarios, improving the reliability and accuracy of the intelligent driving system's testing.

[0119] It should be noted that the above embodiments are merely illustrative of the type of second sensor data and are not intended to limit the type of second sensor data.

[0120] S205. Test the intelligent driving system according to the driving instructions.

[0121] Specifically, electronic devices can test intelligent driving systems in the following feasible ways: acquire at least one set of target driving behaviors when a test road event occurs in the test scenario; if the driving behavior indicated by the driving command matches at least one set of target driving behaviors, then the intelligent driving system has passed the test when the test road event occurs in the test scenario.

[0122] The target driving behavior can be the standard driving behavior when the road event to be tested occurs in the test scenario. There can be one or more sets of target driving behaviors related to each road event to be tested, and this application embodiment does not limit this. For example, the test scenario includes three lanes, and the vehicle is driving in the middle lane. If the road event to be tested is the addition of a road obstacle to the driving lane, the target driving behavior corresponding to the road event to be tested can include: 1. Braking; 2. Turning on the left turn signal and changing lanes to the left; 3. Turning on the right turn signal and changing lanes to the right.

[0123] It should be noted that the target driving behavior can be a pre-set driving behavior, and the electronic device can also obtain the target driving behavior when the test road event occurs in the test scenario according to any other feasible implementation method. This application embodiment does not limit this.

[0124] For example, an intelligent driving system can generate driving commands based on data from multiple second sensors. Electronic devices can match the driving behavior corresponding to the driving command with the target driving behavior. If the driving behavior corresponding to the driving command matches any set of target driving behaviors, the electronic devices can determine that the intelligent driving system has passed the test when the test road event occurs in the test scenario.

[0125] It should be noted that during actual testing, the electronic device can score the driving instructions generated by the intelligent driving system. The score obtained by the intelligent driving system when a test road event occurs in the test scenario determines whether the intelligent driving system has passed the test. For example, the score of the intelligent driving system can be related to factors such as the time when the intelligent driving system outputs the driving instructions and the driving instructions selected by the intelligent driving system (e.g., different driving behaviors yield different scores). This application embodiment does not limit these factors.

[0126] This application provides a testing method for an intelligent driving system. An electronic device can acquire a high-precision map of the test scenario and various types of first sensor data actually collected by the vehicle in the test scenario. It can also acquire event text related to a road event under test, and determine the scenario features corresponding to the test scenario based on the high-precision map data and the various types of first sensor data. Based on the event text, it determines the characteristics of the road event under test corresponding to the event text. Based on the scenario features, the characteristics of the road event under test, and the various types of first sensor data, it predicts the various types of second sensor data collected by the vehicle's sensors when the road event under test occurs in the test scenario. It inputs the various types of second sensor data to multiple sensors connected to the intelligent driving system and acquires driving commands generated by the intelligent driving system based on the various types of second sensor data. Based on the driving commands, it tests the intelligent driving system. Thus, since the electronic device can simulate various types of second sensor data corresponding to the driving scenario online, it can test the intelligent driving system online, improving the safety of intelligent driving system testing. Furthermore, the electronic device can change the test driving scenario based on the event text related to the road event under test, thereby increasing the types of driving scenarios tested by the intelligent driving system and improving the accuracy of the intelligent driving system test.

[0127] This application fully leverages the nonlinear modeling capabilities of deep learning models, employing CNN, Transformer, and GRU models to dynamically predict road driving scenarios over time. Combined with real, high-precision map data, it enhances the model's realism and effectiveness in generating road scene perception data. Furthermore, this application utilizes prior human descriptions of the road scene to be generated, using these descriptions as model input. This enables the controllable simulation and generation of any subsequent road scene data based on the current initialization scenario, allowing the simulation platform to more comprehensively and richly determine the capabilities of the assisted driving system and uncover potential defects.

[0128] exist Figure 2 Based on the embodiments shown, the following, in conjunction with Figure 6 The method for predicting multiple types of second sensor data in the above-mentioned test method for intelligent driving systems is explained in detail.

[0129] Figure 6This is a schematic diagram illustrating a method for predicting multiple types of second sensor data provided in an embodiment of this application. Please refer to... Figure 6 The method includes:

[0130] S601. Based on high-precision map data and multiple types of first sensor data, determine the scene characteristics corresponding to the test scene.

[0131] Specifically, electronic devices can determine the scene features corresponding to the test scenario based on the following feasible implementation methods: determine the map features corresponding to the high-precision map data and the sensor features corresponding to each type of first sensor data, and perform fusion processing on the map features and multiple sensor features to obtain the scene features.

[0132] Sensor features and map features can include temporal information. For example, in practical applications, high-precision map data and first sensor data can be data with a preset duration (e.g., 3 seconds, 5 seconds, etc.). Therefore, multiple first sensor data and map data also include temporal information. When extracting sensor features corresponding to multiple types of first sensor data, the temporal features of the sensor features can also be extracted.

[0133] Below, in conjunction with Figure 7 The process of acquiring scene features is explained.

[0134] Figure 7 This is a schematic diagram illustrating a process for acquiring scene features according to an embodiment of this application. Please refer to... Figure 7 The data includes: high-precision map data, video data, inertial sequence data, and LiDAR data. The aforementioned data constitutes the first sensor data. The high-precision map data, video data, inertial sequence data, and LiDAR data are input into a CNN network, respectively. The CNN network can output features corresponding to the high-precision map data, video data, inertial sequence data, and LiDAR data. Temporal features are extracted from the features corresponding to the aforementioned data (this can be implemented based on a Transformer temporal network, but this embodiment is not limited to this), and feature fusion processing is performed on the aforementioned features based on a Transformer encoder to obtain scene features. These scene features may include information from the high-precision map data, video information, inertial information, and LiDAR information; the scene features may also include temporal information.

[0135] Thus, based on Figure 7 The illustrated embodiments can acquire accurate map features, thereby improving the accuracy of predicting second sensor data. It should be noted that the above embodiments are merely illustrative of the type of first sensor data and do not limit the type of first sensor data.

[0136] S602. Based on the event text, determine the characteristics of the road event to be tested corresponding to the event text.

[0137] Optionally, the electronic device can process the event text based on a text encoder to obtain the features of the road event to be tested. For example, the electronic device can process the event text based on a BERT encoder, which can output the text features corresponding to the event text, and the electronic device can identify these text features as the features of the road event to be tested.

[0138] S603. Based on scene characteristics, road event characteristics to be measured, and multiple types of first sensor data, predict multiple types of second sensor data.

[0139] Specifically, electronic devices can predict multiple types of second sensor data according to the following feasible implementation: fusing scene features and road event features to be tested with the sensor features corresponding to each type of first sensor data to obtain fused features corresponding to each type of first sensor data; and predicting the second sensor data corresponding to any type of first sensor data based on the fused features corresponding to the first sensor data.

[0140] Specifically, the electronic device can fuse the sensor features corresponding to each type of first sensor data with scene features and road event features to be tested, thereby obtaining the fused features of each type of first sensor data, and predict the second sensor data of that type based on the fused features of each type.

[0141] Below, in conjunction with Figure 8 The process of predicting data from the second sensor is explained.

[0142] Figure 8 This is a schematic diagram illustrating a process for predicting second sensor data, provided as an embodiment of this application. Please refer to... Figure 8The dataset includes: high-precision map data, video data A, inertial sequence data B, and LiDAR data C. Video data A, inertial sequence data B, and LiDAR data C are the first sensor data. The high-precision map data, video data A, inertial sequence data B, and LiDAR data C are input into a CNN network, respectively. The CNN network can output features corresponding to the high-precision map data, video data A, inertial sequence data B, and LiDAR data C. Temporal features are extracted from the features corresponding to the above data (this can be implemented based on a Transformer temporal network, but this embodiment does not limit this), resulting in map features corresponding to the high-precision map data, video features corresponding to video data A, inertial features corresponding to inertial sequence data B, and LiDAR features corresponding to LiDAR data C. Feature fusion processing is performed on the map features, video features, inertial features, and LiDAR features based on a Transformer encoder to obtain scene features.

[0143] Please see Figure 8 The GRU network is fed with the road event features, scene features, and video features corresponding to the event text. The GRU network outputs video data (a). Similarly, it outputs inertial sequence data (b) and LiDAR features (c). Video data (a), inertial sequence data (b), and LiDAR data (c) can all be secondary sensor data. This allows electronic devices to predict sensor data for any driving scenario, thereby increasing the types of driving scenarios tested by intelligent driving systems and improving the accuracy of intelligent driving system testing.

[0144] It should be noted that the above embodiments are merely illustrative examples illustrating the types of sensor data and do not limit the types of sensor data. Since the inputs of the first sensor data and high-precision map data can be data of a preset duration, the second sensor data output by the GRU network is data of the next preset duration. For example, if the high-precision map data, video data A, inertial sequence data B, and lidar data C are data within the 0-T time period, then the video data a, inertial sequence data b, and lidar data c can be data within the T-2T time period.

[0145] This application provides a method for predicting multiple types of second sensor data. Based on high-precision map data and multiple types of first sensor data, the method determines the scene features corresponding to the test scenario. Based on event text, it determines the road event features to be tested corresponding to the event text. Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, it predicts multiple types of second sensor data. In this way, electronic devices can predict sensor data corresponding to any driving scenario, thereby increasing the types of driving scenarios tested by intelligent driving systems and improving the accuracy of intelligent driving system testing.

[0146] Based on any of the above embodiments, the following, in conjunction with Figure 9 The training process of the above driving scenario generation model will be explained.

[0147] Figure 9 This is a schematic diagram illustrating the training process of a driving scene generation model provided in an embodiment of this application. Please refer to [link / reference]. Figure 9 The dataset includes high-precision map data, video data, inertial sequence data, and LiDAR data. The high-precision map data, video data, inertial sequence data, and LiDAR data are the actual first-sensor data within the time period from time T to time T1. The high-precision map data (T-T1), video data (T-T1), inertial sequence data (T-T1), and LiDAR data (T-T1) are respectively input into the CNN network. The CNN network can output the features corresponding to the high-precision map data (T-T1), the video data (T-T1), the inertial sequence data (T-T1), and the LiDAR data (T-T1).

[0148] Please see Figure 9 Temporal features are extracted from the features corresponding to the aforementioned data (this can be implemented based on a Transformer temporal network, which is not limited in this embodiment), resulting in map features corresponding to high-precision map data (T-T1), video features corresponding to video data (T-T1), inertial features corresponding to inertial sequence data (T-T1), and lidar features corresponding to lidar data (T-T1). Feature fusion processing is performed on the map features, video features, inertial features, and lidar features based on a Transformer encoder to obtain scene features. Text describing the test scene within the T-T1 time period is obtained, corresponding to the high-precision map data, video data, inertial sequence data, and lidar data within the aforementioned T-T1 time period. The text features corresponding to the text of the test scene are determined, and a first loss between the text features and the scene features is determined.

[0149] Please see Figure 9The system acquires event texts of actual road events occurring within the time period T1-T2 and determines the corresponding road event features. The road event features, scene features, and video features corresponding to the event texts are input into a GRU network, which outputs video data (T1-T2). Similarly, the road event features, scene features, and inertial features corresponding to the event texts are input into the GRU network, which outputs inertial sequence data (T1-T2). Finally, the road event features, scene features, and LiDAR features corresponding to the event texts are input into the GRU network, which outputs LiDAR data (T1-T2). The video data (T1-T2), inertial sequence data (T1-T2), and LiDAR data (T1-T2) can be second sensor data within the predicted time period T-T1.

[0150] Please see Figure 9 The system acquires real video data, real inertial sequence data, and real lidar data for the T1-T2 time period. Based on these data, and in conjunction with the video data (T1-T2), inertial sequence data (T1-T2), and lidar data (T1-T2), it determines the second loss. (Electronic equipment) Figure 9 (Not shown in the image) A loss function can be constructed based on the first loss and the second loss, and the driving scene generation model can be trained based on this loss function. In this way, the electronic device can predict the data collected by the vehicle's sensors in any driving scene based on the driving scene generation model. Therefore, the electronic device can test the intelligent driving system according to any driving scene, such as extreme driving scenarios that are difficult to achieve in offline testing. This can increase the diversity of driving scenarios for intelligent driving system testing and improve the accuracy of intelligent driving system testing.

[0151] It should be noted that the training of the controllable driving scene generation model is specifically divided into four parts according to the network structure: a multimodal data feature extraction network (CNN), a temporal coding network (transformer-temporal), a multimodal fusion coding network (transformer-encoder), and a cyclic road scene data decoding network (GRU-decoder). According to feature learning, it is divided into two stages: the current road scene learning and description stage and the temporal road scene prediction stage. The first learning phase involves inputting 5 seconds of road scene data from various modalities into the generative model. Each modal data is processed by its corresponding feature extraction network (CNN), such as map data feature extraction, IMU feature extraction, visual image feature extraction, and laser point cloud feature extraction. The multimodal data features from each time series are then extracted using a transformer, and a transformer-encoder network is used for multimodal feature fusion learning. Based on the fused road features from the time series, a current road scene feature descriptor is output. This descriptor is compared with the ground truth of manually labeled text descriptions to calculate the loss, which is used to train the model's scene description capability. The second learning phase utilizes the temporal feature data from each modality, fusing subsequent temporal text description data and high-precision map data. A GRU recurrent neural network is then used to predict subsequent road perception data. The predicted data is compared with the ground truth of each modal perception data to calculate the loss, which is used to train the model's scene prediction capability. During training, a MAE strategy is employed, randomly erasing some perception data to improve the model's ability to generate high-quality road scene data even when some sensor signals from the vehicle are lost.

[0152] The driving scene generation model (CNN+Transformer model) in this application can use static visual data and some multimodal perception data as initial states, combined with high-precision map data corresponding to the current scene and preset text information, to controllably generate dynamic scene driving data through a recurrent network GRU model.

[0153] The intelligent assisted driving system automatic test simulation platform proposed in this application can automatically and in multiple scenarios conduct quantitative tests on intelligent assisted driving systems through deep generative models, providing a reliable, objective, and quantifiable test tool for the development of intelligent assisted driving technology, facilitating the discovery of defects and potential problems, and promoting the development of the industry.

[0154] Figure 10 This is a schematic diagram of a testing device for an intelligent driving system provided in an embodiment of this application. Please refer to [link / reference]. Figure 10 The testing device 101 for the intelligent driving system includes a first acquisition module 101, a second acquisition module 102, a prediction module 103, and a processing module 104, wherein:

[0155] The first acquisition module 101 is used to acquire high-precision map data of the test scene and multiple types of first sensor data, wherein the multiple types of first sensor data are data collected by various types of sensors in the vehicle in the test scene;

[0156] The second acquisition module 102 is used to acquire event text related to the road event to be tested;

[0157] The prediction module 103 is used to predict, based on high-precision map data, the multiple types of first sensor data, and the event text, the multiple types of second sensor data collected by the multiple types of sensors in the vehicle when the road event to be tested occurs in the test scenario.

[0158] The processing module 104 is used to process the data from the multiple types of second sensors according to the intelligent driving system to obtain the driving instructions output by the intelligent driving system, and to test the intelligent driving system according to the driving instructions.

[0159] In one possible implementation, the prediction module 103 is specifically used for:

[0160] Based on the high-precision map data and the multi-type first sensor data, the scene characteristics corresponding to the test scene are determined;

[0161] Based on the event text, determine the characteristics of the road event to be tested corresponding to the event text;

[0162] Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, predict the multiple types of second sensor data.

[0163] In one possible implementation, the prediction module 103 is specifically used for:

[0164] Determine the map features corresponding to the high-precision map data, and the sensor features corresponding to each type of the first sensor data, wherein the sensor features and the map features include time-series information;

[0165] The scene features are obtained by fusing the map features and multiple sensor features.

[0166] In one possible implementation, the prediction module 103 is specifically used for:

[0167] The scene features and the road event features to be tested are fused with the sensor features corresponding to each type of the first sensor data to obtain the fused features corresponding to each type of the first sensor data.

[0168] For any type of first sensor data, predict the second sensor data corresponding to the first sensor data based on the fusion features corresponding to the first sensor data.

[0169] In one possible implementation, the processing module 104 is specifically used for:

[0170] The various types of second sensor data are input to multiple sensors connected to the intelligent driving system;

[0171] The system acquires driving instructions generated by the intelligent driving system, which are determined by the intelligent driving system based on multiple types of second sensor data fed back from multiple connected sensors.

[0172] In one possible implementation, the processing module 104 is specifically used for:

[0173] Determine the sensor type for each of the aforementioned sensors;

[0174] According to the sensor type, input the corresponding second sensor data to each sensor.

[0175] In one possible implementation, the processing module 104 is specifically used for:

[0176] Acquire at least one set of target driving behaviors when the road event to be tested occurs in the test scenario;

[0177] If the driving behavior indicated by the driving command matches the at least one set of target driving behaviors, then the intelligent driving system passes the test when the test road event occurs in the test scenario.

[0178] The testing device for the intelligent driving system provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0179] The testing device for the intelligent driving system shown in this application embodiment can be a chip, hardware module, processor, etc. Of course, the testing device for the intelligent driving system can be in other forms, and this application embodiment does not specifically limit it.

[0180] Figure 11 A schematic diagram of the hardware structure of the electronic device provided in this application. Please refer to [link / reference]. Figure 11The electronic device 110 may include a processor 111 and a memory 112, wherein the processor 111 and the memory 112 can communicate; for example, the processor 111 and the memory 112 communicate via a communication bus 113, the memory 112 is used to store program instructions, and the processor 111 is used to call the program instructions in the memory to execute the test method of the intelligent driving system shown in any of the above method embodiments.

[0181] Optionally, the electronic device 110 may also include a communication interface, which may include a transmitter and / or a receiver.

[0182] Optionally, the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0183] This application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, include, as in the first aspect, and various possible testing methods for intelligent driving systems involved in the first aspect.

[0184] This application embodiment may also provide a computer program product, the computer program product including a computer program, which, when executed by a processor, performs test methods for intelligent driving systems as described in the first aspect and various possible aspects of the first aspect.

[0185] All or part of the steps in the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable memory. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned memory (storage medium) includes: read-only memory (ROM), RAM, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof.

[0186] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processing unit of a general-purpose computer, special-purpose computer, embedded processor, or other programmable electronic device to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable electronic device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0187] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable electronic device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0188] These computer program instructions may also be loaded onto a computer or other programmable electronic device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the functions specified in one or more boxes. Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of this application. Therefore, if these modifications and variations of the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations. In this application, the term "comprising" and its variations can refer to a non-limiting inclusion; the term "or" and its variations can refer to "and / or". The terms "first," "second," etc., in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. In this application, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

Claims

1. A testing method for an intelligent driving system, characterized in that, include: Acquire high-precision map data and multiple types of first sensor data for the test scenario. The multiple types of first sensor data are data collected by various types of sensors in the vehicle in the test scenario. Obtain the event text related to the road event to be tested; Based on the high-precision map data, the multiple types of first sensor data, and the event text, predict the multiple types of second sensor data collected by the various types of sensors in the vehicle when the road event to be tested occurs in the test scenario. The intelligent driving system processes the data from the various second sensors to obtain driving commands output by the intelligent driving system, and then tests the intelligent driving system based on the driving commands.

2. The method according to claim 1, characterized in that, Based on the high-precision map data, the multiple types of first sensor data, and the event text, predict the multiple types of second sensor data collected by various types of sensors in the vehicle when the test road event occurs in the test scenario, including: Based on the high-precision map data and the multi-type first sensor data, the scene characteristics corresponding to the test scene are determined; Based on the event text, determine the characteristics of the road event to be tested corresponding to the event text; Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, predict the multiple types of second sensor data.

3. The method according to claim 2, characterized in that, Based on the high-precision map data and the multiple types of first sensor data, the scene features corresponding to the test scene are determined, including: Determine the map features corresponding to the high-precision map data, and the sensor features corresponding to each type of the first sensor data, wherein the sensor features and the map features include time-series information; The scene features are obtained by fusing the map features and multiple sensor features.

4. The method according to claim 2 or 3, characterized in that, Based on the scene features, the road event features to be tested, and the multiple types of first sensor data, predict the multiple types of second sensor data, including: The scene features and the road event features to be tested are fused with the sensor features corresponding to each type of the first sensor data to obtain the fused features corresponding to each type of the first sensor data. For any type of first sensor data, predict the second sensor data corresponding to the first sensor data based on the fusion features corresponding to the first sensor data.

5. The method according to any one of claims 1-3, characterized in that, The intelligent driving system processes the data from the various second sensors to obtain driving commands output by the intelligent driving system, including: The various types of second sensor data are input to multiple sensors connected to the intelligent driving system; The system acquires driving instructions generated by the intelligent driving system, which are determined by the intelligent driving system based on multiple types of second sensor data fed back from multiple connected sensors.

6. The method according to claim 5, characterized in that, Inputting the various types of second sensor data to multiple sensors connected to the intelligent driving system, including: Determine the sensor type for each of the aforementioned sensors; According to the sensor type, input the corresponding second sensor data to each sensor.

7. The method according to any one of claims 1-3, characterized in that, The intelligent driving system is tested according to the driving command, including: Acquire at least one set of target driving behaviors when the road event to be tested occurs in the test scenario; If the driving behavior indicated by the driving command matches the at least one set of target driving behaviors, then the intelligent driving system passes the test when the test road event occurs in the test scenario.

8. A testing device for an intelligent driving system, characterized in that, It includes a first acquisition module, a second acquisition module, a prediction module, and a processing module, wherein: The first acquisition module is used to acquire high-precision map data of the test scene and multiple types of first sensor data, wherein the multiple types of first sensor data are data collected by various types of sensors in the vehicle in the test scene; The second acquisition module is used to acquire event text related to the road event to be tested; The prediction module is used to predict, based on high-precision map data, the multiple types of first sensor data, and the event text, the multiple types of second sensor data collected by the various types of sensors in the vehicle when the road event to be tested occurs in the test scenario. The processing module is used to process the data from the multiple types of second sensors according to the intelligent driving system to obtain the driving instructions output by the intelligent driving system, and to test the intelligent driving system according to the driving instructions.

9. An electronic device, characterized in that, include: Processor, memory; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the test method for the intelligent driving system as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the test method for the intelligent driving system as described in any one of claims 1-7.

Images