Automatic driving function test system, method and device based on vehicle-road cloud integration

Abstract

The invention discloses an automatic driving function test system, method and device based on vehicle-road cloud integration, and the system comprises a test management module which is used for configuring initial test parameters, a test scene and an automatic driving model, completing the initialization of each stage of test passing standard and maximum test times, and issuing a test starting instruction; the reinforcement learning module is used for adjusting the initial test parameters based on a reinforcement learning algorithm to obtain optimal test parameters, replacing the initial test parameters in the test starting instruction with the optimal test parameters, and issuing the updated test starting instruction to the simulation test module; the simulation test module is used for calling the scene library to execute a simulation test, and synchronizing the optimal test parameters to the cloud when the test frequency is smaller than the maximum test frequency and the test result meets the standard; and the HIL test module is used for carrying out HIL test by utilizing the optimal test parameters and the automatic driving model, and generating an OTA deployment package for real vehicle test when the test times are smaller than the maximum test times and the test result meets the standard.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and more specifically, to an autonomous driving function testing system, method, and apparatus based on vehicle-road-cloud integration. Background Technology

[0002] The commercialization of autonomous driving technology relies on comprehensive, high-coverage, and highly reliable functional testing and verification across all scenarios. Current autonomous driving testing often employs a phased, independent testing model, with simulation testing, HIL (Hardware-in-Loop) testing, and real-vehicle testing executed independently. Manual coordination of parameter configuration and data transfer is required between phases. Therefore, this existing phased, independent testing model suffers from fragmented processes, high reliance on manual intervention, delayed parameter adjustments, and poor traceability, resulting in low testing efficiency, high costs, incomplete scenario coverage, and significant difficulties in troubleshooting. Summary of the Invention

[0003] This application provides an autonomous driving function testing system, method, and device based on vehicle-road-cloud integration, which can solve the technical problems of fragmented process, high dependence on manual labor, delayed parameter adjustment, and poor traceability in the phased independent testing mode. It realizes unmanned execution of the entire process of "reinforcement learning-simulation-HIL-OTA-real vehicle" and can automatically adjust test parameters to form a closed-loop optimization based on test results.

[0004] The specific technical solution is as follows: In a first aspect, embodiments of this application provide an autonomous driving function testing system based on vehicle-road-cloud integration, the system being applied in the cloud, and the system comprising: The test management module is used to initialize the system by configuring initial test parameters, test scenario, autonomous driving model, test pass standards for each stage, and maximum number of tests after receiving a test task, and to issue a test start command to the reinforcement learning module. The test start command includes the initial test parameters, the test scenario, the autonomous driving model, the test pass standards for each stage, and the maximum number of tests. The reinforcement learning module is used to obtain the initial test parameters in the test start command, and dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters. After replacing the initial test parameters in the test start command with the optimal test parameters, the updated test start command is sent to the simulation test module. The simulation test module is used to receive the updated test start command, and call the scenario library to execute the simulation test according to the updated test start command. When the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria of the simulation test stage, the optimal test parameters are synchronized to the preset storage space in the cloud. The Hardware-in-the-Loop (HIL) simulation test module is used to load the optimal test parameters synchronized in the cloud and the autonomous driving model configured by the test management module. It uses the optimal test parameters and the autonomous driving model to simulate the operation of vehicle hardware to achieve HIL testing. If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL testing phase, it generates an OTA deployment package including the autonomous driving model and transmits the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real-vehicle testing based on the OTA deployment package.

[0005] In one possible implementation, the test management module is further configured to receive test results reported by at least one of the simulation test module, the HIL test module, the real test vehicle, and the roadside equipment interacting with the real test vehicle. The simulation test module is also used to send a test termination command to the test management module when the number of simulation tests is equal to the maximum number of tests. The HIL test module is also used to send a test termination command to the test management module when the number of HIL tests is equal to the maximum number of tests. The test management module is also used to generate a standardized test report based on the received test results when the test termination instruction is received.

[0006] In one possible implementation, the simulation testing module is further configured to send a restart command to the reinforcement learning module when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation testing phase. The HIL testing module is also used to send a restart command to the reinforcement learning module when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the passing criteria for the HIL testing phase. The real test vehicle is also used to send a restart command to the reinforcement learning module when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test phase. The reinforcement learning module is further configured to, upon receiving the restart instruction, dynamically adjust the optimal test parameters based on the reinforcement learning algorithm to obtain updated optimal test parameters, so as to retest based on the updated optimal test parameters.

[0007] In one possible implementation, the system further includes: The digital twin module is used to visualize the entire testing process based on a digital twin model.

[0008] In one possible implementation, the digital twin module is also used to: collect key operational data from the test management module, the reinforcement learning module, the simulation test module, the HIL test module, the real test vehicle, and the roadside equipment in real time; when a certain dimension of data is detected to exceed a preset abnormal threshold, automatically generate a visual warning sign and simultaneously trigger a warning signal to the test management module; The test management module is also used to perform target operations based on the warning level represented by the warning signal when the warning signal is received. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning module to adjust parameters.

[0009] Secondly, embodiments of this application provide a method for testing autonomous driving functions based on vehicle-road-cloud integration, the method comprising: Upon receiving the test task, initialization is completed by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage, and maximum number of tests. The initial test parameters are dynamically adjusted based on the reinforcement learning algorithm to obtain the optimal test parameters; Simulation tests are performed by calling the scenario library according to the optimal test parameters, the test scenario, and the autonomous driving model. If the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria for the simulation test stage, the vehicle hardware operation process is simulated using the optimal test parameters and the autonomous driving model to perform HIL tests. If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase, an OTA deployment package including the autonomous driving model is generated, and the OTA deployment package is transmitted to the real test vehicle so that the real test vehicle can conduct real vehicle testing based on the OTA deployment package.

[0010] In one possible implementation, the method further includes: Receive at least one of the following: simulation test results, HIL test results, real vehicle test results, and test data reported by roadside equipment that interacts with the real test vehicle; If the number of simulation tests equals the maximum number of tests, and / or if the number of HIL tests equals the maximum number of tests, a standardized test report is generated based on the received test results.

[0011] In one possible implementation, the method further includes: If the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation test phase, and / or if the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the pass criteria for the HIL test phase, and / or if the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test phase, the optimal test parameters are dynamically adjusted based on the reinforcement learning algorithm to obtain updated optimal test parameters, so that the test can be re-performed based on the updated optimal test parameters.

[0012] In one possible implementation, the method further includes: The entire testing process is visualized using a digital twin model.

[0013] In one possible implementation, the method further includes: Key operational data from each stage are collected in real time. When a certain dimension of data exceeds a preset abnormal threshold, a visual warning sign is automatically generated, and a target operation is executed according to the warning level. The target operation includes pausing the test, resetting parameters, or triggering the reinforcement learning module to adjust parameters.

[0014] Thirdly, embodiments of this application provide an autonomous driving function testing device based on vehicle-road-cloud integration, the device comprising: The initialization unit is used to complete the initialization by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests after receiving the test task; The adjustment unit is used to dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters; The simulation testing unit is used to execute simulation tests by calling the scenario library according to the optimal test parameters, the test scenario, and the autonomous driving model. The HIL test unit is used to perform HIL testing by simulating the operation of vehicle hardware using the optimal test parameters and the autonomous driving model, provided that the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria for the simulation test phase. The generation unit is used to generate an OTA deployment package including an autonomous driving model when the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase. The transmission unit is used to transmit the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real vehicle testing based on the OTA deployment package.

[0015] In one possible implementation, the device further includes: The receiving unit is used to receive at least one of the following: simulation test results, HIL test results, real vehicle test results, and test data reported by roadside equipment that interacts with the real test vehicle; The generation unit is further configured to generate a standardized test report based on the received test results when the number of simulation tests equals the maximum number of tests, and / or when the number of HIL tests equals the maximum number of tests.

[0016] In one possible implementation, the adjustment unit is further configured to dynamically adjust the optimal test parameters based on the reinforcement learning algorithm to obtain updated optimal test parameters when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation test stage, and / or when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the pass criteria for the HIL test stage, and / or when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test stage, so as to retest based on the updated optimal test parameters.

[0017] In one possible implementation, the device further includes: The display unit is used to visualize the entire testing process based on a digital twin model.

[0018] In one possible implementation, the device further includes: The data acquisition unit is used to collect key operational data from each stage in real time. The generation unit is also used to automatically generate a visual warning sign when it detects that data in a certain dimension exceeds a preset abnormal threshold. The execution unit is used to perform target operations according to the warning level. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning module to adjust parameters.

[0019] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any possible implementation of the first aspect.

[0020] Fifthly, embodiments of this application provide an electronic device, which includes: One or more processors; The processor is coupled to a storage device for storing one or more programs; When one or more programs are executed by one or more processors, the electronic device performs the method as described in any possible implementation of the second aspect.

[0021] In a sixth aspect, embodiments of this application provide a computer program product containing instructions that, when executed on a computer or processor, cause the computer or processor to perform the method described in any possible implementation of the second aspect.

[0022] As can be seen from the above scheme, the autonomous driving function testing system, method and device based on vehicle-road-cloud integration provided in this application embodiment, through the test management module to coordinate the initialization and command issuance of the entire process, and with the help of the reinforcement learning module to dynamically optimize test parameters, link the simulation test module, HIL test module and real test vehicle to achieve seamless connection and unmanned execution of each stage of "reinforcement learning-simulation-HIL-OTA-real vehicle", which not only eliminates the error and efficiency bottleneck caused by manual intervention and significantly reduces test costs, but also forms a closed-loop adaptive optimization of the entire process based on the test results of each stage through the reinforcement learning algorithm, effectively improving the coverage and relevance of test scenarios and accurately discovering potential faults. At the same time, relying on deep vehicle-road-cloud collaboration to open up the data link of "roadside-cloud-vehicle", it ensures seamless interoperability and consistency of test data at each stage, and comprehensively improves the efficiency, reliability and comprehensiveness of autonomous driving function testing.

[0023] Furthermore, the technical effects that can be achieved by the embodiments of this application include: 1. Based on achieving full-process unmanned operation, adaptive closed-loop optimization, and deep vehicle-road-cloud collaboration, the test management module receives all test results reported by the simulation test module, HIL test module, real test vehicles, and roadside equipment. When the maximum number of tests is reached in each test stage, a standardized test report is generated based on the termination command, which greatly improves the traceability of the test process and the convenience of troubleshooting. 2. In each stage of simulation testing, HIL testing, and real vehicle testing, when the number of tests has not exceeded the maximum limit and the results do not meet the passing standard of the corresponding stage, the corresponding test execution entity automatically sends a restart command to the reinforcement learning module. This triggers the reinforcement learning module to dynamically iterate and optimize the current optimal test parameters based on the reinforcement learning algorithm and generate updated optimal parameters, thereby driving the test to restart. This mechanism not only allows adaptive closed-loop optimization to run through every key node of the entire testing process, effectively avoiding test interruptions or fault omissions caused by improper parameter configuration in a single instance, but also significantly improves the ability to accurately capture complex scenarios and potential faults through continuous iteration and refinement of parameters. At the same time, no manual intervention or decision-making is required throughout the process, which not only continues the high efficiency advantage of the fully unmanned process, but also further enhances the targeting and reliability of the test, ensuring that the test process can autonomously complete parameter optimization and test retry within the preset rules, significantly improving the overall test success rate and the effectiveness of the test results.

[0024] 3. Based on the digital twin model, the entire testing process of "reinforcement learning - simulation - HIL - OTA - real vehicle" is visualized. This not only makes the originally unmanned testing process intuitive and monitorable, allowing staff to keep track of the testing progress, dynamic parameter adjustments, and test result feedback at each stage in real time, but also accurately recreates key scenarios and potential fault occurrence links in the testing process. Combined with standardized test reports, this further improves the traceability of the testing process and the convenience of problem troubleshooting. At the same time, combined with the deep collaboration mechanism of vehicle-road-cloud, the data flow and business flow of "roadside-cloud-vehicle" are presented in a visual form, making the testing collaboration process more transparent and controllable, further strengthening the standardization and rigor of the testing process, and providing more comprehensive support for the efficient advancement and quality assurance of autonomous driving function testing.

[0025] 4. The digital twin module, by collecting real-time key operational data from the test management module, reinforcement learning module, various test execution modules, real test vehicles, and roadside equipment across all dimensions, constructs a dynamic data perception system for the entire testing process. This not only further enhances the visualization and monitoring capabilities of the testing process but also automatically generates intuitive visual warning indicators and simultaneously sends warning signals to the test management module when any dimension of data exceeds a preset anomaly threshold. Based on the warning level represented by the warning signal, the test management module precisely executes target operations such as test pause, parameter reset, or triggering the reinforcement learning module to adjust parameters. This achieves rapid response and intelligent handling of test anomalies, effectively avoiding the waste of test resources or distortion of test results caused by the escalation of anomalies. It ensures the stable and orderly progress of the unmanned testing process and further improves the safety, reliability, and overall efficiency of testing by proactively avoiding potential risks, thus perfecting the closed-loop management capabilities of the vehicle-road-cloud integrated testing system. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0027] Figure 1 A schematic diagram of the structure of an autonomous driving function testing system based on vehicle-road-cloud integration provided in an embodiment of this application; Figure 2 A schematic diagram of another autonomous driving function testing system based on vehicle-road-cloud integration provided in this application embodiment; Figure 3 A flowchart illustrating a vehicle-road-cloud integrated autonomous driving function testing method provided in this application embodiment. Figure 4 This is a block diagram of an autonomous driving function testing device based on vehicle-road-cloud integration, provided as an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0029] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0030] This application provides an autonomous driving function testing system based on vehicle-road-cloud integration. The system is applied in the cloud, such as... Figure 1 As shown, the system includes: The test management module 110 is used to initialize the test by configuring initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests after receiving the test task, and to issue a test start command to the reinforcement learning module 120. The test start command includes initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests. The reinforcement learning module 120 is used to obtain the initial test parameters in the test start command, and dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters. After replacing the initial test parameters in the test start command with the optimal test parameters, the updated test start command is sent to the simulation test module 130. The simulation test module 130 is used to receive the updated test start command and call the scenario library to execute the simulation test according to the updated test start command. If the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria of the simulation test stage, the optimal test parameters are synchronized to the preset storage space in the cloud. The Hardware-in-the-Loop (HIL) test module 140 is used to load the optimal test parameters synchronized in the cloud and the autonomous driving model configured by the test management module 110. It uses the optimal test parameters and the autonomous driving model to simulate the operation of the vehicle hardware and realize HIL testing. If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria of the HIL test stage, it generates an OTA deployment package including the autonomous driving model and transmits the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real vehicle testing based on the OTA deployment package.

[0031] In this embodiment, the user can submit a test task based on the test management module 110, configure the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests to complete the initialization. After the initialization is completed, the test management module 110 can automatically issue a test start command to the reinforcement learning module 120, and the administrator can issue a test start command to the reinforcement learning module 120 based on the test management module 110.

[0032] Among them, the test tasks refer to the verification requirements of autonomous driving functions that need to be performed, such as "reliability test of adaptive cruise function of a certain model of vehicle in high-speed scenario" and "verification of left turn obstacle avoidance function at complex intersections" and other specific test objectives.

[0033] Initial test parameters include vehicle operating parameters (such as initial vehicle speed and acceleration threshold), sensor parameters (such as radar detection distance and camera frame rate), and test environment parameters (such as road friction coefficient and light intensity), which are the basis for test execution.

[0034] Test scenarios refer to simulated operating environments and interaction scenarios for autonomous vehicles, such as urban traffic congestion scenarios, highway overtaking scenarios, rainy night driving scenarios, and pedestrian crossing scenarios at intersections. These scenarios need to be selected or configured from the scenario library.

[0035] The pass standards for each stage of testing refer to the qualification criteria set for different testing stages such as "simulation testing, HIL testing, and real vehicle testing". For example, quantitative or qualitative requirements such as "1000 consecutive runs without collision failure" in simulation testing, "sensor data transmission delay ≤50ms in HIL testing" and "obstacle avoidance response time ≤0.3s in real vehicle testing".

[0036] The maximum number of tests refers to the upper limit threshold set to avoid infinite testing loops (such as 3 or 5 times). If the number of tests in a certain stage reaches this threshold and still fails to meet the passing standard, the test will be terminated or the optimization process will be restarted.

[0037] The test start command is a "command package" generated by the test management module 110 that triggers subsequent tests. It contains all the core information configured during the initialization phase, ensuring that the module receiving the command (such as the reinforcement learning module) can clearly understand the test objectives, rules, and basic configuration.

[0038] The reinforcement learning module 120 is the core unit in the system responsible for dynamically optimizing test parameters. Based on the reinforcement learning algorithm, it adjusts parameters according to the test results at each stage, improving the relevance and effectiveness of the test. Reinforcement learning is a machine learning algorithm based on an "environment interaction-reward feedback" mechanism. In this testing system, the core is to use the "results" of each test stage (such as whether the pass criteria are met, the frequency of fault occurrence) as "reward / penalty signals," continuously iterating and adjusting test parameters to ultimately find the parameter combination that maximizes test effectiveness (such as covering more complex scenarios and accurately triggering potential faults), rather than training the autonomous driving model itself. Dynamic adjustment differs from traditional static parameter configuration. The reinforcement learning module 120 does not rely on human intervention but autonomously iterates and optimizes parameters based on real-time data generated during the test (results from subsequent test stages). This adjustment process runs throughout the entire testing process, not just during the initialization phase. The optimal test parameters are a combination of parameters that can balance "test efficiency, scenario coverage, and fault detection probability" after iterative optimization through reinforcement learning algorithms (for example, for rainy and slippery road scenarios, optimized sensor detection distance parameters, vehicle braking response thresholds, etc., can more easily trigger the anti-slip control logic vulnerabilities of the autonomous driving system). Its core is to adapt to the current test scenario and test target, rather than an absolutely fixed "optimal value".

[0039] Autonomous driving models are computational models trained on massive amounts of real-world data (including multimodal sensor inputs, driving decisions, and vehicle control signals). Their core objective is to simulate or surpass human driving cognition and operational capabilities. They can directly learn environmental characteristics, driving strategies, and vehicle control principles from perceived signals, achieving a mapping from environmental perception to driving decisions. This is equivalent to digitally replicating the human driver's "perception-decision-execution" closed loop at the algorithmic level, forming the algorithmic foundation for intelligent decision-making and control in autonomous driving systems.

[0040] The scenario library is a database that stores a massive number of standardized and structured test scenarios. It contains the core parameters of the scenarios (such as road type, traffic participant status, weather conditions, and emergency triggering rules). It supports on-demand calling, combination, or custom expansion according to the test objectives. For example, scenarios such as "straight driving on urban roads + pedestrian crossing" and "following another vehicle on highways + sudden braking of the vehicle in front" have preset parameters and are stored in the library for direct use by the simulation test module.

[0041] After receiving the test start command, the simulation test module 130 automatically calls the scenario library to execute the test and collects simulation data such as roadside perception, vehicle operation, and scenario interaction in real time.

[0042] HIL testing is a testing technology that integrates "real hardware" and "virtual simulation". Its core is to connect the key hardware of the autonomous driving system (such as the vehicle controller ECU (Electronic Control Unit), sensors, actuators and other physical components) into the test loop, and simulate the vehicle's operating environment, road scenarios and data interaction logic through a virtual model. This allows the real hardware to operate in a virtual environment that is close to the actual working conditions. It avoids the limitation of pure simulation testing that "deviates from the characteristics of real hardware", and solves the problem of "high risk and high cost" of direct real vehicle testing. It is a core transition link connecting virtual verification and real vehicle verification.

[0043] The HIL test module 140 establishes communication with the real vehicle hardware through hardware interfaces (such as CAN (Controller Area Network) bus, Ethernet interface), converts the scene data generated by the autonomous driving model (such as sensor detection data, road resistance data) into electrical signals that the hardware can recognize and inputs them to the real hardware. At the same time, it collects the output response of the hardware (such as the controller's control commands, the actuator's action feedback), forming a closed loop of "virtual scene - real hardware - data feedback", accurately reproducing the working state of the real hardware in actual driving.

[0044] OTA (Over-the-Air) deployment packages are standardized data compressed packages containing autonomous driving models (core software such as decision-making and control algorithms) that have been tested and verified by HIL, optimal test parameters, scenario configuration information, etc. They support remote transmission to real test vehicles via wireless communication (such as 5G, vehicle-to-everything). There is no need for manual on-site disassembly of the vehicle or connection of equipment for configuration, which enables rapid and lossless deployment of test parameters and models.

[0045] Real test vehicles are physical vehicles equipped with complete autonomous driving hardware systems (such as physical components like LiDAR, cameras, and onboard controllers) and basic software platforms. They serve as the final verification vehicle for the testing process, used to verify the actual performance of autonomous driving functions on real roads or closed test tracks. They also receive OTA deployment packages issued by HIL test module 140 and initiate real vehicle testing.

[0046] The autonomous driving function testing system based on vehicle-road-cloud integration provided in this application embodiment coordinates the entire process of initialization and command issuance through the test management module, dynamically optimizes test parameters with the help of the reinforcement learning module, and links the simulation test module, HIL test module and real test vehicle to achieve seamless connection and unmanned execution of each stage of "reinforcement learning-simulation-HIL-OTA-real vehicle". It not only eliminates the error and efficiency bottleneck caused by human intervention and significantly reduces test costs, but also forms a closed-loop adaptive optimization of the entire process based on the test results of each stage through the reinforcement learning algorithm, effectively improving the coverage and relevance of test scenarios and accurately discovering potential faults. At the same time, relying on deep vehicle-road-cloud collaboration to open up the data link of "roadside-cloud-vehicle", it ensures seamless interoperability and consistency of test data at each stage, and comprehensively improves the efficiency, reliability and comprehensiveness of autonomous driving function testing.

[0047] In one possible implementation, the test management module 110 is also used to receive test results reported by at least one of the simulation test module 130, the HIL test module 140, the real test vehicle, and the roadside equipment that interacts with the real test vehicle. The simulation test module 130 is also used to send a test termination command to the test management module 110 when the number of simulation tests equals the maximum number of tests. The HIL test module 140 is also used to send a test termination command to the test management module 110 when the number of HIL tests equals the maximum number of tests. The test management module 110 is also used to generate a standardized test report based on the received test results when a test termination instruction is received.

[0048] In this embodiment of the application, the intelligent terminal devices deployed on the roadside in the vehicle-road-cloud integrated architecture include roadside cameras, lidar, millimeter-wave radar, and RSU (Road Side Unit). Their core functions are to perceive the road environment in real time (such as the location of traffic participants and road condition information) and to communicate with real test vehicles (transmitting environmental data and collaborative instructions). The test data they report can serve as a supplementary verification basis for the real vehicle test results, making the test results more comprehensive (for example, if the roadside device perceives that the vehicle has not avoided an obstacle in time, it can be cross-checked with the test results reported by the vehicle itself).

[0049] A standardized test report is a test result document generated following a unified data format, content framework, and evaluation criteria. Its core contents include basic test information (test task, scenario, parameters), test process at each stage (number of times, execution status), test results (whether the pass criteria are met, fault type and occurrence point), and data statistics (such as the number of parameter optimization iterations and the average response latency). It is structured, traceable, and comparable, avoiding the problems of "disorganized format and fragmented data" in traditional test reports. It makes it easier for testers to quickly locate problems and evaluate test results, and also provides standardized data support for subsequent test optimization or autonomous driving model iteration.

[0050] To avoid infinite test loops, both the simulation test module 130 and the HIL test module 140 are configured with "test count threshold control" logic. When the number of tests performed by the module reaches the "maximum number of tests" preset by the test management module 110 (regardless of whether the test results meet the pass criteria), a "test termination command" will be automatically sent to the test management module 110 to clearly inform that the current stage of testing has reached its limit, triggering the process termination.

[0051] Based on the realization of fully unmanned operation, adaptive closed-loop optimization, and deep vehicle-road-cloud collaboration, the embodiments of this application fully receive test results reported by the simulation test module, HIL test module, real test vehicles and roadside equipment through the test management module, and generate standardized test reports based on termination commands when the maximum number of tests is reached in each test stage, which greatly improves the traceability of the test process and the convenience of problem troubleshooting.

[0052] In one possible implementation, the simulation testing module 130 is further configured to send a restart command to the reinforcement learning module 120 when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation testing phase. The HIL testing module 140 is also used to send a restart command to the reinforcement learning module 120 when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the passing criteria of the HIL testing phase. The real test vehicle is also used to send a restart command to the reinforcement learning module 120 when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test phase. The reinforcement learning module 120 is also used to dynamically adjust the optimal test parameters based on the reinforcement learning algorithm when a restart command is received, so as to obtain the updated optimal test parameters and retest based on the updated optimal test parameters.

[0053] In this embodiment, the "fault adaptive iteration mechanism" of the testing system is a core extension of the full-process closed-loop optimization, realizing an unmanned autonomous loop of "test failure - parameter adjustment - retesting": Triggering condition determination: Simulation test module 130, HIL test module 140 and real test vehicle will monitor their own test status in real time. When the conditions are met that "the current number of tests has not exceeded the preset maximum number of tests" (there is still room for optimization and retry) and "the test result has not met the pass standard for this stage" (insufficient parameter adaptability caused test failure), a restart command will be automatically generated to clarify the parameter optimization requirements for the current test stage.

[0054] Command flow logic: Regardless of which testing stage (virtual simulation, hardware verification, real vehicle testing) the above situation occurs, the restart command will be directly sent to reinforcement learning module 120 (the core of system parameter optimization) to ensure that optimization requirements can be transmitted quickly without manual intervention for judgment or forwarding.

[0055] Parameter Iterative Optimization: After receiving a restart command, the reinforcement learning module 120 uses the "feedback learning logic" of the reinforcement learning algorithm to treat the failure result of this test (such as fault type or unmet target) as a "penalty signal," and traces back to check the adaptation problem of the original optimal test parameters, thereby making targeted dynamic adjustments to the parameters. For example, if the HIL test fails to meet the controller response delay target, the algorithm will optimize relevant parameters such as "data transmission baud rate" and "control command trigger threshold" to generate updated optimal test parameters.

[0056] Retest Trigger: After parameter optimization is completed, the system will restart the currently failed test phase based on the updated optimal test parameters to verify whether the parameter adjustment can solve the previous test failure problem, forming a micro-loop of "test-feedback-optimization-retest".

[0057] In this embodiment, when the number of tests has not exceeded the maximum limit and the results do not meet the passing standard for the corresponding stage in each stage of simulation testing, HIL testing, and real vehicle testing, the corresponding test execution entity automatically sends a restart command to the reinforcement learning module. This triggers the reinforcement learning module to dynamically iterate and optimize the current optimal test parameters based on the reinforcement learning algorithm and generate updated optimal parameters, thereby driving the test to restart. This mechanism not only allows adaptive closed-loop optimization to run through every key node of the entire testing process, effectively avoiding test interruptions or fault omissions caused by improper parameter configuration in a single instance, but also significantly improves the ability to accurately capture complex scenarios and potential faults through continuous iteration and refinement of parameters. At the same time, no manual intervention or decision-making is required throughout the process, which not only continues the high efficiency advantage of unmanned operation of the entire process, but also further enhances the targeting and reliability of the test, ensuring that the test process can autonomously complete parameter optimization and test retry within the preset rules, significantly improving the overall test success rate and the effectiveness of the test results.

[0058] In one possible implementation, such as Figure 2 As shown, the system also includes: The digital twin module 150 is used to visualize the entire testing process based on a digital twin model.

[0059] In this embodiment, the digital twin model is not a simple virtual replica, but a dynamic mapping model that integrates vehicle physical characteristics, roadside environmental elements, cloud command flow logic, and the operating rules of each test module. It synchronizes in real time the command status of the test management module, the parameter optimization trajectory of the reinforcement learning module, the test execution progress of the simulation / HIL / real vehicle, and the interaction data between the real vehicle and roadside equipment, forming a virtual mirror that is "synchronized within seconds" with the real testing process, ensuring the authenticity and real-time nature of the visualization.

[0060] Visualization refers to transforming abstract data (such as parameter adjustment values, test progress percentages, fault occurrence points, and data flow paths) throughout the entire testing process into intuitive graphical representations. These representations include real-time dynamic flowcharts, parameter change curves, 3D simulation interfaces of test scenarios, and module operation status indicators, allowing staff to quickly grasp the overall testing situation without analyzing raw data.

[0061] This application's embodiments visualize the entire testing process of "reinforcement learning - simulation - HIL - OTA - real vehicle" based on a digital twin model. This not only makes the originally unmanned testing process intuitive and monitorable, allowing staff to grasp the testing progress, parameter dynamics, and test result feedback at each stage in real time, but also accurately recreates key scenarios and potential fault occurrence links in the testing process. Combined with standardized test reports, it further enhances the traceability of the testing process and the convenience of problem troubleshooting. At the same time, combined with the deep collaboration mechanism of vehicle-road-cloud, it presents the data flow and business flow of "roadside-cloud-vehicle" in a visual form, making the testing collaboration process more transparent and controllable, further strengthening the standardization and rigor of the testing process, and providing more comprehensive support for the efficient advancement and quality assurance of autonomous driving function testing.

[0062] In one possible implementation, the digital twin module 150 is also used to: collect key operational data from the test management module 110, reinforcement learning module 120, simulation test module 130, HIL test module, real test vehicles, and roadside equipment in real time; when a certain dimension of data is detected to exceed a preset abnormal threshold, a visual warning sign is automatically generated, and a warning signal is simultaneously triggered to the test management module 110. The test management module 110 is also used to perform target operations based on the warning level represented by the warning signal when a warning signal is received. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning module 120 to adjust parameters.

[0063] In this application embodiment, key operational data refers to core data that reflects the core working status of each module and device and the effectiveness of the test process. Specific dimensions include: module operation status data (such as the instruction issuance delay of the test management module and the parameter optimization iteration speed of the reinforcement learning module), parameter data (such as the current test parameter values ​​and the dynamic parameters of the autonomous driving model), interaction data (such as the data transmission delay of "vehicle-road-cloud" and the data packet loss rate), and test process data (such as the fault trigger frequency of simulation test and the hardware response time of HIL test), etc., which are the core basis for judging whether the system is operating normally.

[0064] The preset anomaly thresholds are "reasonable range boundaries" for key data in various dimensions, based on industry standards, system design indicators, and historical testing experience for autonomous driving testing. For example, the data transmission delay threshold is set to 50ms (exceeding this threshold is considered an anomaly), the reinforcement learning module parameter optimization timeout threshold is set to 10s (exceeding this timeout is considered an anomaly), and the valid range of test parameters is set to [0.1, 1.0] (exceeding this range is considered an anomaly). These are the core reference standards for anomaly monitoring.

[0065] Visualized early warning labels are intuitive anomaly alerts generated by the digital twin module and presented in a visual display interface. For example, anomaly modules / devices are marked with flashing red icons, anomaly data dimensions and exceeding values ​​are displayed in pop-ups, and warning levels are distinguished by color gradients (green-yellow-red). This allows staff to quickly locate the anomaly and its severity without analyzing the raw data, thus achieving anomaly visualization.

[0066] The warning level is a hierarchical standard based on the severity and scope of the anomaly (such as minor anomaly, moderate anomaly, and severe anomaly). For example, "parameters slightly exceeding the effective range but not affecting test progress" is a minor anomaly, "data transmission delay exceeding the standard causing test data distortion" is a moderate anomaly, and "module operation crash or hardware failure risk" is a severe anomaly. It is the core basis for the test management module to execute differentiated responses.

[0067] Targeted actions refer to the standardized handling procedures formulated by the test management module for different warning levels, which are essentially a "graded anomaly control" strategy. Minor anomalies trigger parameter adjustments, moderate anomalies trigger parameter resets, and severe anomalies trigger test suspension, avoiding the loss of testing efficiency or increased risk caused by a "one-size-fits-all" approach.

[0068] The digital twin module provided in this application constructs a dynamic data perception system for the entire testing process by collecting real-time key operational data from the test management module, reinforcement learning module, various test execution modules, real test vehicles, and roadside equipment. This not only further enhances the visualization and monitoring capabilities of the testing process but also automatically generates intuitive visual warning indicators and simultaneously sends warning signals to the test management module when any dimension of data exceeds a preset abnormal threshold. Based on the warning level represented by the warning signal, the test management module accurately executes target operations such as test suspension, parameter reset, or triggering the reinforcement learning module to adjust parameters. This achieves rapid response and intelligent handling of test anomalies, effectively avoiding the waste of test resources or distortion of test results caused by the expansion of anomalies. It ensures the stable and orderly progress of the unmanned testing process and further improves the safety, reliability, and overall efficiency of the test by avoiding potential risks in advance, thus perfecting the closed-loop management capability of the vehicle-road-cloud integrated testing system.

[0069] Based on the above system embodiments, another embodiment of this application provides a method for testing autonomous driving functions based on vehicle-road-cloud integration. This method is applied in the cloud, such as... Figure 3 As shown, the method includes: S210: After receiving the test task, initialize by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage, and maximum number of tests.

[0070] Users can submit test tasks via the cloud, configuring initial test parameters, test scenarios, autonomous driving models, pass criteria for each stage of testing, and the maximum number of tests to complete the initialization. Initial test parameters include vehicle operating parameters (such as initial speed and acceleration thresholds), sensor parameters (such as radar detection range and camera frame rate), and test environment parameters (such as road friction coefficient and light intensity), which form the basis for test execution. Test scenarios refer to the simulated operating environment and interaction scenarios of autonomous vehicles. Pass criteria for each stage of testing refer to the qualification conditions set for different test stages such as simulation testing, HIL testing, and real-vehicle testing. The maximum number of tests is an upper limit threshold set to avoid infinite testing loops.

[0071] S220: The initial test parameters are dynamically adjusted based on the reinforcement learning algorithm to obtain the optimal test parameters.

[0072] Reinforcement learning is a machine learning algorithm based on an "environment interaction-reward feedback" mechanism. In this testing system, the core is to use the "results" of each testing stage (such as whether the pass criteria are met, the frequency of fault occurrence) as "reward / penalty signals" to continuously iterate and adjust the testing parameters, ultimately finding the parameter combination that maximizes the testing effect (such as covering more complex scenarios and accurately triggering potential faults), rather than training the autonomous driving model itself. Dynamic adjustment differs from traditional static parameter configuration. It means that the reinforcement learning process does not rely on human intervention, but rather autonomously iterates and optimizes the parameters based on real-time data generated during the testing process (the results of feedback from subsequent testing stages). The adjustment process runs through the entire testing process, not just during the initialization phase. The optimal testing parameters are the parameter combinations that, after iterative optimization by the reinforcement learning algorithm, can balance "testing efficiency, scenario coverage, and fault detection probability" (for example, for slippery roads in rainy weather, optimized sensor detection distance parameters and vehicle braking response thresholds can more easily trigger vulnerabilities in the anti-slip control logic of the autonomous driving system). Its core is to adapt to the current testing scenario and testing objective, rather than an absolutely fixed "optimal value".

[0073] S230: Based on the optimal test parameters, test scenario, and autonomous driving model, call the scenario library to perform simulation tests. If the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria for the simulation test stage, use the optimal test parameters and autonomous driving model to simulate the operation of the vehicle hardware and perform HIL tests.

[0074] The autonomous driving model is a computational model trained on massive amounts of real-world data (including multimodal sensor inputs, driving decisions, and vehicle control signals). Its core objective is to simulate or surpass human driving cognition and operational capabilities. It can directly learn environmental characteristics, driving strategies, and vehicle control laws from perception signals, realizing the mapping from environmental perception to driving decisions. This is equivalent to digitally reproducing the "perception-decision-execution" closed loop of a human driver at the algorithmic level, and it forms the algorithmic foundation for autonomous driving systems to achieve intelligent decision-making and control.

[0075] The scenario library is a database that stores a massive number of standardized and structured test scenarios. It contains the core parameters of the scenarios (such as road type, traffic participant status, weather conditions, and emergency triggering rules). It supports on-demand calling, combination, or custom expansion according to the test objectives. For example, scenarios such as "straight driving on urban roads + pedestrian crossing" and "following another vehicle on highways + sudden braking of the vehicle in front" have preset parameters and are stored in the library for direct use during simulation testing.

[0076] HIL testing is a testing technology that integrates "real hardware" and "virtual simulation". Its core is to connect the key hardware of the autonomous driving system (such as the vehicle controller ECU, sensors, actuators and other physical components) into the test loop, and simulate the vehicle's operating environment, road scenarios and data interaction logic through a virtual model. This allows the real hardware to run in a virtual environment that is close to the actual working conditions. It avoids the limitation of pure simulation testing that "deviates from the characteristics of real hardware", and solves the problem of "high risk and high cost" of direct real vehicle testing. It is a core transition link connecting virtual verification and real vehicle verification.

[0077] S240: If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase, generate an OTA deployment package including the autonomous driving model and transmit the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real vehicle tests based on the OTA deployment package.

[0078] The OTA deployment package is a standardized data compressed package containing autonomous driving models (core software such as decision-making and control algorithms) that have been tested and verified by HIL, optimal test parameters, and scenario configuration information. It supports remote transmission to real test vehicles via wireless communication (such as 5G and vehicle-to-everything), without the need for manual on-site disassembly of vehicles or connection of equipment for configuration, thus enabling rapid and lossless deployment of test parameters and models.

[0079] Real-world test vehicles are physical vehicles equipped with complete autonomous driving hardware systems (such as LiDAR, cameras, vehicle controllers, and other physical components) and basic software platforms. They serve as the final verification vehicle for the testing process, used to verify the actual performance of autonomous driving functions on real roads or closed test tracks, and to receive OTA deployment packages and initiate real-world vehicle testing.

[0080] The autonomous driving function testing method based on vehicle-road-cloud integration provided in this application first completes the initial configuration of initial test parameters, test scenarios, pass standards for each stage, and maximum number of tests. Then, it dynamically adjusts the initial test parameters based on reinforcement learning algorithms to obtain optimal test parameters. Subsequently, it sequentially executes simulation tests, HIL tests, and transmits OTA deployment packages to real test vehicles. This full-process design achieves seamless connection and unmanned advancement of each stage of "reinforcement learning - simulation - HIL - OTA - real vehicle". It not only eliminates the error and efficiency bottlenecks caused by manual intervention and significantly reduces testing costs, but also forms a closed-loop adaptive adjustment of the entire process through reinforcement learning-driven parameter dynamic optimization, effectively improving the coverage and relevance of test scenarios and accurately discovering potential faults. At the same time, relying on the vehicle-road-cloud integrated architecture to open up the data link of "roadside - cloud - vehicle", it ensures seamless interoperability and consistency of test data at each stage, and comprehensively improves the efficiency, reliability, and comprehensiveness of autonomous driving function testing.

[0081] In one possible implementation, the method further includes: receiving at least one of simulation test results, HIL test results, real vehicle test results, and test data reported by roadside equipment interacting with the real test vehicle; and generating a standardized test report based on the received test results when the number of simulation tests equals the maximum number of tests, and / or when the number of HIL tests equals the maximum number of tests.

[0082] In this embodiment of the application, the intelligent terminal devices deployed on the roadside in the vehicle-road-cloud integrated architecture include roadside cameras, lidar, millimeter-wave radar, roadside units (RSUs), etc. The core functions are to perceive the road environment in real time (such as the location of traffic participants and road condition information) and to conduct vehicle-road communication with real test vehicles (transmitting environmental data and collaborative instructions). The test data reported by the devices can serve as a supplementary verification basis for the real vehicle test results, making the test results more comprehensive (for example, if the roadside device perceives that the vehicle has not avoided obstacles in time, it can be cross-checked with the test results reported by the vehicle itself).

[0083] A standardized test report is a test result document generated following a unified data format, content framework, and evaluation criteria. Its core contents include basic test information (test task, scenario, parameters), test process at each stage (number of times, execution status), test results (whether the pass criteria are met, fault type and occurrence point), and data statistics (such as the number of parameter optimization iterations and the average response latency). It is structured, traceable, and comparable, avoiding the problems of "disorganized format and fragmented data" in traditional test reports. It makes it easier for testers to quickly locate problems and evaluate test results, and also provides standardized data support for subsequent test optimization or autonomous driving model iteration.

[0084] To avoid infinite test loops, simulation tests and HIL tests each have "test count threshold control" logic. When the number of tests reaches the preset "maximum number of tests" (regardless of whether the test results meet the pass criteria), a process termination operation is triggered.

[0085] This application's embodiments, based on achieving seamless connection and unmanned advancement across all stages, closed-loop adaptive optimization, and data collaboration and interoperability, ensure the comprehensiveness and completeness of test information collection by receiving multi-dimensional test data from simulation tests, HIL tests, real vehicle tests, and roadside equipment reports. Simultaneously, a test termination command is triggered when the simulation or HIL test reaches its maximum number of iterations, and a standardized test report is generated based on all received test results. This not only ensures the test process concludes in an orderly manner within preset rules, improving the standardization and controllability of the test, but also clearly presents the entire test process and results through a standardized report format, significantly improving the traceability of test data and the convenience of problem troubleshooting. It further enhances the closed-loop management capabilities of the vehicle-road-cloud integrated testing system, providing accurate and reliable basis for quality assessment and subsequent optimization of autonomous driving function tests, and comprehensively strengthening the rigor and practicality of the test.

[0086] In one possible implementation, the method further includes: when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation test phase, and / or when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the pass criteria for the HIL test phase, and / or when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test phase, dynamically adjusting the optimal test parameters based on a reinforcement learning algorithm to obtain updated optimal test parameters, so as to retest based on the updated optimal test parameters.

[0087] In this embodiment, the "fault adaptive iteration mechanism" of the testing system is a core extension of the full-process closed-loop optimization, realizing an unmanned autonomous loop of "test failure - parameter adjustment - retesting": Triggering conditions: Simulation test, HIL test, and real vehicle test will monitor their own test status in real time. When the conditions are met that "the current number of tests has not exceeded the preset maximum number of tests" (there is still room for optimization and retry) and "the test result has not met the pass standard for this stage" (insufficient parameter adaptability caused test failure), the reinforcement learning model will be triggered to iteratively update the optimal test parameters.

[0088] Parameter Iterative Optimization: Based on the "feedback learning logic" of reinforcement learning algorithms, the results of the current test failure (such as fault type, unmet target) are used as "penalty signals" to trace back the adaptation problem of the original optimal test parameters, and then make targeted dynamic adjustments to the parameters. For example, if the HIL test fails to meet the controller response delay target, the algorithm will optimize relevant parameters such as "data transmission baud rate" and "control command trigger threshold" to generate updated optimal test parameters.

[0089] Retest Trigger: After parameter optimization is completed, the system will restart the currently failed test phase based on the updated optimal test parameters to verify whether the parameter adjustment can solve the previous test failure problem, forming a micro-loop of "test-feedback-optimization-retest".

[0090] This application embodiment, when simulation testing, HIL testing, or real vehicle testing fails to meet the corresponding stage pass standards and does not exceed the maximum number of tests, dynamically iterates and adjusts the optimal test parameters based on reinforcement learning algorithms and restarts the test. This not only ensures that the closed-loop adaptive optimization mechanism runs through every key node of the entire testing process, effectively avoiding test interruptions or fault omissions caused by improper parameter configuration in a single instance, but also significantly improves the ability to accurately capture complex scenarios and potential faults through continuous parameter refinement. At the same time, no human intervention is required for decision-making or adjustment throughout the entire process, continuing the high efficiency advantage of fully unmanned operation. This ensures that the testing process can autonomously complete parameter optimization and retry within preset rules, significantly improving the test success rate and the validity of the results. It further improves the adaptive capability and fault tolerance of the vehicle-road-cloud integrated testing system, providing a more solid guarantee for the comprehensive verification of autonomous driving functions.

[0091] In one possible implementation, the method further includes: visually displaying the entire testing process based on a digital twin model.

[0092] In this embodiment, the digital twin model is not a simple virtual replica, but a dynamic mapping model that integrates vehicle physical characteristics, roadside environmental elements, cloud command flow logic, and operational rules for each testing phase. It synchronizes in real time the parameter optimization trajectory of reinforcement learning, the test execution progress of simulation / HIL / real vehicle, and the interaction data between the real vehicle and roadside equipment, forming a virtual mirror that is "synchronized within seconds" with the real testing process, ensuring the authenticity and real-time nature of the visualization.

[0093] Visualization refers to transforming abstract data (such as parameter adjustment values, test progress percentages, fault occurrence points, and data flow paths) throughout the entire testing process into intuitive graphical representations. These representations include real-time dynamic flowcharts, parameter change curves, 3D simulation interfaces of test scenarios, and module operation status indicators, allowing staff to quickly grasp the overall testing situation without analyzing raw data.

[0094] This application's embodiments visualize the entire testing process by introducing a digital twin model. This not only allows for a clear presentation of the progress status, dynamic parameter adjustment process, test result feedback, and data interaction links between the roadside, cloud, and vehicle at each stage of "reinforcement learning - simulation - HIL - OTA - real vehicle," but also facilitates real-time monitoring of the overall testing process by staff. Furthermore, it accurately recreates key scenarios and potential fault trajectories during the testing process. Combined with full test data and standardized reports, it further enhances the traceability of the testing process and the convenience of problem troubleshooting. Simultaneously, it makes the collaborative testing logic under the vehicle-road-cloud integrated architecture more transparent and controllable, further strengthening the standardization and rigor of the testing process. This provides more intuitive and comprehensive support for the efficient management, quality assessment, and subsequent optimization of autonomous driving function testing, and improves the full-process visualized management and control capabilities of the testing system.

[0095] In one possible implementation, the method further includes: collecting key operational data from each stage in real time; automatically generating a visual warning sign when data in a certain dimension exceeds a preset abnormal threshold; and performing target operations according to the warning level. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning model to adjust parameters.

[0096] In this application embodiment, key operational data refers to core data that can reflect the core working status of each module and device and the effectiveness of the test process. Specific dimensions include: operational status data of each test stage, parameter data (such as the current test parameter values ​​and the dynamic parameters of the autonomous driving model), interaction data (such as the data transmission delay of "vehicle-road-cloud" and data packet loss rate), test process data (such as the fault trigger frequency of simulation test and the hardware response time of HIL test), etc., which are the core basis for judging whether the system is operating normally.

[0097] The preset anomaly thresholds are "reasonable range boundaries" for key data in various dimensions, based on industry standards, system design indicators, and historical testing experience for autonomous driving testing. For example, the data transmission delay threshold is set to 50ms (exceeding this threshold is considered an anomaly), the reinforcement learning parameter optimization timeout threshold is set to 10s (exceeding this timeout is considered an anomaly), and the valid range of test parameters is set to [0.1, 1.0] (exceeding this range is considered an anomaly). These are the core reference standards for anomaly monitoring.

[0098] Visualized early warning labels are intuitive forms of anomaly alerts generated by digital twin models. They are presented in a visual display interface, for example, by marking abnormal modules / devices with flashing red icons, displaying abnormal data dimensions and exceeding values ​​in pop-ups, and using color gradients (green-yellow-red) to distinguish warning levels. This allows staff to quickly locate the location and severity of anomalies without analyzing raw data, thus achieving anomaly visualization.

[0099] The warning level is a hierarchical standard based on the severity and scope of the anomaly (such as minor anomaly, moderate anomaly, and severe anomaly). For example, "parameters slightly exceeding the effective range but not affecting test progress" is a minor anomaly, "data transmission delay exceeding the standard causing test data distortion" is a moderate anomaly, and "module operation crash or hardware failure risk" is a severe anomaly. It is the core basis for the test management module to execute differentiated responses.

[0100] This application embodiment establishes a multi-dimensional data monitoring mechanism by collecting key operational data from each stage of "reinforcement learning-simulation-HIL-OTA-real vehicle" in real time. When any dimension of data exceeds a preset anomaly threshold, an intuitive visual warning sign is automatically generated, and target operations such as test suspension, parameter reset, or triggering reinforcement learning model parameter adjustment are precisely executed according to the warning level. This not only achieves rapid perception and intelligent handling of test anomalies, effectively avoiding waste of test resources, extended test cycles, or distorted results caused by the expansion of anomalies, but also ensures the stable and orderly progress of the unmanned testing process. Furthermore, it strengthens the risk resistance capability and adaptive control level of the testing system, making test management under the vehicle-road-cloud integrated architecture more timely, accurate, and safe. It improves the closed-loop management of the entire chain from data monitoring and anomaly warning to handling response, providing a more solid guarantee for the high-quality advancement of autonomous driving function testing.

[0101] Based on the above method embodiments, another embodiment of this application provides an autonomous driving function testing device based on vehicle-road-cloud integration. This device can be applied to the cloud, such as... Figure 4 As shown, the device includes: The initialization unit 310 is used to complete the initialization by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests after receiving the test task; The adjustment unit 320 is used to dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters; The simulation test unit 330 is used to call the scenario library to perform simulation tests according to the optimal test parameters, the test scenario and the autonomous driving model; HIL test unit 340 is used to perform HIL testing by simulating the operation of vehicle hardware using the optimal test parameters and the autonomous driving model when the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria of the simulation test stage. The generation unit 350 is used to generate an OTA deployment package including an autonomous driving model when the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase. The transmission unit 360 is used to transmit the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real vehicle testing based on the OTA deployment package.

[0102] In one possible implementation, the device further includes: The receiving unit is used to receive at least one of the following: simulation test results, HIL test results, real vehicle test results, and test data reported by roadside equipment that interacts with the real test vehicle; The generation unit 350 is further configured to generate a standardized test report based on the received test results when the number of simulation tests equals the maximum number of tests, and / or when the number of HIL tests equals the maximum number of tests.

[0103] In one possible implementation, the adjustment unit 320 is further configured to dynamically adjust the optimal test parameters based on the reinforcement learning algorithm to obtain updated optimal test parameters when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation test stage, and / or when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the pass criteria for the HIL test stage, and / or when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test stage, so as to retest based on the updated optimal test parameters.

[0104] In one possible implementation, the device further includes: The display unit is used to visualize the entire testing process based on a digital twin model.

[0105] In one possible implementation, the device further includes: The data acquisition unit is used to collect key operational data from each stage in real time. The generation unit 350 is also used to automatically generate a visual warning sign when it detects that data in a certain dimension exceeds a preset abnormal threshold. The execution unit is used to perform target operations according to the warning level. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning module to adjust parameters.

[0106] The autonomous driving function testing device based on vehicle-road-cloud integration provided in this application embodiment first completes the initial configuration of initial test parameters, test scenarios, pass standards for each stage, and maximum number of tests. Then, it dynamically adjusts the initial test parameters based on reinforcement learning algorithms to obtain optimal test parameters. Subsequently, it sequentially executes simulation tests, HIL tests, and transmits OTA deployment packages to real test vehicles. This full-process design achieves seamless connection and unmanned advancement of each stage of "reinforcement learning - simulation - HIL - OTA - real vehicle". It not only eliminates the errors and efficiency bottlenecks caused by manual intervention and significantly reduces testing costs, but also forms a closed-loop adaptive adjustment of the entire process through reinforcement learning-driven parameter dynamic optimization. This effectively improves the coverage and relevance of test scenarios and accurately discovers potential faults. At the same time, relying on the vehicle-road-cloud integrated architecture, it opens up the data link between "roadside - cloud - vehicle" to ensure seamless interoperability and consistency of test data at each stage, comprehensively improving the efficiency, reliability, and comprehensiveness of autonomous driving function testing.

[0107] Based on the above method embodiments, another embodiment of this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method as described in any of the above embodiments.

[0108] Based on the above method embodiments, another embodiment of this application provides an electronic device or computer device, including: One or more processors; The processor is coupled to a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the electronic device or computer device performs the method as described in any of the above embodiments.

[0109] Based on the above embodiments, another embodiment of this application provides a computer program product, which includes instructions that, when executed on a computer or processor, cause the computer or processor to perform the method described in any of the above embodiments.

[0110] The above-described device and system embodiments correspond to the method embodiments and have the same technical effects. For detailed descriptions, please refer to the method embodiments. The device embodiments are derived from the method embodiments; detailed descriptions can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application.

[0111] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An autonomous driving function testing system based on vehicle-road-cloud integration, characterized in that, The system is applied in the cloud and includes: The test management module is used to initialize the system by configuring initial test parameters, test scenario, autonomous driving model, test pass standards for each stage, and maximum number of tests after receiving a test task, and to issue a test start command to the reinforcement learning module. The test start command includes the initial test parameters, the test scenario, the autonomous driving model, the test pass standards for each stage, and the maximum number of tests. The reinforcement learning module is used to obtain the initial test parameters in the test start command, and dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters. After replacing the initial test parameters in the test start command with the optimal test parameters, the updated test start command is sent to the simulation test module. The simulation test module is used to receive the updated test start command, and call the scenario library to execute the simulation test according to the updated test start command. When the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria of the simulation test stage, the optimal test parameters are synchronized to the preset storage space in the cloud. The Hardware-in-the-Loop (HIL) simulation test module is used to load the optimal test parameters synchronized in the cloud and the autonomous driving model configured by the test management module. It uses the optimal test parameters and the autonomous driving model to simulate the operation of the vehicle hardware to achieve HIL testing. If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL testing phase, it generates an OTA deployment package including the autonomous driving model and transmits the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real-vehicle testing based on the OTA deployment package.

2. The system according to claim 1, characterized in that, The test management module is also used to receive test results reported by at least one of the simulation test module, the HIL test module, the real test vehicle, and the roadside equipment that interacts with the real test vehicle. The simulation test module is also used to send a test termination command to the test management module when the number of simulation tests is equal to the maximum number of tests. The HIL test module is also used to send a test termination command to the test management module when the number of HIL tests is equal to the maximum number of tests. The test management module is also used to generate a standardized test report based on the received test results when the test termination instruction is received.

3. The system according to claim 1, characterized in that, The simulation testing module is also used to send a restart command to the reinforcement learning module when the number of simulation tests is less than the maximum number of tests and the simulation test results do not meet the pass criteria for the simulation testing phase. The HIL testing module is also used to send a restart command to the reinforcement learning module when the number of HIL tests is less than the maximum number of tests and the HIL test results do not meet the passing criteria for the HIL testing phase. The real test vehicle is also used to send a restart command to the reinforcement learning module when the number of real vehicle tests is less than the maximum number of tests and the real vehicle test results do not meet the pass criteria for the real vehicle test phase. The reinforcement learning module is further configured to, upon receiving the restart instruction, dynamically adjust the optimal test parameters based on the reinforcement learning algorithm to obtain updated optimal test parameters, so as to retest based on the updated optimal test parameters.

4. The system according to any one of claims 1-3, characterized in that, The system also includes: The digital twin module is used to visualize the entire testing process based on a digital twin model.

5. The system according to claim 4, characterized in that, The digital twin module is also used to: collect key operating data from the test management module, the reinforcement learning module, the simulation test module, the HIL test module, the real test vehicle, and the roadside equipment in real time; when a certain dimension of data is detected to exceed a preset abnormal threshold, a visual warning sign is automatically generated, and a warning signal is simultaneously triggered to the test management module. The test management module is also used to perform target operations based on the warning level represented by the warning signal when the warning signal is received. The target operations include pausing the test, resetting parameters, or triggering the reinforcement learning module to adjust parameters.

6. A method for testing autonomous driving functions based on vehicle-road-cloud integration, characterized in that, The method includes: Upon receiving the test task, initialization is completed by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage, and maximum number of tests. The initial test parameters are dynamically adjusted based on the reinforcement learning algorithm to obtain the optimal test parameters; Simulation tests are performed by calling the scenario library according to the optimal test parameters, the test scenario, and the autonomous driving model. If the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria for the simulation test stage, the vehicle hardware operation process is simulated using the optimal test parameters and the autonomous driving model to perform HIL tests. If the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase, an OTA deployment package including the autonomous driving model is generated, and the OTA deployment package is transmitted to the real test vehicle so that the real test vehicle can conduct real-vehicle testing based on the OTA deployment package.

7. The method according to claim 6, characterized in that, The method further includes: Receive at least one of the following: simulation test results, HIL test results, real vehicle test results, and test data reported by roadside equipment that interacts with the real test vehicle; If the number of simulation tests equals the maximum number of tests, and / or if the number of HIL tests equals the maximum number of tests, a standardized test report is generated based on the received test results.

8. A testing device for autonomous driving functions based on vehicle-road-cloud integration, characterized in that, The device includes: The initialization unit is used to complete the initialization by configuring the initial test parameters, test scenario, autonomous driving model, test pass standards for each stage and maximum number of tests after receiving the test task; The adjustment unit is used to dynamically adjust the initial test parameters based on the reinforcement learning algorithm to obtain the optimal test parameters; The simulation testing unit is used to execute simulation tests by calling the scenario library according to the optimal test parameters, the test scenario, and the autonomous driving model. The HIL test unit is used to perform HIL testing by simulating the operation of vehicle hardware using the optimal test parameters and the autonomous driving model, provided that the number of simulation tests is less than the maximum number of tests and the simulation test results meet the pass criteria for the simulation test phase. The generation unit is used to generate an OTA deployment package including an autonomous driving model when the number of HIL tests is less than the maximum number of tests and the HIL test results meet the pass criteria for the HIL test phase. The transmission unit is used to transmit the OTA deployment package to the real test vehicle so that the real test vehicle can conduct real vehicle testing based on the OTA deployment package.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 6-7.

10. An electronic device, characterized in that, The electronic device includes: One or more processors; The processor is coupled to a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the electronic device performs the method as described in any one of claims 6-7.

Images