An automatic driving debugging calibration simulation test method, device and storage medium

Abstract

The application discloses an automatic driving debugging calibration simulation test method and device and a storage medium. The method comprises the following steps: in response to an automatic driving test instruction initiated by a user, generating simulation perception data based on a virtual sensor model and a virtual chassis model driven by the automatic driving test instruction; performing a closed-loop simulation test on the simulation perception data by using a preset perception algorithm, a planning decision algorithm and a control algorithm; recording trajectory data, control data and perception output data in the closed-loop simulation test process, quantitatively evaluating the trajectory data, the control data and the perception output data to obtain quantitative evaluation results; calculating a multi-dimensional calibration score according to the quantitative evaluation results, and generating a multi-dimensional calibration effect report based on the multi-dimensional calibration score. The application realizes the technical effect of efficient, safe and low-cost virtual verification of calibration quality by mapping real calibration parameters to a simulation environment and performing closed-loop testing and quantitative scoring.

Description

Technical Field

[0001] This application relates to the field of autonomous driving testing, and in particular to an autonomous driving debugging, calibration, simulation testing method, equipment, and storage medium. Background Technology

[0002] The research and testing of autonomous driving systems heavily rely on the precise calibration of various sensors and chassis actuators. Traditional debugging and verification processes are primarily conducted in real-world vehicle environments, collecting data through road tests and manually analyzing the calibration results. This approach is not only time-consuming and costly, but also limited by site, weather, and traffic conditions, making it difficult to cover diverse extreme or dangerous scenarios and posing significant safety risks. Furthermore, due to the lack of standardized quantitative evaluation methods, the assessment of calibration quality often depends on engineers' experience, resulting in strong subjectivity and difficulty in ensuring consistency.

[0003] With the development of simulation technology, conducting partial testing in virtual environments has become an auxiliary means to improve efficiency. However, common simulation tests are mostly based on idealized general models, whose sensor and vehicle dynamics parameters differ significantly from the specific calibration conditions of the actual vehicle under test. This disconnect between model and real vehicle parameters means that simulation results cannot accurately reflect the real vehicle's performance under specific calibration conditions. The simulation environment becomes a functional verification tool, rather than an effective platform for evaluating and predicting calibration quality. Therefore, the advantages of simulation testing have not been fully utilized in the crucial calibration and debugging process.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide an autonomous driving debugging and calibration simulation test method, equipment and storage medium, which aims to solve the technical problem that existing simulation tests cannot effectively evaluate the calibration quality of real vehicles.

[0006] To achieve the above objectives, this application proposes an autonomous driving debugging, calibration, and simulation testing method, the method comprising: In response to the user-initiated autonomous driving test command, the virtual sensor model and virtual chassis model are driven to generate simulation perception data based on the autonomous driving test command; The preset perception algorithm, planning and decision-making algorithm, and control algorithm are used to perform closed-loop simulation tests on the simulated perception data. The trajectory data, control data, and sensing output data during the closed-loop simulation test are recorded, and the trajectory data, control data, and sensing output data are quantitatively evaluated to obtain the quantitative evaluation results. The multidimensional calibration score is calculated based on the quantitative evaluation results, and a multidimensional calibration effect report based on the multidimensional calibration score is generated. The multidimensional calibration score includes sensor calibration score, chassis calibration score and comprehensive performance score.

[0007] In one embodiment, the step of generating simulation perception data based on a user-initiated autonomous driving test command by driving a virtual sensor model and a virtual chassis model according to the autonomous driving test command includes: According to the autonomous driving test instruction, a preset test scenario is loaded, and the scenario parameters of the test scenario include at least one of road type, weather conditions, lighting conditions and traffic flow density. Based on the test scenario, the virtual sensor model is driven to generate initial simulation perception data, which includes at least one of point cloud data, image data, detection point data, inertial data, and positioning data. The virtual chassis model is driven to simulate vehicle dynamics response using the test scenario and the initial simulation perception data, thereby obtaining the simulation perception data.

[0008] In one embodiment, the step of performing closed-loop simulation testing on the simulated sensing data by executing preset perception algorithms, planning and decision-making algorithms, and control algorithms includes: The simulation perception data is processed by the perception algorithm to output environmental perception results, which include obstacle information, lane line information and traffic sign information. The planning and decision-making algorithm performs behavioral decisions based on the environmental perception results, the high-precision map information of the test scenario, and the preset driving task to generate a target trajectory. The target trajectory is calculated using the control algorithm to output longitudinal and lateral control commands, and the virtual chassis model is driven based on the longitudinal and lateral control commands to perform closed-loop testing.

[0009] In one embodiment, prior to the step of responding to a user-initiated autonomous driving test command, the method further includes: The calibration parameters of a real autonomous vehicle are obtained, and the calibration parameters are standardized to obtain simulation compatibility parameters. The simulation compatibility parameters are synchronously updated to the virtual sensor model parameters and virtual chassis model parameters of the simulation software.

[0010] In one embodiment, the step of synchronously updating the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software includes: Based on the sensor calibration parameters in the simulation compatibility parameters, update the configuration parameters of the virtual sensor model group. The virtual sensor model group includes at least a virtual lidar model, a virtual millimeter-wave radar model, a virtual camera model, a virtual inertial measurement unit model, and a virtual global positioning system model. The configuration parameters include the intrinsic parameter attributes, installation pose attributes, and transformation relationship parameters of each model relative to the vehicle coordinate system. Based on the chassis calibration parameters in the simulation compatibility parameters, update the configuration parameters of the virtual chassis model group. The virtual chassis model group includes at least a virtual vehicle dynamics model and a virtual actuator model. The configuration parameters include the geometric properties, mass properties, and tire properties of the virtual vehicle dynamics model, as well as the dynamic response characteristics and control logic parameters of the virtual actuator model.

[0011] In one embodiment, after the step of synchronizing the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software, the method further includes: After the parameters are synchronized and updated, a preset verification scenario is run in the simulation test environment to drive the updated virtual sensor model and virtual chassis model to perform parameter synchronization tests. Analyze the response data in the test results, and determine whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results; If the response data is determined to meet the expected test objective, the steps for responding to the user-initiated autonomous driving test command are executed.

[0012] In one embodiment, after analyzing the response data in the test results and determining whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results, the method further includes: If the response data is determined not to meet the expected test target, a parameter synchronization fine-tuning strategy is executed to make the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test target. The step of implementing the parameter synchronization fine-tuning strategy to ensure that the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives includes: Based on the comparison between the response data and the expected data, one or more simulation compatibility parameters that cause performance deviations are located in the comparison results. The simulation compatibility parameters include sensor calibration parameters and chassis calibration parameters. Within the preset parameter adjustment range, the one or more simulation compatibility parameters are iteratively adjusted, and the preset benchmark verification scenario is rerun after each iteration until the response parameters obtained after rerun meet the expected test target, at which point the iteration stops.

[0013] In one embodiment, the step of analyzing the response data in the test results and determining whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results includes: Based on the output of the virtual sensor model in the simulation test environment, the measured values ​​based on perception indicators are calculated. The perception indicators include at least one of target detection rate, positioning error, and lane line recognition accuracy. Based on the control response of the virtual chassis model in the simulation test environment, the measured values ​​of the control indicators are calculated, including at least one of trajectory tracking deviation, speed / acceleration response error, and steering angle tracking error. The measured values ​​are compared with the corresponding expected functional thresholds, and the results are used to determine whether the expected test targets are met. The expected functional thresholds include the expected thresholds for perception functions and the expected thresholds for control functions.

[0014] In addition, to achieve the above objectives, this application also proposes an autonomous driving debugging and calibration simulation test device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the autonomous driving debugging and calibration simulation test method described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the autonomous driving debugging, calibration, simulation, and testing method described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: The technical solution of this application, in response to a user-initiated autonomous driving test command, drives a virtual sensor model and a virtual chassis model to generate simulated perception data based on the autonomous driving test command; executes a preset perception algorithm, planning and decision-making algorithm, and control algorithm to perform closed-loop simulation testing on the simulated perception data; records trajectory data, control data, and perception output data during the closed-loop simulation test process; performs quantitative evaluation on the trajectory data, control data, and perception output data to obtain quantitative evaluation results; calculates a multi-dimensional calibration score based on the quantitative evaluation results; and generates a multi-dimensional calibration effect report based on the multi-dimensional calibration score, wherein the multi-dimensional calibration score includes a sensor calibration score, a chassis calibration score, and a comprehensive performance score.

[0017] This application achieves the technical effect of efficient, safe, and low-cost virtual verification of calibration quality by mapping real calibration parameters to a simulation environment and performing closed-loop testing and quantitative scoring. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of the autonomous driving debugging, calibration, and simulation testing method of this application; Figure 2 This is a detailed process diagram based on step S10 in the first embodiment; Figure 3 This is a detailed process diagram based on step S20 in the first embodiment; Figure 4 This is a flowchart illustrating the second embodiment of the autonomous driving debugging, calibration, and simulation testing method of the present invention. Figure 5 This is a detailed schematic diagram of step S60 in the first embodiment; Figure 6 This is a flowchart illustrating the third embodiment of the autonomous driving debugging, calibration, simulation, and testing method of the present invention. Figure 7 This is a detailed schematic diagram of step S80 in the first embodiment; Figure 8 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the autonomous driving debugging, calibration, simulation, and testing method in this application embodiment.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] In related technologies, the calibration and verification of autonomous driving systems mainly rely on real vehicle road tests, which are costly, inefficient, and risky. Simulation tests, on the other hand, use idealized general models and cannot accurately reflect the specific calibration state of real vehicles, making it difficult to effectively evaluate the calibration quality.

[0024] Based on the aforementioned deficiencies in related technologies, this application proposes an autonomous driving debugging, calibration, and simulation testing method. In this method, a virtual-real mapping simulation testing environment is established, and the sensor and chassis calibration parameters of the real vehicle are synchronized to the simulation software. This drives high-fidelity virtual sensors and chassis models to generate simulation data, executes closed-loop algorithm testing, and performs multi-dimensional quantitative evaluation. Finally, a calibration effect report based on sensor calibration scores, chassis calibration scores, and comprehensive performance scores is generated, achieving efficient, safe, and standardized evaluation of calibration quality.

[0025] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0026] Based on this, the embodiments of this application provide an autonomous driving debugging, calibration, and simulation testing method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the autonomous driving debugging, calibration, and simulation testing method of this application. In this embodiment, the autonomous driving debugging, calibration, and simulation testing method includes steps S10 to S40: Step S10: In response to the autonomous driving test command initiated by the user, drive the virtual sensor model and virtual chassis model to generate simulation perception data based on the autonomous driving test command; In this embodiment, when the system receives an autonomous driving test command initiated by the user, it drives a set of virtual models closely mapped to the calibration state of a real vehicle, thereby generating high-fidelity simulation perception data. This serves as the starting point for virtual-real combined testing. This data processing logic focuses on how to transform real-world calibration parameters into computable inputs in the virtual environment.

[0027] Users select or configure test tasks through the test management interface and trigger commands. Upon receiving the command, the system loads the corresponding test scenario and a pre-built virtual vehicle model. This virtual vehicle model comprises two core components: a series of virtual sensor models and a virtual chassis model. The source of the parameters for these models is crucial; they are not preset ideal values, but rather derived from a previously completed calibration process using real autonomous vehicles.

[0028] Specifically, the establishment and driving of virtual sensor models involves several key technologies. During the initialization phase, the system has established corresponding digital models based on the physical specifications of real sensors such as LiDAR, millimeter-wave radar, cameras, IMUs, and GPS. These models can simulate their measurement principles, noise characteristics, and physical limitations. When performing this step, the system does not use the default parameters of these models, but dynamically loads the specific parameters collected and synchronized during the real vehicle calibration phase. For example, it loads the intrinsic parameter matrix and distortion coefficients obtained through checkerboard calibration for the virtual camera; configures the calibrated intrinsic parameters such as beam angle and distance offset for the virtual LiDAR; and sets the precise installation position and attitude transformation matrix relative to the vehicle coordinate system for all virtual sensors, determined through extrinsic parameter calibration. When driving these models to generate data, the simulation's 3D scene engine calculates the raw data that each virtual sensor should "perceive" based on the vehicle's real-time state in the virtual world, such as image pixel matrices with specific lens distortion, point cloud coordinates containing simulated noise and calibration errors, and radar target lists simulating Doppler effects and RCS characteristics. These data together constitute the simulation perception data stream.

[0029] Similarly, the driving logic of the virtual chassis model follows a similar pattern to that of the virtual sensor model. Its dynamic parameters, such as vehicle mass, wheelbase, moment of inertia, and tire lateral stiffness, as well as actuator model parameters, such as the steering system's transmission ratio and delay, and the braking system's response curve, are all derived from the chassis calibration results of a real vehicle. During simulation testing, this model is responsible for providing a realistic controlled object model for subsequent control algorithms, and its dynamic response characteristics strive to be consistent with the performance of a real vehicle under the same calibration conditions.

[0030] Therefore, this data processing logic is a "parameter injection" and "real-time simulation" process based on real calibration parameters. This ensures that the input starting point for simulation testing, namely the simulated perception data, reproduces to the greatest extent the data characteristics that sensors in a real vehicle under specific calibration conditions should output, laying a reliable data foundation for the subsequent objective evaluation of the actual effect of this calibration scheme in a virtual environment.

[0031] Step S20: Execute the preset perception algorithm, planning and decision-making algorithm and control algorithm to perform closed-loop simulation test on the simulation perception data; This embodiment takes the high-fidelity simulation perception data generated in the above steps and, driven by this data, executes the complete autonomous driving algorithm stack, forming a software-in-the-loop closed-loop simulation test from perception to control. The core data processing logic of this step is to place the real algorithm code in a virtual but parameter-real simulation environment to run it, observe and record its overall behavior and performance, thereby evaluating the quality of the underlying calibration parameters supporting these algorithms.

[0032] The process begins with the execution of the perception algorithm. The system inputs the generated simulated images, point clouds, radar data, etc., into the perception module of the autonomous driving system under test. These modules include algorithms for object detection, lane recognition, multi-object tracking, sensor fusion, and localization. Because the input simulated data carries characteristics introduced by real-world calibration, such as actual camera distortion and LiDAR installation offset, the output of the perception algorithm, such as the bounding box, type, speed, and position of the target object, as well as the vehicle's own localization pose, will directly reflect the impact of the current calibration parameters on perception accuracy. If the calibration is poor, even if the algorithm itself is excellent, significant performance degradation may occur at this stage, such as localization drift and target association errors.

[0033] Next, the planning and decision-making algorithm operates based on the above perception results. The environment understanding module integrates perception information to construct an understanding of the virtual driving scenario. The behavior decision-making module then makes decisions such as following, changing lanes, and stopping based on the perception results, traffic rules, and preset driving tasks. The path planning and motion planning modules further transform the decisions into a specific, smooth, and executable motion trajectory. This process verifies the robustness and decision rationality of the upstream algorithm under non-ideal perception input. Perception noise or bias caused by calibration errors may force the planning module to make more conservative or less smooth plans.

[0034] Finally, the control algorithm receives the target trajectory or speed commands output by the planning layer. Control algorithms, such as lateral steering controllers and longitudinal speed and braking controllers, calculate the specific actuator control quantities. These control commands are not directly applied to the real vehicle but are output to the established, parameter-synchronized virtual chassis model. Based on its dynamic characteristics and actuator response characteristics, the virtual chassis model simulates and calculates the changes in the vehicle's motion state under the new control commands, including new position, attitude, and speed.

[0035] The aforementioned cycle, in which the "perception-planning-control" algorithm outputs control commands, drives the virtual vehicle model to change its state, and subsequently influences the generation of the next frame of simulation perception data, constitutes a complete closed loop. It is through executing this series of interconnected data processing and calculation processes that the complete behavioral chain that autonomous driving software might exhibit under real calibration conditions can be reproduced in the virtual environment. This allows the subtle influences of calibration parameters to be amplified and transmitted through the algorithm chain, ultimately manifesting in the vehicle's overall motion trajectory and control quality, providing a comprehensive data source for subsequent quantitative evaluation.

[0036] Step S30: Record the trajectory data, control data, and sensing output data during the closed-loop simulation test, and perform a quantitative evaluation on the trajectory data, control data, and sensing output data to obtain a quantitative evaluation result; This embodiment systematically records, extracts, and quantifies the massive amounts of process data generated during closed-loop simulation testing, transforming simulated phenomena such as vehicle behavior and intermediate algorithm outputs into measurable and comparable objective indicators. Its data processing logic involves extracting key performance evidence reflecting calibration quality from time-series simulation logs, performing structured calculations on this evidence, and generating preliminary evaluation results.

[0037] During closed-loop testing, the system activates a high-frequency, multi-channel data logger. The recorded data is primarily divided into three categories: trajectory data, control data, and perception output data. Trajectory data originates from the state output of the virtual chassis model, recording the simulated vehicle's high-precision pose, velocity, acceleration, and other kinematic information at each time point. Control data records the target command values ​​for throttle, brake, and steering angle calculated by the control algorithm, as well as the simulated values ​​of the actual actuator responses in the virtual chassis model. Perception output data records the deliverables of the perception algorithm at various times, such as a list of all detected objects and their attributes, the vehicle's own localization results, and the geometric description of the lane lines.

[0038] After recording is complete, the quantitative evaluation phase begins. This process involves in-depth benchmarking calculations and statistical analysis of the three types of data based on a predefined series of evaluation indicators. For example, in evaluating the perception output data, the system compares the "Ground Truth" in the simulation environment—the true state and attributes of all objects in the virtual world—frame-by-frame with the output of the perception algorithm. By calculating the recall and precision of target detection, measuring the estimation errors of target position and velocity, and analyzing the deviation and distribution of the localization results from the actual trajectory, the perception accuracy is quantified. These accuracy indicators are directly related to the quality of sensor intrinsic and extrinsic parameter calibration.

[0039] The evaluation of trajectory and control data focuses on the overall control performance of the vehicle. The system uses a preset reference trajectory or desired speed as a benchmark to calculate the statistical quantities of lateral and longitudinal deviations between the actual trajectory and the reference trajectory. Simultaneously, it analyzes the tracking error and response time of speed control, the steady-state error and overshoot of steering control, and the distance control accuracy during braking. These control performance indicators are profoundly affected by chassis calibration parameters; for example, the accuracy of steering delay compensation and the realism of the braking system response model will all be revealed in these quantitative results.

[0040] Data processing employs key techniques such as time alignment, data interpolation, filtering and smoothing, and extensive statistical calculations. Ultimately, it outputs a set of structured quantitative evaluation results, including the mean, variance, maximum value, and proportion of errors exceeding thresholds. These results are no longer raw operational logs, but refined data evidence directly addressing core questions such as "how accurate is the perception" and "how stable is the control," providing direct input for the next step of comprehensive scoring.

[0041] Step S40: Calculate the multidimensional calibration score based on the quantitative evaluation results, and generate a multidimensional calibration effect report based on the multidimensional calibration score. The multidimensional calibration score includes sensor calibration score, chassis calibration score, and comprehensive performance score.

[0042] This embodiment represents the summary and presentation stage of the testing process. Its core data processing logic involves weighting, aggregating, and interpreting the obtained quantitative evaluation results, transforming them into a multi-dimensional calibration scoring system, and automatically generating a structured calibration performance report. This step aims to transform complex test data into intuitive and decision-making conclusions, achieving standardized and quantitative evaluation of calibration quality.

[0043] First, a multi-dimensional calibration score is calculated based on the quantitative evaluation results. The scoring system is typically designed with a three-layer structure. The bottom layer consists of sensor calibration scores and chassis calibration scores. The sensor calibration score is not a single value, but rather a comprehensive evaluation of multiple sub-indicators strongly correlated with calibration in the perception accuracy assessment. For example, it is generated by weighted fusion of target detection accuracy scores, absolute and relative positioning error scores, and consistency scores between different sensors, ultimately producing one or more scores that represent the overall calibration quality of the sensor system. Similarly, the chassis calibration score is mainly derived from various results of the control performance evaluation, such as trajectory tracking accuracy scores, speed control stability scores, and steering and braking response accuracy scores, and is obtained after comprehensive calculation.

[0044] After obtaining the calibration scores for the sensor and chassis subsystems, a comprehensive performance score is further calculated. This score reflects the overall impact of the calibration work on the vehicle's autonomous driving capabilities. During the calculation, different weights are assigned to the sensor and chassis calibration scores according to project requirements, and then a weighted sum is performed. In addition, higher-level indicators reflecting the overall system performance may be directly incorporated, such as task completion success rate in typical scenarios and average driving comfort indicators. Ultimately, the comprehensive performance score provides an overall evaluation of the system's performance under the current calibration scheme.

[0045] Based on these scores, the system triggers an automatic report generation module. This module organizes and populates the report according to a predefined template, including a test overview, detailed quantitative evaluation results, calculated scores, and a level assessment based on the scores and thresholds. Key technical methods used in report generation include data visualization, such as automatically generating a score radar chart to visually display the strengths and weaknesses of each dimension; plotting performance trend charts to show the changes of the same indicator in different scenarios; and creating error distribution histograms for in-depth analysis of problem characteristics. The final part of the report presents conclusions and recommendations. Based on the score level and specific deductions, the system matches and outputs targeted improvement suggestions from a pre-defined knowledge base. For example, it might point out that inaccurate joint extrinsic parameter calibration of the LiDAR and camera caused the target fusion to jump, or suggest recalibrating the median and delay parameters of the steering system.

[0046] Therefore, this embodiment completes the final transformation from data to insight. By using an algorithmic scoring and report generation process, it replaces subjective judgments based on engineers' personal experience, ensuring that the effectiveness of each calibration task is presented and measured in a standardized, quantitative, and documented form. This greatly improves the consistency and efficiency of the evaluation process and provides a clear direction for subsequent optimization iterations.

[0047] Furthermore, you can also view Figure 2 , Figure 2 This is a detailed process diagram based on step S10 in the first embodiment. Figure 2 The step of generating simulation perception data by driving the virtual sensor model and virtual chassis model based on the user-initiated autonomous driving test command includes S11-13: Step S11: Load a preset test scenario according to the autonomous driving test instruction. The scenario parameters of the test scenario include at least one of road type, weather conditions, lighting conditions and traffic flow density. Step S12: Based on the test scenario, drive the virtual sensor model to generate initial simulation perception data, which includes at least one of point cloud data, image data, detection point data, inertial data, and positioning data. Step S13: Using the test scenario and the initial simulation perception data, drive the virtual chassis model to simulate vehicle dynamics response to obtain the simulation perception data.

[0048] This embodiment reveals the decomposed actions and data processing logic from receiving instructions to generating a complete simulation perception data chain. These three steps constitute an ordered process of scene initialization, sensor simulation, and chassis simulation. Its core lies in using a parameterized scene to drive a high-fidelity model and gradually synthesize closed-loop data for algorithm testing.

[0049] In the initialization phase of the test, the processing logic involves parsing and constructing a computable virtual driving environment based on user commands. When the system responds to an autonomous driving test command initiated by the user through the test interface, it first interprets the test intent implied in the command. The command typically includes the identifier or configuration parameters of the test scenario. Then, based on this information, the corresponding test scenario data is loaded from a pre-generated and well-managed scenario library. This scenario not only includes a static, high-precision 3D road model and semantic map, but more importantly, it contains a series of configurable dynamic and environmental parameters. These scenario parameters are key variables driving subsequent simulation operations. For example, road type determines the road's curvature, slope, and lane rules; weather conditions directly affect the simulation effect of the sensor model, such as the simulation of camera visibility and lidar penetration capabilities in rain and fog; lighting conditions control the position and intensity of the virtual sun, used to generate image shadows and exposure effects at different times; and traffic flow density parameters are used to dynamically generate and schedule surrounding vehicles, pedestrians, and other moving elements in the scenario, constructing a realistic traffic interaction environment. After loading, the simulation engine initializes the world state, preparing for the "placement" of the virtual vehicle and the "perception" of the sensors.

[0050] Within the loaded test scenario, its data processing logic simulates the process of a real sensor "observing" the current virtual environment and outputting raw data under specific calibration conditions. The system driver has already injected a set of virtual sensor models with real calibration parameters. Each model performs independent calculations based on its physical principles and current parameters. For example, the virtual camera model performs geometric transformations and degradation processing on the image rendered by the 3D scene engine based on its loaded intrinsic parameters and distortion coefficients, outputting image data that simulates real optical characteristics. The virtual LiDAR model emits simulated laser beams into the scene according to its calibrated beam layout and scanning mode, and calculates the distance and reflection intensity of the return point based on ray tracing and object material properties, forming point cloud data. The millimeter-wave radar model simulates the transmission and reception of electromagnetic waves, generating detection point data containing target distance, velocity, and angle through signal processing algorithms. Simultaneously, the IMU virtual model outputs acceleration and angular velocity inertial data with simulated zero bias and noise based on the motion state of the virtual vehicle; the GPS virtual model, combined with the satellite visibility and signal interference conditions set in the scene, outputs positioning data with corresponding errors. The generation of the initial simulation perception data strictly follows the time synchronization mechanism to ensure that the data streams of each sensor are aligned on the timestamp, simulating the hardware synchronization effect in the real system.

[0051] The data processing logic connecting perception and vehicle dynamics integrates scene information and initial perception data, calculating the complete state response of the vehicle system through a high-fidelity dynamics model, thereby enriching and closing the simulation perception data chain. The multi-source initial perception data and the physical environment defined by the test scenario are used as inputs. A virtual chassis model is activated and driven here. This model already includes mass, dimensions, tire, suspension, and actuator characteristic parameters synchronized from real vehicle calibration. The system first calculates the contact mechanics between the tires and the ground based on scene information, such as road slope, friction coefficient, and the vehicle's current position. Simultaneously, the model "receives" preliminary control commands from upstream algorithms or test scripts. More importantly, the model's own state changes form an internal feedback loop. For example, vehicle acceleration leads to pitch changes, and steering causes roll and yaw movements. These subtle changes in vehicle posture, velocity, and acceleration, calculated in real-time by the dynamics model, are immediately fed back into the entire simulation loop. These updated vehicle states, along with the continuously generated initial simulation perception data, constitute the defined, final output of the complete simulation perception data to the subsequent autonomous driving algorithm stack. Therefore, this step ensures that the vehicle's movement in the virtual environment conforms to the laws of physics and that its dynamic characteristics are consistent with those of the real calibrated vehicle, so that the control commands of subsequent algorithms can be verified on a real "controlled object" model.

[0052] Furthermore, you can also view Figure 3 , Figure 3 This is a detailed process diagram based on step S20 in the first embodiment. Figure 3 The step of performing closed-loop simulation testing on the simulation perception data by executing preset perception algorithms, planning and decision-making algorithms, and control algorithms includes S21~23: Step S21: The simulation perception data is processed by the perception algorithm to output the environmental perception result, which includes obstacle information, lane line information and traffic sign information. Step S22: The planning and decision-making algorithm is used to make behavioral decisions based on the environmental perception results, the high-precision map information of the test scenario, and the preset driving task to generate a target trajectory. Step S23: Calculate the target trajectory using the control algorithm to output longitudinal control commands and lateral control commands, and drive the virtual chassis model based on the longitudinal control commands and lateral control commands to perform closed-loop testing.

[0053] This embodiment further breaks down the closed-loop simulation test execution process, specifically describing how the autonomous driving algorithm stack processes and responds sequentially through three core stages—perception, planning and decision-making, and control—after receiving simulation perception data, ultimately forming instructions to drive the virtual vehicle and completing the closed loop. These three steps constitute the core computational chain of software-in-the-loop testing, and its data processing logic reflects the complete transformation path from raw data to control behavior.

[0054] The generated simulated perception data is analyzed and understood, transforming it into structured environmental cognition. The data processing logic involves inputting heterogeneous data streams, such as point clouds, images, and millimeter-wave radar detection points, from simulated real sensors into the corresponding perception algorithm modules. These modules perform a series of calculations, including target detection, segmentation, tracking, and localization, ultimately outputting a unified, machine-understandable environmental perception result.

[0055] In practice, various perception algorithm modules are launched in parallel or sequentially to process their corresponding data streams. For example, the object detection algorithm processes image data generated by the camera's virtual model, identifying the bounding boxes and categories of obstacles such as vehicles, pedestrians, and cyclists in the image through models such as convolutional neural networks. It also processes LiDAR point cloud data, obtaining the location and geometric dimensions of obstacles in 3D space through point cloud segmentation and clustering algorithms. The lane detection algorithm focuses on extracting the geometric features of lane lines from images or point clouds, outputting the type, curvature, and position information of the lane lines relative to the vehicle. The traffic sign recognition algorithm analyzes images, identifying speed limit, stop, and turn signs and interpreting their meanings. Simultaneously, the localization algorithm integrates the output of the GPS virtual model, IMU inertial data, and the matching results of LiDAR point clouds and high-precision maps to estimate the vehicle's precise pose and speed in the global coordinate system in real time. The outputs of these different algorithms do not exist in isolation. Sensor fusion algorithms serve as a crucial component, using synchronized spatiotemporal calibration parameters from each sensor to correlate, match, and estimate the detection results from cameras, LiDAR, and millimeter-wave radar within a unified spatiotemporal framework. This eliminates false detections or missed detections from a single sensor, resulting in a more stable and accurate obstacle tracking list. Ultimately, the output environmental perception result is a comprehensive data structure containing a dynamic attribute list of all tracked obstacles at the current moment, a structured description of lane lines, semantic information of traffic signs, and the vehicle's high-precision positioning status, providing reliable environmental input for subsequent decision-making and planning.

[0056] Based on the understanding of environmental perception results, combined with global task and map information, intelligent driving decisions are made and an executable trajectory is planned. Its data processing logic lies in combining environmental cognition with driving objectives, and through multi-level planning algorithms, generating a future movement path that not only complies with traffic rules and safety requirements, but also efficiently completes the driving task.

[0057] The implementation of this step begins at the decision-making level. The behavior decision-making module comprehensively analyzes the environmental perception results, the high-precision map semantic information attached when the test scenario is loaded, and the driving task preset for this test. The map information provides prior knowledge such as lane connection relationships, intersection structure, and traffic rules. The decision-making module assesses the current traffic situation, such as judging the intention of the vehicles ahead, evaluating the safety and necessity of changing lanes, and deciding whether to go straight or turn at an intersection. This decision-making process is based on rule-based state machines, probabilistic models, or more complex reinforcement learning models. After determining the high-level behavior, the motion planning module performs global path planning, planning a rough reference path from the starting point to the destination based on the high-precision map. Then, within a local range, it combines the real-time perceived dynamic obstacle trajectories, the vehicle's state, and decision commands to generate a detailed trajectory. This process solves an optimization problem that satisfies vehicle dynamics constraints, avoids collisions, and ensures ride comfort, ultimately outputting a smooth sequence of trajectory points covering the next few seconds, i.e., the target trajectory. This target trajectory accurately describes the position, heading angle, speed, and even acceleration that the vehicle expects to reach at each future moment, serving as a bridge connecting the decision intention and the underlying control.

[0058] The target trajectory given by the planning layer is transformed into specific control commands that can be understood by the vehicle's actuators, driving the virtual chassis model to complete the action execution, thereby closing the simulation test loop. Its data processing logic lies in decomposing the desired trajectory tracking problem into longitudinal speed / distance control and lateral direction control through control algorithms, and calculating the physical quantities required by the underlying actuators.

[0059] During implementation, the control algorithm receives the generated target trajectory as input. This algorithm typically consists of a lateral controller and a longitudinal controller. The lateral controller is responsible for controlling the vehicle's steering; its core function is to calculate the steering angle required to minimize the lateral deviation between the vehicle's actual trajectory and the target trajectory. It employs a pure tracking algorithm based on a geometric model or a linear quadratic regulator based on a dynamic model to calculate the target front wheel angle in real time. The longitudinal controller controls the vehicle's speed by adjusting the drive torque and braking force. Based on the desired speed curve in the target trajectory and the vehicle's actual speed, it calculates the target acceleration and then outputs specific throttle opening or braking pressure commands through the throttle and brake mapping relationship. These calculated longitudinal and lateral control commands together constitute the drive commands for the virtual chassis model. The virtual chassis model, with its dynamics and actuator models infused with real calibration parameters, calculates the precise changes in the vehicle's motion state under the new commands, including updates to position, attitude, and speed, based on the simulated physical laws and actuator response characteristics within its internal model. The updated vehicle status is immediately fed back into the world state of the simulation environment, which in turn affects the scene "seen" by the virtual sensor model in the next simulation cycle, generating new simulation perception data. This cycle repeats continuously, forming a complete closed-loop test from perception to control to environmental feedback. Through this closed-loop operation, the adaptability and control accuracy of the control algorithm on a specific calibrated chassis model are fully tested.

[0060] Furthermore, you can also view Figure 4 , Figure 4 This is a flowchart illustrating the second embodiment of the autonomous driving debugging, calibration, and simulation testing method of the present invention, based on the shown... Figure 4 Prior to the step of responding to the user-initiated autonomous driving test command, steps S50-60 are also included: Step S50: Obtain the calibration parameters of the real autonomous vehicle, and obtain the simulation compatibility parameters after standardizing the calibration parameters; Step S60: Synchronously update the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software.

[0061] This embodiment is a crucial preparatory step for implementing the entire testing method, establishing a precise parameter bridge between the real physical world and the virtual simulation environment. These two steps focus on data acquisition, transformation, and synchronization. The core data processing logic involves taking the specific parameter set obtained after debugging and calibration of the real vehicle, performing a series of normalization and mapping processes, and finally injecting it into the corresponding model in the simulation software without loss and with accuracy. This ensures that the starting point for subsequent simulation testing is established on the actual calibration state.

[0062] At the starting point of the data processing flow, first-hand calibration data from real vehicles is acquired and preprocessed. This step begins with the data acquisition process, obtaining raw calibration parameters by connecting to the vehicle's calibration software, in-vehicle network, or electronic control unit. These parameters are heterogeneous and dispersed, mainly including two categories: sensor calibration parameters and chassis calibration parameters. Sensor calibration parameters further include intrinsic and extrinsic parameters; intrinsic parameters include the focal length, principal point coordinates, and distortion coefficient matrix of the camera, the distance and angle calibration tables of each channel within the LiDAR, and the zero bias and scaling factor of the IMU; extrinsic parameters describe the rigid transformation relationship between various sensors, i.e., the rotation matrix and translation vector from the sensor coordinate system to the vehicle coordinate system. Chassis calibration parameters are derived from the debugging results of the drive-by-wire system, such as the angle transfer function and delay time constant of the steering system, the pressure-torque response curve of the braking system, the torque response characteristics of the drive system, and key physical quantities involved in the vehicle dynamics model, such as mass, wheelbase, and tire lateral stiffness.

[0063] After obtaining these raw parameters, they are standardized. This data processing logic requires transforming data from diverse sources, with varying formats and potentially inconsistent units, into a set of internally consistent "simulation-compatible parameters" that the simulation software can directly recognize and load. This process first unifies the format, for example, by standardizing all extrinsic parameter transformation matrices to a 4x4 homogeneous coordinate transformation format and standardizing angle units to radians. Secondly, coordinate system unification is performed, ensuring that all spatial parameters are based on a predefined vehicle coordinate system aligned with the global coordinate system of the simulation environment. Next, unit unification is performed, converting all physical quantities to the SI or conventional units specified by the simulation engine. Furthermore, data integrity and validity are verified, checking for missing necessary parameters and ensuring that values ​​are within reasonable physical ranges. The final output simulation-compatible parameters are a structured dataset or configuration file that fully preserves all information from the real vehicle calibration while meeting the input interface specifications of the simulation model.

[0064] The processed simulation compatibility parameters are reliably and accurately synchronized and updated to the corresponding model in the simulation software, completing the final alignment of the virtual and real states. This step involves a process of "parameter injection" and "model refresh."

[0065] In practice, the simulation-compatible parameter package is transmitted to the simulation environment via the configuration interface or data communication link provided by the simulation software. For virtual sensor models, the sensor list in the parameter package is traversed, and the intrinsic and extrinsic parameters of each sensor are loaded into the corresponding virtual model instances such as LiDAR, camera, and millimeter-wave radar. For example, the intrinsic parameter matrix calibrated by the camera is assigned to the projection model of the virtual camera renderer, and the extrinsic parameter transformation matrix is ​​set to the installation posture of the virtual sensor on the simulated vehicle 3D model. For virtual chassis models, the system updates the properties of the multibody dynamics model with the calibrated dynamic parameters and replaces the default response curves of the original actuator models with actuator response characteristic parameters. During parameter loading, conflict detection is performed, such as checking whether the new parameters conflict with the original constraints of the model, and performing necessary model re-initialization operations to ensure that the new parameters are fully effective.

[0066] After synchronization and updating, a verification phase is conducted. The updated virtual vehicle model is run in a simple benchmark scenario, and its behavioral output is quickly compared and analyzed with the baseline data recorded by a real vehicle under the same calibration state and in a similar scenario. This aims to verify the overall effectiveness of parameter synchronization from a macroscopic perspective, ensuring that the virtual model has sufficient consistency with the real vehicle in key response characteristics, such as steering transient response and braking deceleration build-up. Only when the synchronization verification is successful, confirming that the simulation environment has successfully reproduced the specific calibration state of the real vehicle, is the entire system considered ready to accept and execute subsequent user test commands.

[0067] Therefore, the above processing steps together constitute an automated pipeline from "real vehicle data" to "simulation image". Through standardized data processing and a reliable synchronization mechanism, the fundamental problem of the disconnect between model and real vehicle parameters is solved, making all subsequent simulation-based tests, evaluations, and scores reflect the actual calibration quality. This is the technical foundation for the establishment and effectiveness of the entire method.

[0068] Furthermore, you can also view Figure 5 , Figure 5 This is a detailed process diagram based on step S60 in the first embodiment. Figure 5 The step of synchronously updating the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software includes S61~62: Step S61: Update the configuration parameters of the virtual sensor model group according to the sensor calibration parameters in the simulation compatibility parameters. The virtual sensor model group includes at least a virtual lidar model, a virtual millimeter-wave radar model, a virtual camera model, a virtual inertial measurement unit model, and a virtual global positioning system model. The configuration parameters include the intrinsic parameter attributes, installation pose attributes, and transformation relationship parameters of each model relative to the vehicle coordinate system. Step S62: Update the configuration parameters of the virtual chassis model group according to the chassis calibration parameters in the simulation compatibility parameters. The virtual chassis model group includes at least a virtual vehicle dynamics model and a virtual actuator model. The configuration parameters include the geometric properties, mass properties and tire properties of the virtual vehicle dynamics model, as well as the dynamic response characteristics and control logic parameters of the virtual actuator model.

[0069] This embodiment, based on the specific details of the parameter synchronization update process, defines how to accurately configure standardized real vehicle calibration data into the corresponding virtual model set in the simulation environment, specifically for the sensor system and chassis system. The core data processing logic of these two steps lies in "targeted distribution" and "model reconstruction," that is, accurately matching and writing parameters according to their types into the attribute configuration of specific virtual models, thereby constructing a digital surrogate that corresponds one-to-one with the real vehicle calibration state on the simulation side.

[0070] When constructing a virtual sensor array that fully maps to the real vehicle's perception system, the data processing logic involves parsing the structured sensor calibration parameters in the simulation-compatible parameter package and, based on this, updating the internal configuration of each member in the virtual sensor model group one by one. The virtual sensor model group is a collection covering the main sensor types for autonomous driving, typically including at least a virtual LiDAR model, a virtual millimeter-wave radar model, a virtual camera model, a virtual inertial measurement unit model, and a virtual global positioning system model.

[0071] In practice, the system iterates through the parameter list and performs the following update operations for each sensor identifier. First, the intrinsic parameters of each model are updated. For example, for a virtual camera model, its calibrated intrinsic parameter matrix and distortion coefficients are loaded; these parameters directly determine the projection geometry and optical distortion effects when the virtual camera renders images. For a virtual LiDAR model, its internal calibration table is loaded, precisely setting the vertical angle, horizontal angular resolution, and zero-point offset for each scan line. For a virtual IMU model, parameters such as the zero bias and scaling factor of its accelerometer and gyroscope are loaded. Second, the mounting pose attributes are updated. Based on the calibrated extrinsic parameter data—that is, the transformation parameters from the sensor coordinate system to the vehicle coordinate system—the system sets precise mounting position coordinates (x, y, z) and mounting angles (pitch, roll, yaw) for each virtual sensor model on the simulated vehicle's 3D model. This step is crucial, ensuring that in the simulation, the LiDAR's scanning cone angle, the camera's field of view, and the millimeter-wave radar's beam pointing are completely consistent with the physical mounting on the real vehicle. Finally, based on the aforementioned intrinsic parameters and installation pose, the system will uniformly calculate and confirm the final transformation relationship matrix of all sensors relative to a uniformly defined vehicle coordinate system within the simulation engine, ensuring that multi-sensor data fusion has a correct spatiotemporal alignment basis.

[0072] In constructing a virtual vehicle dynamic model that matches the dynamics and control system of a real vehicle chassis, the data processing logic involves injecting the chassis calibration parameters from the simulation-compatible parameter package into each sub-model of the virtual chassis model group to reproduce the handling and response characteristics of a real vehicle. The virtual chassis model group mainly consists of two core parts: a virtual vehicle dynamics model and a virtual actuator model.

[0073] In practice, updating the configuration parameters of the virtual vehicle dynamics model involves multiple physical levels. Geometric attributes such as wheelbase, track width, front and rear overhang lengths, and center of gravity height are directly updated, as these are fundamental to determining the vehicle's steering geometry and stability. Mass attributes such as vehicle mass, sprung mass, and moments of inertia about each axle are also updated, directly affecting the vehicle's acceleration, braking, and cornering inertia. Tire attributes are crucial for dynamics simulation, and parameters requiring updating include the tire's rolling radius, longitudinal slip stiffness, and lateral stiffness characteristic curves, which determine the force transmission relationship between the tire and the virtual road surface. For the virtual actuator model, the updates involve its dynamic response characteristics and control logic parameters. For example, the steering system model will update its transmission ratio, mechanical backlash, and response delay model from control commands to wheel angle; the braking system model will update its hydraulic or electric brake pressure build-up characteristic curves, braking force distribution logic, and response delay; and the drive system model will update the torque response spectrum and external characteristic parameters of the motor or engine. All these parameter updates are designed to ensure that the dynamic responses of the virtual chassis model, such as longitudinal acceleration, lateral acceleration, and yaw rate, when faced with the same control commands, are as close as possible to the performance of a real calibrated vehicle in the physical world.

[0074] After the update, the virtual vehicle within the simulation software transforms from a general model into a "digital twin" carrying specific calibration parameters of a real vehicle. This provides an accurate and reliable model foundation for subsequent closed-loop simulation tests based on real calibration conditions. This process emphasizes the completeness and accuracy of parameter updates and is one of the core elements ensuring the effectiveness of the entire simulation testing method.

[0075] Furthermore, you can also view Figure 6 , Figure 6 This is a flowchart illustrating the third embodiment of the autonomous driving debugging, calibration, and simulation testing method of the present invention, based on the shown... Figure 6 After the step of synchronizing and updating the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software, the method further includes steps S70~S90: Step S70: After confirming that the parameter synchronization update is completed, run the preset verification scenario in the simulation test environment to drive the updated virtual sensor model and virtual chassis model to perform parameter synchronization test. Step S80: Analyze the response data in the test results, and determine whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results; Step S90: If it is determined that the response data meets the expected test target, execute the step of responding to the autonomous driving test command initiated by the user.

[0076] Step S100: If it is determined that the response data does not meet the expected test target, execute the parameter synchronization fine-tuning strategy to make the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test target; This embodiment constitutes the key quality verification and closed-loop control links after parameter synchronization updates. The data processing logic of this series of steps is as follows: after injecting the real calibration parameters into the simulation model, a preset, standardized verification process is used to actively verify the accuracy of the virtual-real mapping and the effectiveness of the model update. Only when the behavior of the virtual model is confirmed to be consistent with the expected goals is the entire system considered reliable and enters the formal testing phase; if it is not consistent, an automated fine-tuning mechanism is triggered until the requirements are met, thereby ensuring that the benchmark for all subsequent evaluations is correct and stable.

[0077] The core of the verification process execution phase is to drive the updated model through a "trial run" in a controlled simulation environment. The data processing logic for this step involves designing and executing a set of verification scenarios that comprehensively stimulate the key characteristics of the model, and systematically collecting all response data during the operation. Once the system confirms that the parameter synchronization update operation is complete, it automatically loads one or more preset verification scenarios. These verification scenarios are carefully designed and typically include several parts: first, a baseline static scenario, such as placing the vehicle in a virtual reproduction of a clearly defined calibration site, used to verify the sensor installation extrinsic parameters; second, simple dynamic scenarios, such as the vehicle performing uniform straight-line driving, fixed-radius circular driving, or simple acceleration and braking processes, used to stimulate the basic dynamic response of the chassis. The system drives the updated virtual sensor model and virtual chassis model to run in this scenario. During the operation, not only is the vehicle's final trajectory recorded, but multiple data streams are also collected and time-aligned simultaneously, including: the raw simulation data output by each virtual sensor (such as images and point clouds), key intermediate states within the sensor model, all kinematic states calculated by the virtual chassis model (position, velocity, acceleration, attitude angles, etc.), and the commands and response values ​​of the actuator model. These data constitute the complete response data packet used for subsequent analysis.

[0078] In the analysis and decision-making phase of the verification process, the data processing logic involves conducting a multi-dimensional and quantitative benchmarking analysis of the collected response data against a set of "expected test targets," and making a pass / fail judgment based on the analysis results. The expected test targets are not subjective feelings, but rather derived from two objective bases: first, standard performance data (i.e., the "Ground Truth" baseline) collected and stored during the real vehicle calibration process in similar benchmark scenarios; and second, ideal response values ​​calculated based on theoretical models and calibration parameters. The analysis process employs automated algorithms. For example, at the perception level, the system matches the images generated by the virtual camera in a static scene with theoretical projections or real vehicle reference images, calculates reprojection errors, and verifies the accuracy of the camera's internal and external parameter synchronization; it also compares the point cloud scanned by the virtual LiDAR on a known-size calibration object with the standard model to analyze the point cloud shape and distance deviations. At the control level, the system analyzes the deviation between the actual trajectory radius of the vehicle during uniform circular motion and the expected radius calculated by the theoretical steering model, evaluating the synchronization effect of the steering system parameters. It also analyzes the rise time, steady-state error, and overshoot of the vehicle speed response under step acceleration or braking commands, comparing these with real vehicle test data or theoretically expected curves. The system comprehensively checks whether the deviations of various indicators are within preset tolerance thresholds, and ultimately determines whether the basic perception and control functions of the virtual model meet the expected test objectives through a decision logic (if all key indicators pass, the entire system passes).

[0079] The branch paths executed based on the judgment results constitute a decision-making closed loop. This loop includes a positive path and a negative feedback and correction path. In the positive path, if the system determines that the response data meets the expected test objectives, it means that parameter synchronization is successful and the virtual-real mapping relationship is accurately established. At this point, the system generates a verification-passed status flag and officially enters a standby state, ready to receive and execute various autonomous driving test commands subsequently initiated by the user. This marks the successful completion of the preliminary preparation phase.

[0080] In the negative feedback and correction path, if the response data is determined not to meet the expected test target, it indicates a deviation in parameter synchronization or incomplete model mapping. In this case, the system will not report an error or terminate; instead, it will automatically execute a parameter synchronization fine-tuning strategy. The data processing logic of this strategy is based on the results of deviation analysis, performing targeted, iterative parameter optimization. For example, if the perception data analysis shows a systematic shift in the lidar point cloud, the fine-tuning strategy may add a small correction to the already synchronized extrinsic parameter transformation matrix and re-drive the verification scenario test. If the dynamic response differs from expectations, it may make small adjustments to the tire side stiffness or actuator delay parameters. This fine-tuning process is automated and iterative. After adjustment, the system reruns the simplified verification scenario, performs analysis and judgment again, forming an inner loop of "verification-analysis-fine-tuning-re-verification" until the response data reaches the expected target. In this way, the system possesses a certain degree of self-calibration capability, overcoming minor inconsistencies caused by calibration data measurement noise, model simplification errors, or transmission conversion processes. Ultimately, it ensures that the simulation model can serve as a high-fidelity digital substitute for the actual calibration state, providing a solid and reliable foundation for subsequent evaluation.

[0081] The step of implementing the parameter synchronization fine-tuning strategy to ensure that the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives includes: Based on the comparison between the response data and the expected data, one or more simulation compatibility parameters that cause performance deviations are located in the comparison results. The simulation compatibility parameters include sensor calibration parameters and chassis calibration parameters. Within the preset parameter adjustment range, the one or more simulation compatibility parameters are iteratively adjusted, and the preset benchmark verification scenario is rerun after each iteration until the response parameters obtained after rerun meet the expected test target, at which point the iteration stops.

[0082] The core data processing logic of the above-described parameter synchronization fine-tuning strategy lies in constructing a closed-loop optimization process with feedback and self-correction capabilities. When the model response after initial synchronization deviates from the expected target, the system does not rely on manual intervention but can automatically diagnose the source of the deviation and perform fine-tuning and iterative adjustments to the corresponding simulation compatibility parameters until the performance of the virtual model is calibrated to within an acceptable tolerance range. This process ensures the final accuracy of the virtual-real mapping and is a key link in improving the robustness and automation level of the entire simulation testing method.

[0083] The fine-tuning strategy begins with precise attribution of deviations. The system first performs deeper data mining based on comparative analysis results to pinpoint the specific parameters causing performance deviations. Its processing logic involves tracing the observed macroscopic performance deviations back to errors in one or more simulation-compatible parameters. The system invokes a built-in deviation analysis algorithm, which quantitatively compares the specific response data recorded during testing with the expected data item by item. For example, if the feature point reprojection error of the virtual camera image exhibits a specific spatial distribution pattern in a static calibration board verification scenario, the algorithm will map it to a potential deviation in specific elements of the camera intrinsic parameter matrix (such as focal length or principal point coordinates), or inaccuracy in the lens distortion coefficient. If the vehicle exhibits a stable steering radius error in a uniform circular test, the algorithm will analyze whether this is mainly due to inaccurate steering system transmission ratio parameters or deviations in the setting of basic geometric parameters such as the vehicle's wheelbase. By analyzing the correlation between various error modes such as lateral deviation, longitudinal deviation, and angular deviation and the physical meaning of the parameters, the system can identify one or a few key parameters that are most likely to need adjustment from a large set of simulation-compatible parameters. These parameters may belong to sensor calibration parameters or chassis calibration parameters.

[0084] After successfully locating the target parameters, the system enters the iterative adjustment phase. The processing logic in this phase involves applying a small correction to the target parameters within a preset, safe parameter adjustment range, then re-driving the model for verification testing, and determining the next action based on the new results, forming an automated "adjustment-verification-evaluation" cycle. The preset parameter adjustment range is crucial; it is typically set based on prior engineering knowledge, such as the possible measurement error range of the calibration parameters or the model's sensitivity analysis results, to prevent the adjustment process from becoming ill-conditioned or producing unreasonable parameter values.

[0085] In practice, the system first calculates an initial correction based on the magnitude and direction of the initial deviation using a predefined adjustment strategy. For example, for extrinsic translation error, proportional adjustment might be used; for response delay, linear compensation might be applied. Next, the system automatically updates the corresponding model with the adjusted parameters in the simulation environment and reruns the preset benchmark verification scenario used in the previous steps. After the run, response data is collected again, and its compliance with the expected target is calculated. The system compares this result with the result of the previous iteration: if the compliance has significantly improved but not yet fully met the target, fine-tuning continues in the same direction with possible adjustment step sizes; if the compliance has decreased or the change is not significant, the optimization algorithm may be adjusted, trying different correction directions or strategies. This process continues to loop until the response data generated by the latest iteration is determined to meet the expected test target. Once the target is met, the iteration loop immediately stops, the system records the final adjusted parameter set, and uses it as the benchmark model configuration for all subsequent formal simulation tests. Through this automated iterative fine-tuning based on data and models, the system can effectively compensate for minor systematic errors that are unavoidable during parameter synchronization, ultimately ensuring that the virtual digital twin is highly faithful to its corresponding real physical entity in terms of key functional characteristics.

[0086] Furthermore, you can also view Figure 7 , Figure 7 This is a detailed process diagram based on step S80 in the first embodiment. Figure 7 The step of analyzing the response data in the test results and determining whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results includes S81~83: Step S81: Based on the output of the virtual sensor model in the simulation test environment, calculate the measured value based on the perception index, which includes at least one of the following: target detection rate, positioning error, and lane line recognition accuracy. Step S82: Based on the control response of the virtual chassis model in the simulation test environment, calculate the measured values ​​based on the control indicators, wherein the control indicators include at least one of trajectory tracking deviation, speed / acceleration response error, and steering angle tracking error; Step S83: Compare the measured value with the corresponding expected functional threshold, and determine whether it meets the expected test target based on the comparison result. The expected functional threshold includes the expected threshold for perception function and the expected threshold for control function.

[0087] This embodiment refines the performance compliance assessment process, providing a specific quantitative path from raw response data to a clear conclusion of pass or fail. Its data processing logic involves extracting the complex multimodal data stream generated by simulation into a few key, measurable performance indicators, and comparing the measured values ​​of these indicators with pre-set objective thresholds, thereby replacing subjective judgment and achieving standardized and automated acceptance decisions.

[0088] The core of quantitative evaluation of data output from virtual sensor models is calculating measured values ​​of indicators that reflect the fundamental capabilities of the perception system. During the verification scenario, the virtual sensor model continuously outputs simulation data streams, such as image sequences, point cloud frames, and location messages. The system synchronously records these data along with predefined or real-time generated "real-world" data within the verification scenario. Based on this paired data, the system automatically performs the following calculations: For target detection rate, the algorithm matches the list of obstacles detected and output by the virtual camera or LiDAR with the actual ground truth of obstacles in the scene in each frame, counting the number of correct detections, missed detections, and false detections, and then calculating specific values ​​such as recall and precision as measured values. For positioning error, the system aligns the output of the positioning module provided by the virtual sensor model (such as GPS / IMU fusion pose or SLAM results) with the high-precision ground truth trajectory of the vehicle in the simulation environment, calculating the position deviation and heading angle deviation point by point, and calculating their mean, root mean square, etc., as measured error values. To assess lane line recognition accuracy, the algorithm compares the lane line geometric parameters output by the virtual perception system with the actual geometric model of the lane lines in the scene, calculating the error in estimating the lateral position of the lane lines within the effective distance, or the accuracy of lane line type recognition. This calculation process relies on precise timestamp alignment, spatial coordinate transformation, and mature detection and evaluation algorithms, ultimately outputting a structured dataset of real-world perception metrics.

[0089] The core of the quantitative evaluation of control response data from the virtual chassis model is calculating measured values ​​of indicators reflecting the vehicle's basic performance in execution and tracking. During the verification scenario, the system records two sets of key time-series data: one set consists of standard control commands or reference trajectories preset in the verification scenario or issued by a simple controller, including target speed and target path; the other set consists of the actual motion response generated by the virtual chassis model, including the vehicle's actual speed, acceleration, position, yaw angle, and the actual state of the actuators. Based on these two sets of data, the system automatically performs the following analyses: For trajectory tracking deviation, the algorithm compares the vehicle's actual driving trajectory with the reference trajectory, calculates the lateral deviation at each sampling point in the normal direction, and calculates its maximum value and root mean square value as measured indicators. For speed or acceleration response error, during the period when the vehicle executes acceleration or braking commands, the algorithm calculates the difference between the actual speed curve and the target speed curve, analyzing steady-state error, response time, or overshoot as measured values. For steering angle tracking error, when the verification scenario requires a specific steering operation, the algorithm compares the target steering angle command with the front wheel steering angle of the model's actual response, calculating its following error and delay time. These calculations are all based on time series analysis, filtering, and performance measurement methods in control theory.

[0090] In the decision-making phase of the entire judgment process, the data processing logic compares the calculated measured values ​​of various indicators one by one with a predefined set of "functional expected thresholds" stored in the system. This threshold set is divided into two parts: the expected threshold for perception function and the expected threshold for control function. Each specific indicator has a corresponding upper or lower limit. For example, the expected threshold for perception function may stipulate that, in a specific verification scenario, the recall rate of target detection must not be lower than 95%, and the root mean square value of the positioning error must not be greater than 0.1 meters; the expected threshold for control function may stipulate that the root mean square value of the lateral deviation of trajectory tracking must not be greater than 0.15 meters, and the adjustment time of the velocity step response must not exceed 2 seconds.

[0091] During the comparison and calibration process, the system generates a Boolean value for each indicator: if the measured value is better than or equal to its corresponding expected functional threshold, the indicator is deemed to meet the requirements; otherwise, it is deemed not to meet the requirements. The final overall judgment logic is usually based on a comprehensive analysis of the judgment results of all key indicators. A typical strategy is to require that the judgment results of all specified core perception indicators and core control indicators be "compliant," and the system ultimately determines that the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives. If any core indicator fails to meet the standard, the overall judgment is "non-compliant." This judgment method based on quantitative thresholds and clear logic completely eliminates subjectivity and ambiguity in the evaluation process, providing a clear and consistent decision-making basis for whether the system is qualified to enter formal testing.

[0092] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the autonomous driving debugging, calibration, simulation and testing method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0093] This application provides an autonomous driving debugging and calibration simulation test device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the autonomous driving debugging and calibration simulation test method in the above embodiment 1.

[0094] The following is for reference. Figure 8 The diagram illustrates a structural schematic of an autonomous driving debugging, calibration, and simulation test device suitable for implementing embodiments of this application. The autonomous driving debugging, calibration, and simulation test device in this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8 The autonomous driving debugging, calibration, and simulation test equipment shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0095] like Figure 8As shown, the autonomous driving debugging, calibration, and simulation test equipment may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the autonomous driving debugging, calibration, and simulation test equipment. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the autonomous driving debugging, calibration, and simulation test equipment to exchange data wirelessly or via wired communication with other devices. Although the autonomous driving debugging, calibration, and simulation test equipment with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0096] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0097] The autonomous driving debugging and calibration simulation test equipment provided in this application, employing the autonomous driving debugging and calibration simulation test method in the above embodiments, can solve the technical problem that existing simulation tests cannot effectively evaluate the calibration quality of real vehicles. Compared with the prior art, the beneficial effects of the autonomous driving debugging and calibration simulation test equipment provided in this application are the same as those of the autonomous driving debugging and calibration simulation test method provided in the above embodiments, and other technical features in this autonomous driving debugging and calibration simulation test equipment are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0098] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0099] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0100] This application provides a storage medium, which is a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the autonomous driving debugging, calibration, simulation, and testing method in the above embodiments.

[0101] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0102] The aforementioned computer-readable storage medium may be included in the autonomous driving debugging, calibration, simulation, and testing equipment; or it may exist independently and not be assembled into the autonomous driving debugging, calibration, simulation, and testing equipment.

[0103] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the autonomous driving debugging, calibration, simulation, and testing equipment, the autonomous driving debugging, calibration, simulation, and testing equipment implements the technical content of the autonomous driving debugging, calibration, simulation, and testing method embodiment shown above.

[0104] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0105] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0106] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0107] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described autonomous driving debugging and calibration simulation test method. This solves the technical problem that existing simulation tests cannot effectively evaluate the calibration quality of real vehicles. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the autonomous driving debugging and calibration simulation test method provided in the above embodiments, and will not be repeated here.

Claims

1. A simulation test method for debugging, calibration, and testing autonomous driving, characterized in that, The autonomous driving debugging, calibration, and simulation testing method includes the following steps: In response to the user-initiated autonomous driving test command, the virtual sensor model and virtual chassis model are driven to generate simulation perception data based on the autonomous driving test command; The preset perception algorithm, planning and decision-making algorithm, and control algorithm are used to perform closed-loop simulation tests on the simulated perception data. The trajectory data, control data, and sensing output data during the closed-loop simulation test are recorded, and the trajectory data, control data, and sensing output data are quantitatively evaluated to obtain the quantitative evaluation results. The multidimensional calibration score is calculated based on the quantitative evaluation results, and a multidimensional calibration effect report based on the multidimensional calibration score is generated. The multidimensional calibration score includes sensor calibration score, chassis calibration score and comprehensive performance score.

2. The autonomous driving debugging, calibration, simulation, and testing method as described in claim 1, characterized in that, The step of responding to a user-initiated autonomous driving test command and generating simulation perception data based on the autonomous driving test command to drive the virtual sensor model and virtual chassis model includes: According to the autonomous driving test instruction, a preset test scenario is loaded, and the scenario parameters of the test scenario include at least one of road type, weather conditions, lighting conditions and traffic flow density. Based on the test scenario, the virtual sensor model is driven to generate initial simulation perception data, which includes at least one of point cloud data, image data, detection point data, inertial data, and positioning data. The virtual chassis model is driven to simulate vehicle dynamics response using the test scenario and the initial simulation perception data, thereby obtaining the simulation perception data.

3. The autonomous driving debugging, calibration, simulation, and testing method as described in claim 1, characterized in that, The step of performing closed-loop simulation testing on the simulated sensing data by executing preset perception algorithms, planning and decision-making algorithms, and control algorithms includes: The simulation perception data is processed by the perception algorithm to output environmental perception results, which include obstacle information, lane line information and traffic sign information. The planning and decision-making algorithm performs behavioral decisions based on the environmental perception results, the high-precision map information of the test scenario, and the preset driving task to generate a target trajectory. The target trajectory is calculated using the control algorithm to output longitudinal and lateral control commands, and the virtual chassis model is driven based on the longitudinal and lateral control commands to perform closed-loop testing.

4. The autonomous driving debugging, calibration, simulation, and testing method as described in claim 1, characterized in that, Prior to the step of responding to a user-initiated autonomous driving test command, the method further includes: The calibration parameters of a real autonomous vehicle are obtained, and the calibration parameters are standardized to obtain simulation compatibility parameters. The simulation compatibility parameters are synchronously updated to the virtual sensor model parameters and virtual chassis model parameters of the simulation software.

5. The autonomous driving debugging, calibration, and simulation testing method as described in claim 4, characterized in that, The steps of synchronizing and updating the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software include: Based on the sensor calibration parameters in the simulation compatibility parameters, update the configuration parameters of the virtual sensor model group. The virtual sensor model group includes at least a virtual lidar model, a virtual millimeter-wave radar model, a virtual camera model, a virtual inertial measurement unit model, and a virtual global positioning system model. The configuration parameters include the intrinsic parameter attributes, installation pose attributes, and transformation relationship parameters of each model relative to the vehicle coordinate system. Based on the chassis calibration parameters in the simulation compatibility parameters, update the configuration parameters of the virtual chassis model group. The virtual chassis model group includes at least a virtual vehicle dynamics model and a virtual actuator model. The configuration parameters include the geometric properties, mass properties, and tire properties of the virtual vehicle dynamics model, as well as the dynamic response characteristics and control logic parameters of the virtual actuator model.

6. The autonomous driving debugging, calibration, and simulation testing method as described in claim 4, characterized in that, After the step of synchronizing and updating the simulation compatibility parameters to the virtual sensor model parameters and virtual chassis model parameters of the simulation software, the method further includes: After the parameters are synchronized and updated, a preset verification scenario is run in the simulation test environment to drive the updated virtual sensor model and virtual chassis model to perform parameter synchronization tests. Analyze the response data in the test results, and determine whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results; If the response data is determined to meet the expected test objective, the steps for responding to the user-initiated autonomous driving test command are executed.

7. The autonomous driving debugging, calibration, and simulation testing method as described in claim 6, characterized in that, After analyzing the response data in the test results and determining whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results, the method further includes: If the response data is determined not to meet the expected test target, a parameter synchronization fine-tuning strategy is executed to make the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test target. The step of implementing the parameter synchronization fine-tuning strategy to ensure that the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives includes: Based on the comparison between the response data and the expected data, one or more simulation compatibility parameters that cause performance deviations are located in the comparison results. The simulation compatibility parameters include sensor calibration parameters and chassis calibration parameters. Within the preset parameter adjustment range, the one or more simulation compatibility parameters are iteratively adjusted, and the preset benchmark verification scenario is rerun after each iteration until the response parameters obtained after rerun meet the expected test target, at which point the iteration stops.

8. The autonomous driving debugging, calibration, and simulation testing method as described in claim 6, characterized in that, The step of analyzing the response data in the test results and determining whether the basic perception and control functions of the virtual sensor model and the virtual chassis model meet the expected test objectives based on the analysis results includes: Based on the output of the virtual sensor model in the simulation test environment, the measured values ​​based on perception indicators are calculated. The perception indicators include at least one of target detection rate, positioning error, and lane line recognition accuracy. Based on the control response of the virtual chassis model in the simulation test environment, the measured values ​​of the control indicators are calculated, including at least one of trajectory tracking deviation, speed / acceleration response error, and steering angle tracking error. The measured values ​​are compared with the corresponding expected functional thresholds, and the results are used to determine whether the expected test targets are met. The expected functional thresholds include the expected thresholds for perception functions and the expected thresholds for control functions.

9. An autonomous driving debugging, calibration, and simulation testing device, characterized in that, The autonomous driving debugging, calibration, and simulation test equipment stores a computer program, which, when executed by a processor, implements the autonomous driving debugging, calibration, and simulation test method according to any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the autonomous driving debugging, calibration, simulation, and testing method according to any one of claims 1-8.

Images