Autonomous vehicle key signal test method and system based on real data driving

Abstract

The invention discloses an automatic driving vehicle key signal testing method and system based on real data driving. According to the method, data are collected in a real road scene, test cases are automatically generated, a standard test case library is constructed, hierarchical or parallel testing is carried out in virtual simulation, hardware-in-the-loop and real vehicle environments, and result feedback and closed-loop iteration are achieved through automatic analysis. The system comprises a data collection module, a test case generation module, a standard test case library module, a test execution module and a result feedback module. According to the invention, the problems of insufficient scene coverage and low authenticity and efficiency of the existing test method are solved, the comprehensiveness, authenticity and test efficiency of the test scene are improved, and the safety and reliability of the key signal system of the automatic driving vehicle are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and particularly relates to a method and system for testing key signals of an autonomous driving vehicle driven by real data, which are used for comprehensively, efficiently and accurately testing key signal systems such as steering, braking, and body control of an autonomous driving vehicle. Background Art

[0002] In the field of testing key signal systems of autonomous driving vehicles, the existing technologies mainly adopt a single or partial combination of traditional on-vehicle testing, virtual simulation testing, and hardware-in-the-loop (HIL) testing, but there are many defects: 1. Insufficient coverage and efficiency of traditional testing methods: Limited scenario coverage: On-vehicle testing relies on manually designed scenarios and is difficult to cover long-tail scenarios such as extreme weather and high-risk accidents, such as "ghost vehicles" and emergency braking. The closed-site testing has a low degree of restoration of complex environments such as dynamic traffic flow and multi-vehicle interaction, resulting in the consistency between the test results and the real road being less than 95%. This is because manually designed scenarios cannot comprehensively consider various complex situations on the actual road and are difficult to simulate the real driving environment.

[0003] Cost and cycle issues: On-vehicle testing consumes a large amount of time and funds. It takes several years to verify a mileage of millions of kilometers, and high-risk scenarios cannot be tested due to regulatory restrictions. This makes the cost of on-vehicle testing high and ineffective in testing some key high-risk scenarios, affecting the comprehensiveness of the testing.

[0004] 2. Bottlenecks in hardware-in-the-loop (HIL) technology: Difficulties in co-verifying multiple systems: In the dynamic performance verification of key signal systems such as steering and braking by existing HIL platforms, it is difficult to accurately simulate the dynamic characteristics of real vehicles, such as steering response delay and braking pressure fluctuation, resulting in deviations between the test results of controllers (such as ESP and EPS) and the actual working conditions. This is due to the technical limitations of HIL platforms in simulating the dynamic characteristics of real vehicles and their inability to fully reproduce various dynamic characteristics of real vehicles.

[0005] Insufficient fault injection and redundancy capabilities: Traditional HIL platforms lack flexible multi-level fault simulation capabilities, such as sensor failures and communication interruptions, and cannot fully verify the safety redundancy mechanism of the system. This makes it impossible to comprehensively evaluate the performance of the system under various fault conditions during the testing process, posing potential safety hazards.

[0006] 3. Low level of intelligence in scenario generation and test cases Singleization of the scenario library: The existing scenario generation relies on manual rule design and cannot efficiently generalize real road test data. Rule-based modeling requires manual parameter adjustment, has poor scalability, and is difficult to support the parameterized combination of thousands of scenarios. This results in the scenario library being unable to adapt to the diversity and complexity of actual roads, and the test coverage is limited.

[0007] Low efficiency of test case generation: Traditional methods lack the ability of adaptive optimization and cannot quickly screen key test scenarios, resulting in a test efficiency that is only 1 / 1000 of the actual requirements. The method of manually designing test cases cannot meet the requirements of fast and efficient testing, and it is difficult to ensure the comprehensiveness and representativeness of test cases. Summary of the Invention

[0008] The present invention aims to solve the problems of insufficient test scenario coverage, insufficient authenticity and representativeness of test cases, long test feedback cycle, and low efficiency in the existing test methods for the key signal system of autonomous driving vehicles, and provides a test method and system for the key signals of autonomous driving vehicles based on real data driving, so as to improve the comprehensiveness, authenticity of test scenarios and test efficiency, and ensure the safety and reliability of the key signal system of autonomous driving vehicles.

[0009] To solve the above technical problems, the present invention is implemented by the following solutions: A test method for the key signals of autonomous driving vehicles based on real data driving, comprising the following steps: Step S1: Data collection: Use a test vehicle equipped with sensors to collect key signal data and environmental information during vehicle driving in a real road scenario, and store them to form a real vehicle database.

[0010] Step S2: Automatic generation of test cases: Extract and screen out typical and representative road test scenario data from the real vehicle database, and then generate a test case set covering normal working conditions, extreme working conditions and edge scenarios through generalization processing of road test scenarios.

[0011] Step S3: Construction of a standard test case library: Establish a set of standard test case libraries including a standard test case library for the steering system, a standard test case library for the throttle / brake system, and a standard test case library for the key signals of the body control system.

[0012] Step S4: Test execution: Send the automatically generated test cases and the cases in the standard test case library to the test vehicle, and the controller in the test vehicle executes corresponding actions according to the scenario requirements in the cases, and records the vehicle state changes and response data.

[0013] Step S5: Result feedback and closed-loop iteration: Analyze the response data of the test vehicle, associate the analysis results with the test scenarios / cases, and feedback them to the test case generation module to optimize the test cases or generate new test scenarios for the next test iteration, thus realizing a closed-loop test process.

[0014] For further optimization, in step S1, by installing a variety of sensors on the test vehicle, such as cameras, radars, IMUs (Inertial Measurement Units), GPS, etc., it is ensured that key signal data and environmental information during vehicle driving can be collected comprehensively. The camera is used to capture road scenes, traffic signs, and obstacles around the vehicle; the radar measures the distance and relative speed between the vehicle and surrounding objects; the IMU obtains attitude information such as the vehicle's acceleration and angular velocity; the GPS locates the vehicle's position and driving trajectory.

[0015] Conduct multi-scenario road tests under different road types (such as urban roads, highways, rural roads, etc.), weather conditions (such as sunny days, rainy days, snowy days, etc.), traffic conditions (such as peak hours, off-peak hours, congestion, etc.), and driving conditions (such as acceleration, deceleration, turning, following, etc.). For example, test the signal data of the vehicle under frequent starts and stops and complex traffic environments during peak hours on urban roads, and test the vehicle state data during high-speed driving on highways.

[0016] The key signal data includes the vehicle's steering angle, angular velocity, throttle opening, acceleration, braking pressure, vehicle speed, and body attitude, and the environmental information includes road markings, traffic signs, and information about obstacles around the vehicle.

[0017] Perform real-time processing and storage of the collected data, and store it in a unified database according to certain formats and specifications to provide a data basis for subsequent analysis and processing.

[0018] For further optimization, in step S2, use the previously collected scenario data to train a model for scenario extraction, and then input all the data in the real vehicle database into the trained model to extract the road test scenarios. The generalization of road test scenarios adopts a scenario generation model based on deep learning to expand diverse scenarios. It specifically includes the following steps: Step S2.1: Road test scenario extraction: Step S2.1.1: Clean the original data stored in the database, remove noise, outliers, and duplicate data to improve data quality. For example, use a filtering algorithm to remove noise data caused by sensor measurement errors, and identify and delete obviously incorrect or unreasonable data records through data verification rules.

[0019] Step S2.1.2: Then, use machine learning algorithms such as clustering analysis and association rule mining to analyze the cleaned data and extract typical and representative scene segments. For example, through clustering analysis, scene patterns such as frequent high-speed turning, emergency braking, and following a vehicle are found; using association rule mining to find key signal data combinations related to specific driving behaviors or environmental conditions.

[0020] Step S2.1.3: Label and classify the extracted scene segments. The labeling content includes scene types (such as turning scenes, braking scenes, etc.), key feature parameters (such as steering angle, vehicle speed, braking pressure, etc.), and environmental information (road type, weather conditions, etc.). Classification can be carried out according to multiple dimensions such as driving behavior, road conditions, and environmental factors for subsequent screening and processing according to different requirements.

[0021] Step S2.2: Road test scene generalization: Step S2.2.1: Expand the extracted typical scenes according to traffic rules, driving common sense, and vehicle dynamics principles. For example, for an emergency braking scene on a dry road surface on a sunny day, a series of similar but different working condition test scenes can be generated by changing parameters such as road surface friction coefficient (simulating a wet and slippery road surface), vehicle initial speed, and braking start distance.

[0022] Step S2.2.2: Adopt deep learning models such as Generative Adversarial Network (GAN) and Variational Autoencoder (VAE) to learn the original scene data and generate new scene data. These models can generate diverse scene variants while retaining the original scene features. For example, use GAN to generate the same driving scene under different weather conditions, or use VAE to generate scenes with different combinations of vehicle speed and acceleration.

[0023] Step S2.2.3: Generation of marginal scenes and extreme working conditions: Pay special attention to the generation of marginal scenes and extreme working conditions, such as the "ghost probe" scene and the braking scene in the case of a vehicle tire blowout. Through the analysis and simulation of historical accident data and the study of vehicle extreme performance, test instances of these high-risk scenes are generated to comprehensively evaluate the response ability of the key signal system of autonomous vehicles in special situations.

[0024] Step S2.3: Test case generation. Combine the generalized scenario data with the corresponding test objectives and expected results to generate complete test cases. For example, for a test case of a steering scenario, clearly set the initial state of the vehicle (speed, position, etc.), steering operations (steering angle, steering speed), and expected vehicle responses (such as steering angle deviation range, body roll angle limit, etc.). Develop detailed test steps for each test case, including pre-test preparations (such as vehicle inspection, equipment calibration), operation sequence and parameter settings during the test, and data analysis requirements after the test. Ensure that testers or automated test systems can execute the tests according to a unified standard.

[0025] Write test cases in a unified format and template, including fields such as test case number, name, description, test steps, expected results, test conditions, etc., for easy management, maintenance, and execution. For example, the test case numbers are arranged according to certain rules for easy searching and reference; the test description clearly elaborates the purpose and scenario background of the test.

[0026] For further optimization, in step S3, the construction of the standard test case library specifically includes: Step S3.1: Define the test scope and objectives based on industry standards to clarify the coverage of the test case library. It includes: steering systems, such as the accuracy and responsiveness of EPS (Electric Power Steering); acceleration and deceleration systems, such as throttle pedal response, brake pedal response; body control systems, such as ESP (Electronic Stability Program), ESC (Electronic Stability Control), EPB (Electronic Parking Brake).

[0027] Step S3.2: Collect and analyze industry standards: Refer to the scenario definition methods specified in existing automotive industry standards or national standards to determine the standard framework and indicators of the test case library. Specifically, reference can be made to GB / T34590-2017 "General Technical Conditions for Automated Driving Systems of Road Vehicles", which defines the general technical specifications for automated driving vehicles, including the basic test requirements for vehicle dynamic behavior, and ISO21448 "Road Vehicles - Safety of the Intended Functionality" (SOTIF standard), which provides detailed guidance on the construction of safety scenarios and boundary condition tests for automated driving systems.

[0028] Step S3.3: Collect road test scenario data, specifically: Collect real road data, including normal scenarios, boundary scenarios, accident or danger scenarios, and extract scenario data from existing standard documents, research results, and vehicle accident case databases; Step S3.4: Parameterization and generalization processing of scenario feature parameters, specifically: Extract key signal parameters, such as vehicle speed, steering angle, acceleration, braking force, road surface conditions, etc., And grading, classifying, and labeling the characteristic parameters to facilitate algorithm generalization processing and standardized management.

[0029] Step S3.5: Generation of standardized test cases, including: Step S3.5.1: Data preprocessing: Filtering and noise reduction, and removing abnormal data from the collected road scene data; Extract the sequence of key vehicle characteristic parameters, such as lateral acceleration, longitudinal speed, steering angle, braking pressure, etc.

[0030] Step S3.5.2: Key feature clustering: Use clustering algorithms (such as K-means or Gaussian mixture clustering GMM) to cluster the road scene characteristic parameters, and identify typical representative scenarios and boundary scenarios; The purpose of clustering is to clarify the distribution ranges of the central features of typical scenarios and marginal operating conditions.

[0031] Step S3.5.3: Feature generalization algorithm: Use the generative adversarial network GAN or variational autoencoder VAE algorithm to generalize the data of typical scenarios, and generate more similar scenarios with varying parameters; Algorithm input: The key parameter vector of the scenario, such as X = [speed, steering angle, acceleration, braking force, road surface friction coefficient], X = [speed, steering angle, acceleration, braking force, road surface friction coefficient]; Algorithm output: A set of generalized scenario vectors to enhance the coverage of the test scenario library.

[0032] Step S3.5.4: Test case screening algorithm: Evaluate the test importance of the generalized scenarios through an importance-based automatic screening algorithm (such as genetic algorithm, particle swarm optimization PSO, etc.); Use the automatic screening algorithm for iterative optimization, and select the most critical and representative scenarios to add to the standard use case library; Step S3.5.5: Standard formatting output, specifically: Output of the standardized test scenario document, including: scenario number, scenario type, target test parameters, expected output results; And the test execution process and specifications, including the scenario entry method, exit method, and action execution timing.

[0033] In this application, the so-called "standardization" not only means compliance with enterprise or internal requirements, but more importantly, compliance with national standards or international industry standards. The method for constructing the standard test case library proposed in this application clearly takes authoritative national and international standards as the basis, and uses artificial intelligence and clustering generalization algorithms to automatically generate scenarios, achieving high coverage, authenticity, and consistency of test cases, and meeting the requirements of the automotive industry for the functional and safety testing of autonomous driving systems.

[0034] For further optimization, in step S4, the test execution stage includes hierarchical or parallel testing of the same test case and standard test case in a virtual simulation platform, a hardware-in-the-loop (HIL) platform, and a real vehicle environment, and seamless switching or parallel operation of the test case on different platforms is achieved through a scheduling module and an interface standardization protocol.

[0035] For further optimization, in step S5, in result feedback and closed-loop iteration, an artificial intelligence automated analysis module is used to compare and statistically analyze the response data of the test vehicle, and extract abnormal signal responses, potential defects, and control failure information in edge scenarios. Specifically, it includes Step S5.1: Simulation test verification: Conduct simulation tests on the generated test cases and standard test cases on a virtual simulation platform to check the feasibility and effectiveness of the test cases. By simulating the operation of the vehicle in different scenarios, verify whether the test cases can accurately detect the performance problems and potential faults of the system. For example, observe whether the response of the steering system meets the expectations and whether there are abnormal situations when the simulated vehicle executes a certain steering test case.

[0036] Step S5.2: Actual vehicle test verification: Select some representative test cases and standard test cases to conduct test verification on an actual vehicle, and further verify the authenticity and reliability of the test cases. Compare the actual test results with the expected results, analyze the reasons for the differences, and optimize and adjust the test cases. For example, if it is found in actual tests that the braking distance of the vehicle exceeds the expectation under a certain braking test case, it is necessary to check whether the parameter settings in the test case are reasonable or whether some influencing factors are omitted.

[0037] Industry experts review the test cases and provide opinions and suggestions based on their experience and professional knowledge. Experts can evaluate from aspects such as the rationality of the test scenario, the clarity of the test objective, and the accuracy of the expected results, providing a reference for the optimization of the test cases. According to the expert feedback, further improve the test cases to ensure their quality and effectiveness.

[0038] A key signal test system for autonomous vehicles driven by real data includes: Data collection module: Used to collect key signal data and environmental information during vehicle driving in real road scenarios and store them; Test case generation module: Includes a road test scenario extraction unit and a road test scenario generalization unit, used to extract real vehicle data from a unified database to generate a test case set covering normal working conditions, extreme working conditions, and edge scenarios; Standard test case library module: Establish a standard test case library set including a standard test case library for the steering system, a standard test case library for the throttle / brake system, and a standard test case library for key signals of the body control system; Test Execution Module: Transfers test cases to the test vehicle, controls the controller in the test vehicle to perform corresponding actions according to the scenario requirements in the cases, and records the vehicle state changes and response data. The test execution module includes a virtual simulation platform, a Hardware-in-the-Loop (HIL) platform, and a real vehicle test unit, and realizes hierarchical or parallel testing of test cases on different platforms through a scheduling module and an interface standardization protocol; Result Feedback Module: Includes an artificial intelligence automated analysis unit, which is used to analyze the response data of the test vehicle, associate the analysis results with the test scenarios / cases, and feedback them to the test case generation module to realize a closed-loop test process.

[0039] For further optimization, the sensors carried by the data collection module include cameras, radars, IMUs, and GPS.

[0040] For further optimization, the road test scenario extraction unit of the test case generation module uses machine learning algorithms, and the road test scenario generalization unit uses a scenario generation model based on deep learning.

[0041] For further optimization, the artificial intelligence automated analysis unit of the result feedback module is used to extract abnormal signal responses, potential defects, and control failure information in marginal scenarios.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention automatically generates and generalizes test scenarios driven by real road data, significantly improving the test coverage, covering extreme working conditions and marginal scenarios that are difficult to cover by traditional test methods, and making the test results closer to actual driving conditions.

[0043] 2. The present invention establishes a standard test case library to ensure the standardization, consistency, and repeatability of the test process, reduces test differences caused by human factors, and facilitates horizontal comparison and verification between different vehicles.

[0044] 3. The present invention reduces the time cost of manual analysis by setting up an automated feedback mechanism, realizes rapid closed-loop iterative optimization, timely discovers and solves problems existing in the test, and improves the test efficiency and the quality of test cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flowchart of the key signal test method for autonomous driving vehicles based on real data driving according to the present invention.

[0046] Figure 2 It is a block diagram of the key signal test system for autonomous driving vehicles based on real data driving according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] The present invention will be further described below in conjunction with embodiments, but it should not be understood that the above-mentioned subject scope of the present invention is limited to the following embodiments. Embodiment 1:

[0048] As Figure 1 shown, a method for testing key signals of an autonomous driving vehicle based on real data driving includes the following steps: Step S1: Data collection: Use a test vehicle equipped with sensors such as cameras, radars, IMUs, and GPSs to conduct road tests in real road scenarios. Under different weather conditions (sunny, rainy, snowy, etc.), different traffic conditions (peak hours, off-peak hours, congested sections, etc.), and different road types (highways, urban roads, rural roads, etc.), key signal data during vehicle driving is collected in real time, including steering angle, angular velocity, throttle opening, acceleration, braking pressure, vehicle speed, body attitude, and environmental information (road markings, traffic signs, information on obstacles around the vehicle). The collected data is stored in a dedicated database after being processed. For example, timestamps are used to mark the data for subsequent query and analysis.

[0049] Step S2: Automatic test case generation stage: Extract "real vehicle data" from the database for "road test scenario extraction". Use machine learning algorithms, such as clustering analysis algorithms, to analyze the collected data and identify typical scenario segments such as high-speed turning, emergency braking, and following driving. Then, through "road test scenario generalization" processing, a scene generation model based on deep learning, such as a generative adversarial network GAN, is used to expand the extracted typical scenarios into diverse scenarios. For example, under wet road surface conditions, according to parameters such as road surface friction coefficient, the emergency braking scenario is generalized to generate test scenarios with different vehicle speeds and different braking start distances. Finally, the "test case generation" module combines the generalized scenario data to generate executable standardized test cases, including detailed descriptions of the test scenarios, expected vehicle responses, and other information.

[0050] Step S3: Standard test case library construction stage: In addition to the automatically generated test cases, construct a set of standardized test case libraries. In the test case library for the steering system, design test cases for different steering angles, steering speeds, road surface conditions, etc.; the test case library for the throttle / brake system includes test scenarios under different throttle openings, braking pressures, vehicle loads, etc.; the test case library for the body control system and other key signals covers test cases for body functions such as headlights, windshield wipers, and windows under different working conditions. These standard cases are written in a unified format and specification for easy management and use.

[0051] Step S4: Test Execution Phase: The automatically generated test cases and standardized test cases are sent to the test vehicle through the standard interface. The test vehicle is equipped with a steering controller, an acceleration controller, a braking controller, and other relevant vehicle body controllers. During the test, each controller performs corresponding actions according to the requirements of the test cases. For example, in a test case simulating a high-speed turn, the steering controller controls the vehicle's steering according to the preset steering angle and speed, and the acceleration controller and the braking controller adjust the vehicle speed according to the test scenario. At the same time, the vehicle records the state changes and response data such as the steering angle, vehicle speed, and acceleration in real time.

[0052] Step S5: Result Feedback and Closed-loop Iteration Phase: After the test vehicle completes the test cases, the real-time data output by each vehicle controller is automatically transmitted to the test result analysis system. The artificial intelligence automated analysis module is used to analyze the data. For example, by comparing the expected response and the actual response, abnormal signal responses and potential defects are identified. If it is found that the vehicle has oversteering in a specific steering test case, the analysis system associates the problem with the corresponding test scenario and test case and feedbacks it to the test case generation module. The test case generation module optimizes the existing test cases according to the feedback results, such as adjusting the parameters of the steering angle or speed, or generating new test scenarios for the next test iteration, so as to achieve continuous optimization and improvement. Embodiment 2:

[0053] As Figure 2 shown, a key signal test system for autonomous vehicles driven by real data includes: Data Collection Module: It is used to collect key signal data and environmental information during the vehicle's driving process in real road scenarios and store them; the sensors carried by the data collection module include cameras, radars, IMUs, and GPSs.

[0054] Test Case Generation Module: It includes a road test scenario extraction unit and a road test scenario generalization unit, which are used to extract real vehicle data from a unified database to generate a test case set covering normal working conditions, extreme working conditions, and edge scenarios; the road test scenario extraction unit uses machine learning algorithms, and the road test scenario generalization unit uses a scenario generation model based on deep learning.

[0055] Standard Test Case Library Module: A standard test case library set including a standard test case library for the steering system, a standard test case library for the throttle / brake system, and a standard test case library for key signals of the vehicle body control system is established; Test execution module: Transmits test cases to the test vehicle, controls the controller in the test vehicle to perform corresponding actions according to the scenario requirements in the cases, and records the vehicle state changes and response data. The test execution module includes a virtual simulation platform, a hardware-in-the-loop (HIL) platform, and a real vehicle test unit, and realizes hierarchical or parallel testing of test cases on different platforms through a scheduling module and an interface standardization protocol; Result feedback module: Includes an artificial intelligence automated analysis unit, which is used to analyze the response data of the test vehicle, associate the analysis results with the test scenario / case, and feedback to the test case generation module to realize a closed-loop test process. The artificial intelligence automated analysis unit is used to extract abnormal signal responses, potential defects, and control failure information under edge scenarios.

[0056] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A method for testing key signals of an autonomous driving vehicle based on real data, characterized in that: The following steps are involved: Step S1: Data collection: using a test vehicle equipped with sensors to collect key signal data and environmental information during vehicle driving in a real road scene, and store them to form a real vehicle database; Step S2: Automatic generation of test cases: extract and screen typical and representative road test scenario data from the real vehicle database, and then generate a test case set covering normal working conditions, extreme working conditions and edge scenarios through generalization of the road test scenarios; Step S3: standard test case library construction: establishing a standard test case library set including a steering system standard test case library, a throttle / brake system standard test case library, and a body control system key signal standard test case library; Step S4: Test execution: The automatically generated test cases and the test cases in the standard test case library are sent to the test vehicle. The controller in the test vehicle performs corresponding actions according to the scenario requirements in the test case and records the vehicle state changes and response data; Step S5: Result feedback and closed-loop iteration: Analyze the response data of the test vehicle, associate the analysis results with the test scenarios / use cases, and feed them back to the use case generation module to optimize the test cases or generate new test scenarios for the next test iteration to achieve a closed-loop test process.

2. The real data-driven automatic driving vehicle key signal testing method according to claim 1 is characterized in that: In step S1, the key signal data include vehicle steering angle, angular velocity, throttle opening, acceleration, brake pressure, vehicle speed and vehicle body posture, and the environmental information includes road markings, traffic signs and obstacle information around the vehicle.

3. The real data-driven automatic driving vehicle key signal testing method according to claim 2 is characterized in that: In step S2, the road test scene data extraction is specifically as follows: using the previously collected scene data to train a model for scene extraction, and then inputting all the data in the real vehicle database into the trained model for extraction; The drive test scenario generalization adopts a deep learning-based scenario generation model to expand diversified scenarios.

4. The method for testing key signals of an autonomous driving vehicle based on real data according to claim 3 is characterized in that: In step S3, the standard test case library is constructed, specifically including: Step S3.1: Define the test scope and objectives according to industry standards and clarify the coverage of the test case library; Step S3.2: Collect and analyze industry standards: Refer to the scenario definition methods specified in existing automotive industry standards or national standards to determine the standard framework and indicators of the test case library; Step S3.3: Collecting road test scenario data, specifically: collecting real road data, including normal scenarios, boundary scenarios, accident or dangerous scenarios, and extracting scenario data from existing standard documents, research results, and vehicle accident case databases; Step S3.4: Parameterization and generalization of scene features, specifically: extracting key signal parameters, grading, classifying, and labeling feature parameters to facilitate algorithm generalization and standardized management; Step S3.5: Generation of standardized test cases, including: Step S3.5.1: Data preprocessing: filtering and denoising the collected road scene data and removing abnormal data, and extracting the vehicle key characteristic parameter sequence; Step S3.5.2: Key feature clustering: clustering the road scene feature parameters using a clustering algorithm to identify typical representative scenes and boundary scenes; Step S3.5.3: Feature generalization algorithm: Use the generative adversarial network (GAN) or variational autoencoder (VAE) algorithm to generalize the data of typical scenes and generate more similar scenes with different parameters. Step S3.5.4: Use case screening algorithm: Evaluate the test importance of the generalized scenarios through an automatic screening algorithm based on importance; use the automatic screening algorithm for iterative optimization to select the most critical and representative scenarios to add to the standard use case library; Step S3.5.5: Standard formatted output, specifically: standardized document output of the test scenario, including: scenario number, scenario type, target test parameters, expected output results; and test execution process and specifications, including scenario entry method, exit method, and action execution timing.

5. The method for testing key signals of an autonomous driving vehicle based on real data according to claim 4 is characterized in that: In step S4, the test execution phase includes hierarchical or parallel testing of the same test case or standard test case in a virtual simulation platform, a hardware-in-the-loop (HIL) platform, and a real vehicle environment, and achieving seamless switching or parallel operation of test cases on different platforms through a scheduling module and an interface standardization protocol.

6. The real data-driven automatic driving vehicle key signal testing method according to claim 3 is characterized in that: In step S5, during the result feedback and closed-loop iteration, the artificial intelligence automated analysis module is used to compare and statistically analyze the response data of the test vehicle to extract abnormal signal responses, potential defects, and control failure information in edge scenarios.

7. A real data-driven autonomous driving vehicle key signal testing system, characterized in that: include: Data collection module: used to collect and store key signal data and environmental information during vehicle driving in real road scenes; Test case generation module: including a road test scenario extraction unit and a road test scenario generalization unit, which are used to extract real vehicle data from a unified database to generate a test case set covering normal working conditions, extreme working conditions and edge scenarios; Standard test case library module: establish a standard test case library set including the steering system standard test case library, the throttle / brake system standard test case library, and the body control system key signal standard test case library; Test execution module: transmits the test case to the test vehicle, controls the controller in the test vehicle to perform corresponding actions according to the scenario requirements in the test case, and records the vehicle state changes and response data. The test execution module includes a virtual simulation platform, a hardware-in-the-loop HIL platform and a real vehicle test unit, and realizes hierarchical or parallel testing of test cases on different platforms through a scheduling module and an interface standardization protocol; Result feedback module: Contains an artificial intelligence automated analysis unit, which is used to analyze the response data of the test vehicle, associate the analysis results with the test scenarios / cases, and feed them back to the test case generation module to achieve a closed-loop test process.

8. The real data-driven automatic driving vehicle key signal testing system according to claim 7, characterized in that: The sensors carried by the data collection module include cameras, radars, IMUs and GPS.

9. The real data-driven automatic driving vehicle key signal testing system according to claim 8, characterized in that: The drive test scenario extraction unit of the test case generation module adopts a machine learning algorithm, and the drive test scenario generalization unit adopts a scenario generation model based on deep learning.

10. The real data-driven automatic driving vehicle key signal testing system according to claim 9, characterized in that: The artificial intelligence automated analysis unit of the result feedback module is used to extract abnormal signal responses, potential defects, and control failure information in edge scenarios.

Images