Simulation test method of unmanned mine car and storage medium

Abstract

The application provides a simulation test method of an unmanned mine car and a storage medium. The simulation test method of the unmanned mine car comprises the following steps: acquiring abnormal data and a to-be-tested control model when the mine car is running; based on working condition characteristics, the abnormal data is analyzed to generate structured vehicle data and a spatial scene model; a simulation vehicle entity is generated and a simulation environment is constructed; a sensor physical model is called to generate virtual perception data, the virtual perception data is input into the to-be-tested control model, and control instructions are output; based on the control instructions and simulation execution, a to-be-tested simulation vehicle is configured to perform a simulation task, and running state data is generated through a mine car dynamics model; the virtual perception data and the running state data are evaluated according to quantitative evaluation indexes, a multi-dimensional evaluation matrix is generated, and defect characteristics are located; an updated to-be-tested control model is acquired, and the simulation task is re-executed to obtain regression test results. The application avoids costs and risks caused by repeated on-road test driving.

Description

Technical Field

[0001] This invention relates to the field of unmanned driving simulation testing technology, and more specifically, to a simulation testing method and storage medium for an unmanned mining truck. Background Technology

[0002] Currently, in related technologies, the verification of control algorithms for unmanned mining trucks mainly relies on real-vehicle testing. Real-vehicle testing is costly, risky, and time-consuming, and it's difficult to safely reproduce occasional hazardous conditions in real-world mining environments. Fault data collected during actual vehicle operation lacks effective conversion methods, making it difficult to create reusable virtual simulation scenarios, hindering the implementation of real-world fault reproduction tests. Furthermore, mining operations are characterized by heavy vehicle loads, steep slopes, and frequent multi-vehicle interactions, making it difficult for simulation methods to simultaneously handle concurrent multi-vehicle operations and achieve high-fidelity simulations of the unique working conditions in mining areas. These issues make it difficult to obtain high-frequency, reliable, and reproducible verification of the control algorithms for unmanned mining trucks. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

[0004] Therefore, the first aspect of the present invention proposes a simulation testing method for unmanned mining trucks.

[0005] A second aspect of the present invention provides a storage medium.

[0006] In view of this, the first aspect of the present invention proposes a simulation testing method for an unmanned mining truck, comprising: acquiring abnormal data and a control model under test during the operation of the mining truck; analyzing the abnormal data based on the operating condition characteristics to generate structured vehicle data and a spatial scene model; generating a simulated vehicle entity based on the structured vehicle data, the simulated vehicle entity including an environmental interaction vehicle and a simulated vehicle under test; constructing a simulation environment based on the simulated vehicle entity and the spatial scene model; in the simulation environment, calling a sensor physical model to generate virtual perception data of the simulated vehicle under test; inputting the virtual perception data into the control model under test and outputting control commands; controlling the simulated vehicle under test to perform simulation operations based on the control commands and simulation execution configuration; calling a mining truck dynamics model, the mining truck dynamics model responding to the control commands to generate operating status data of the simulated vehicle under test; evaluating the virtual perception data and operating status data according to preset quantitative evaluation indicators to generate a multi-dimensional evaluation matrix; locating the defect features of the control model under test based on the multi-dimensional evaluation matrix; obtaining an updated control model under test based on the defect features; and re-executing the simulation operation of the simulated vehicle under test based on the updated control model under test and the spatial scene model to obtain regression test results.

[0007] The simulation testing method for unmanned mining trucks proposed in this invention enables real vehicle anomaly data to be quickly transformed into reproducible virtual simulation scenarios, avoiding the costs and risks associated with repeated road tests, reducing the difficulty in reproducing extreme and dangerous working conditions, improving the verification efficiency and iteration speed of unmanned mining algorithms, and forming a complete data closed loop integrating problem discovery, scenario reproduction, algorithm iteration, and regression verification.

[0008] Optionally, in some technical solutions, abnormal data is analyzed based on operating condition characteristics to generate structured vehicle data and spatial scene models, including: converting the abnormal data into a standard format to generate standard format data; extracting vehicle information from the standard format data to generate structured vehicle data; and extracting scene information from the standard format data to construct a spatial scene model.

[0009] In the above technical solution, the SIL (Software in the Loop) platform organizes the inconsistent abnormal data into structured vehicle data and spatial scene models with separate vehicle and scene dimensions by converting the format, extracting vehicle information and scene information. This improves the efficiency of converting abnormal data into simulation scenes and avoids the problem of data not being able to be used directly in the simulation process due to inconsistent data formats. It also facilitates the separate generation of structured vehicle data and spatial scene models.

[0010] Optionally, in some technical solutions, before constructing the simulation environment based on the simulated vehicle entity and spatial scene model, the simulation testing method for unmanned mining trucks may also include: obtaining test constraint parameters, including time-series period parameters, vehicle density parameters, and vehicle heavy load parameters; and generating simulation execution configuration based on the test constraint parameters and spatial scene model.

[0011] In the above technical solution, by acquiring test constraint parameters and generating simulation execution configuration, the SIL platform makes the execution of simulation jobs configurable and reproducible, thereby improving the consistency of simulation job execution.

[0012] Optionally, in some technical solutions, a simulation environment is constructed based on the simulated vehicle entities and the spatial scene model, including: generating a three-dimensional road network based on the spatial scene model; mapping each simulated vehicle entity to the corresponding coordinate node in the three-dimensional road network to generate the simulation environment.

[0013] In the above technical solution, the SIL platform generates a three-dimensional road network and maps each simulated vehicle entity to the corresponding coordinate node, so that the spatial position of each simulated vehicle entity in the simulation environment corresponds to the spatial position of the mining area road. This improves the consistency between the simulation environment and the real mining area space, avoids the problem of the position setting of each simulated vehicle entity being disconnected from the mining area road space, and also facilitates the completion of multi-entity collaborative interaction simulation between the environmental interaction vehicle and the simulated vehicle under test under the same spatial reference.

[0014] Optionally, in some technical solutions, the sensor physical model includes a ray tracing model and / or an optical rendering model, and the virtual perception data includes 3D point cloud data and simulation image data. In the simulation environment, the sensor physical model is invoked to generate virtual perception data of the simulated vehicle under test, including: in the simulation environment, invoking the ray tracing model to generate 3D point cloud data; and in the simulation environment, invoking the optical rendering model to generate simulation image data.

[0015] In the above technical solution, the SIL platform generates three-dimensional point cloud data and simulation image data through ray tracing model and optical rendering model respectively, so that the virtual perception data has both laser point cloud and visual image perception content. This improves the completeness and realism of the virtual perception data in perceiving the mining environment, and enables the control model under test to complete software-in-the-loop verification under conditions close to the perception of the real mining area.

[0016] Optionally, in some technical solutions, a mining truck dynamics model is invoked. The mining truck dynamics model responds to control commands and generates operating status data of the simulated vehicle under test. This includes: invoking the mining truck dynamics model, which generates operating status data based on control commands and at least one of the following factors: the mining truck's full-load mass parameters, unloaded mass parameters, tire side slip characteristics, and brake pressure delay parameters.

[0017] In the above technical solution, the SIL platform generates operating status data based on the control commands and full-load mass parameters, unloaded mass parameters, tire side slip characteristics and brake pressure delay parameters through the mine car dynamics model. This enables the operating status data to reflect the real motion response of the mine car under different load conditions and braking conditions, improving the consistency between the operating status data and the actual vehicle behavior, and also improving the reliability of the motion simulation results of the vehicle under test.

[0018] Optionally, in some technical solutions, the defect characteristics of the control model under test are located based on a multi-dimensional evaluation matrix, including: determining the abnormal time series interval that triggers the preset risk threshold based on the multi-dimensional evaluation matrix; extracting the target perception data and target control commands corresponding to the abnormal time series interval based on the abnormal time series interval; generating a problem scenario model based on the target perception data and target control commands; reproducing the abnormal situation based on the problem scenario model; and locating the defect characteristics based on the abnormal situation.

[0019] In the above technical solution, the SIL platform determines the abnormal time interval, intercepts target perception data and target control commands, generates a problem scenario model, reproduces the abnormal situation, and locates defect features based on the abnormal situation. This limits the defect localization process to the time period in which the problem occurs, improves the targeting of defect localization, avoids blindly searching for defects in the full simulation data, and avoids repeatedly executing the complete simulation to reproduce occasional failures.

[0020] Optionally, in some technical solutions, an updated control model to be tested is obtained based on the defect characteristics, including: calling a large language model, modifying the control model to be tested according to the defect characteristics, and obtaining the updated control model to be tested.

[0021] In the above technical solution, the SIL platform automates the defect repair process by calling a large language model to modify the control model under test according to the defect characteristics. This improves the efficiency of algorithm iteration, avoids the time overhead caused by manual code modification, and also improves the connection efficiency between defect repair and simulation verification.

[0022] Optionally, in some technical solutions, the preset quantitative evaluation indicators include minimum safe distance, collision time, average traffic speed, and emergency braking distance. Based on the preset quantitative evaluation indicators, the virtual sensing data and operational status data are evaluated to generate a multi-dimensional evaluation matrix, including: evaluating the virtual sensing data based on the minimum safe distance to obtain a minimum safe distance evaluation value; evaluating the virtual sensing data and operational status data based on the collision time to obtain a collision time evaluation value; evaluating the operational status data based on the average traffic speed to obtain an average traffic speed evaluation value; evaluating the operational status data based on the emergency braking distance to obtain an emergency braking distance evaluation value; and generating a multi-dimensional evaluation matrix based on the minimum safe distance evaluation value, collision time evaluation value, average traffic speed evaluation value, and emergency braking distance evaluation value.

[0023] In the above technical solution, the SIL platform evaluates virtual perception data and operational status data from four dimensions: minimum safe distance, collision time, average traffic speed, and emergency braking distance, and generates a multi-dimensional evaluation matrix. This expands the evaluation of simulation results from a single dimension to multiple dimensions, improving the comprehensiveness of the evaluation and the accuracy of subsequent defect location.

[0024] A second aspect of the present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the simulation testing method for the unmanned mining truck in any of the above-described technical solutions. Therefore, this storage medium possesses all the beneficial effects of the simulation testing method for the unmanned mining truck in any of the above-described technical solutions, which will not be elaborated further here.

[0025] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0027] Figure 1 One of the flowcharts for a simulation test method of an unmanned mining truck according to an embodiment of the present invention;

[0028] Figure 2 This is a second flowchart of a simulation testing method for an unmanned mining truck according to an embodiment of the present invention;

[0029] Figure 3 The third flowchart is a simulation test method for an unmanned mining truck according to an embodiment of the present invention;

[0030] Figure 4 The fourth flowchart is a simulation test method for an unmanned mining truck according to an embodiment of the present invention;

[0031] Figure 5 The fifth flowchart is a simulation test method for an unmanned mining truck according to an embodiment of the present invention;

[0032] Figure 6 This is the sixth flowchart of a simulation testing method for an unmanned mining truck according to an embodiment of the present invention;

[0033] Figure 7 This is a structural block diagram of an electronic device according to an embodiment of the present invention;

[0034] Figure 8 This is a structural block diagram of a simulation test system for an unmanned mining truck according to an embodiment of the present invention.

[0035] in, Figure 7 and Figure 8 The correspondence between the reference numerals and component names in the attached drawings is as follows:

[0036] Electronic device 70; memory 702; processor 704; simulation test system for unmanned mining truck 80; first acquisition module 802; parsing module 804; first generation module 806; environment construction module 808; second generation module 810; output module 812; execution module 814; third generation module 816; fourth generation module 818; positioning module 820; second acquisition module 822; regression test module 824. Detailed Implementation

[0037] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0038] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0039] The following reference Figures 1 to 8 A simulation testing method and storage medium for an unmanned mining truck according to some embodiments of the present invention are described.

[0040] The autonomous driving algorithms for mining trucks in open-pit mines need to be repeatedly verified for safety and reliability in scenarios such as complex mining terrain, multi-vehicle interaction, and long-term continuous operation. Real-vehicle testing has drawbacks such as high cost, high risk, and long cycle, and it is difficult to reproduce extremely dangerous working conditions in real mines.

[0041] like Figure 1 As shown, in view of the above, the first aspect of the present invention proposes a simulation testing method for an unmanned mining truck, comprising:

[0042] Step 102: Obtain abnormal data and the control model to be tested during the operation of the mining truck;

[0043] Step 104: Based on the operating condition characteristics, analyze the abnormal data to generate structured vehicle data and spatial scene models;

[0044] Step 106: Generate simulated vehicle entities based on structured vehicle data. The simulated vehicle entities include environmental interaction vehicles and simulated vehicles under test.

[0045] Step 108: Construct the simulation environment based on the simulated vehicle entity and spatial scene model;

[0046] Step 110: In the simulation environment, call the sensor physical model to generate virtual perception data of the simulated vehicle under test;

[0047] Step 112: Input the virtual sensing data into the control model under test and output control commands;

[0048] Step 114: Based on the control commands and simulation execution configuration, control the vehicle under test to perform simulation operations;

[0049] Step 116: Call the mine car dynamics model. The mine car dynamics model responds to the control command and generates the operating status data of the simulated vehicle under test.

[0050] Step 118: Evaluate the virtual sensing data and operational status data according to the preset quantitative evaluation indicators, and generate a multi-dimensional evaluation matrix;

[0051] Step 120: Based on the multidimensional evaluation matrix, locate the defect characteristics of the control model under test;

[0052] Step 122: Based on the defect characteristics, obtain the updated control model to be tested;

[0053] Step 124: Based on the updated control model under test and spatial scene model, re-execute the simulation operation for the simulated vehicle under test to obtain the regression test results.

[0054] This invention proposes a simulation testing method for unmanned mining trucks. The method utilizes a SIL (Software in the Loop) platform running on GPUs (Graphics Processing Units) or high-performance computing clusters. It leverages the Kubernetes (Kubernetes) containerization platform to provide elastic container services, high-speed caching services, object storage, and a GPU computing power pool, supporting distributed simulation and multi-GPU parallel rendering, achieving elastic scaling of computing power and resource isolation. The SIL platform interacts with the scheduling cloud platform, DevOps (Development and Operations) platform, and data closed-loop platform via an Ethernet API (Application Programming Interface). Communication employs either HTTP (Hypertext Transfer Protocol) or MQTT (Message Queuing Telemetry Transport) protocols.

[0055] The SIL platform integrates OpenDrive (ASAM OpenDRIVE, Open Road Network Description Standard Format) or OpenScenario (ASAM OpenSCENARIO, Open Scene Description Standard Format) editors, 3D (Three-Dimensional) mine scene generation, high-precision dynamic models of heavy trucks, and physical-grade sensor simulation capabilities, and supports real-time Web3D (Web Three-Dimensional) preview. The SIL platform's scheduling system layer also provides perception and positioning algorithms and algorithm access interfaces.

[0056] The SIL platform transforms abnormal data from mining truck operation into structured vehicle data and spatial scene models, further generating simulated vehicle entities and constructing a simulation environment. It forms a software-in-the-loop simulation closed loop through sensor physical models, control models under test, and mining truck dynamics models. Then, it locates defect characteristics through quantitative evaluation and obtains updated control models under test. The simulation operation is then re-executed to obtain regression test results, reducing the repeated investment in real vehicle testing, avoiding the problem that extreme and dangerous working conditions are difficult to reproduce in real mining areas, and improving the verification efficiency and reliability of unmanned driving algorithms in mines.

[0057] The SIL platform first acquires abnormal data and the control model under test during the operation of the mining truck, providing input for subsequent construction of simulation scenarios and execution of simulation tasks.

[0058] Among them, abnormal data refers to data collected and stored on a disk during the actual operation of mining vehicles in the mining area. Abnormal data includes data on abnormal operating conditions or information on periods of failure. The control model under test refers to the model to be verified corresponding to the automatic driving control algorithm of the mining vehicle.

[0059] In some embodiments, the SIL platform can directly read abnormal data through the on-board storage interface of vehicles at the mining site; in other embodiments, it can also retrieve abnormal data stored on the actual vehicle through a data closed-loop platform.

[0060] In some embodiments, the source of the control model under test may be the algorithm image package continuously delivered by the DevOps platform; in other embodiments, the source of the control model under test may be the algorithm compiled locally.

[0061] After acquiring the abnormal data, the SIL platform analyzes the abnormal data based on the operating condition characteristics to generate structured vehicle data and spatial scene models, providing a foundation for the subsequent generation of simulated vehicle entities and the construction of simulation environments.

[0062] Among them, working condition characteristics refer to the working condition category of the mining truck during mining operations, structured vehicle data refers to the structured information of the simulated vehicle restored from abnormal data, and spatial scene model refers to the three-dimensional spatial geometric model of the mining area terrain.

[0063] In some embodiments, when the SIL platform parses abnormal data, it can reconstruct the vehicle and scene dimensions from the abnormal data through scene mining; in other embodiments, the SIL platform can also parse abnormal data through multi-source information parsing.

[0064] After generating structured vehicle data, the SIL platform generates simulated vehicle entities based on the structured vehicle data. The simulated vehicle entities include environmental interaction vehicles and simulated vehicles under test, providing vehicle elements for the subsequent construction of the simulation environment.

[0065] Among them, the simulated vehicle entity refers to the digital entity of each vehicle in the mining area simulation scenario, the environmental interactive vehicle refers to the vehicle in the scenario that interacts with the simulated vehicle under test, and the simulated vehicle under test refers to the mining vehicle equipped with the control model under test and used as the object of test.

[0066] In some embodiments, when the SIL platform generates a simulated vehicle entity, it can instantiate the simulated vehicle entity one by one according to the vehicle parameters recorded in the structured vehicle data; in other embodiments, it can also retrieve a vehicle model that matches the structured vehicle data from the vehicle model library to generate the simulated vehicle entity.

[0067] The SIL platform incorporates both the environmental interaction vehicle and the simulated vehicle under test into the simulated vehicle entity, enabling the simulated vehicle under test to complete multi-vehicle interaction verification in a virtual scene that includes the environmental interaction vehicle. This avoids the problem of verifying only a single vehicle and being disconnected from the real multi-vehicle collaborative scenario in the mining area.

[0068] After generating the simulated vehicle entity, the SIL platform constructs a simulation environment based on the simulated vehicle entity and the spatial scene model, providing a virtual operating scenario for the simulated vehicle under test.

[0069] The simulation environment refers to a virtual operating scenario that allows the simulated vehicle under test to drive, perceive, and make decisions. The SIL platform deploys each simulated vehicle entity on the terrain and roads described by the spatial scene model, forming a virtual operating scenario containing multiple mining trucks.

[0070] In some embodiments, when building a simulation environment, the SIL platform can deploy simulated vehicle entities onto the spatial scene model using a scene editor; in other embodiments, when building a simulation environment, the SIL platform can also assign simulated vehicle entities to corresponding positions in the spatial scene model using automated deployment rules.

[0071] After building the simulation environment, the SIL platform calls the sensor physical model in the simulation environment to generate virtual perception data of the simulated vehicle under test, providing perception input for the control model under test.

[0072] Among them, the sensor physical model refers to the model that simulates the physical perception process of the sensors carried by the mining vehicle, and the virtual perception data refers to the data perceived by the sensors carried by the simulated vehicle in the virtual scene.

[0073] In some embodiments, when the SIL platform calls the sensor physical model to generate virtual sensing data, it can generate virtual sensing data through physical optics simulation calculation; in other embodiments, the SIL platform can also generate virtual sensing data through geometric occlusion relationships.

[0074] After generating virtual sensing data, the SIL platform inputs the virtual sensing data into the control model under test (DUT), which then outputs control commands to provide a basis for controlling the simulated vehicle under test. These control commands are the manipulation commands generated by the DUT in response to the driving actions of the simulated vehicle.

[0075] In some embodiments, the SIL platform can input virtual sensing data into the control model under test via shared memory; in other embodiments, the SIL platform can also input virtual sensing data into the control model under test via message queues.

[0076] Virtual sensing data serves as input to the control model under test (DUT), and after processing by the DUT, it forms control commands, thereby completing the software-in-the-loop verification of the DUT during the simulation phase.

[0077] After outputting control commands, the SIL platform controls the vehicle under test to perform simulation tasks based on the control commands and simulation execution configuration. The simulation execution configuration refers to the information used to configure the execution mode of the simulation tasks.

[0078] The SIL platform drives the driving actions of the simulated vehicle under test according to control commands, and sets the execution process of the simulation operation according to the simulation execution configuration, so that the simulated vehicle under test can complete continuous driving and interactive operations in the virtual scene.

[0079] In some embodiments, when the SIL platform controls the vehicle under test to perform simulation operations, it can directly drive the steering and braking of the vehicle under test to perform simulation operations through control commands; in other embodiments, the SIL platform can also perform simulation operations by sending control commands to the control interface of the vehicle under test.

[0080] During the simulation process, the SIL platform calls the mine car dynamics model, which responds to control commands and generates the operating status data of the simulated vehicle under test.

[0081] Among them, the mining truck dynamics model refers to the dynamic model that simulates the motion response of mining vehicles, and the operating status data refers to the motion status data of the simulated vehicle under test during the simulation operation process. The operating status data characterizes the motion information such as the position, attitude and speed of the simulated vehicle under test.

[0082] In some embodiments, the dynamic model of the mining truck responds to control commands to calculate the motion of the simulated vehicle under test, and the operating status data can be calculated by numerical integration; in other embodiments, the operating status data can also be calculated by lookup table interpolation.

[0083] The dynamic model of the mining truck responds to control commands to calculate the motion of the simulated vehicle under test, making the motion process of the simulated vehicle closely resemble the real dynamic response of mining vehicles, thus improving the consistency between simulation results and actual vehicle behavior.

[0084] After generating operational status data, the SIL platform evaluates the virtual sensing data and operational status data according to preset quantitative evaluation indicators, generating a multi-dimensional evaluation matrix to provide a basis for locating the defect characteristics of the control model under test.

[0085] Among them, quantitative evaluation indicators refer to indicators used to quantitatively measure simulation results, and multidimensional evaluation matrix refers to data matrix formed by quantitatively evaluating simulation results from multiple dimensions.

[0086] In some embodiments, the SIL platform can evaluate virtual sensing data and operational status data through quantitative scoring; in other embodiments, the SIL platform can also evaluate virtual sensing data and operational status data through rule engine judgment.

[0087] Through multi-dimensional quantitative evaluation, the SIL platform transforms simulation results into a comparable and analyzable multi-dimensional evaluation matrix, thereby improving the accuracy of defect localization.

[0088] After generating the multidimensional evaluation matrix, the SIL platform uses the matrix to locate the defective features of the control model under test (DUT). These defective features refer to characteristics of the DUT that lead to abnormal control behavior during simulation.

[0089] In some embodiments, when the SIL platform locates defect features, it can do so by performing pattern recognition on the multidimensional evaluation matrix; in other embodiments, the SIL platform can also locate defect features by analyzing the poorly performing evaluation items in the multidimensional evaluation matrix.

[0090] The SIL platform analyzes the multidimensional evaluation matrix to locate the defect features exposed by the control model under test in the simulation operation, so that the control model under test can be updated in the future based on the defect features.

[0091] After locating the defect characteristics, the SIL platform obtains an updated control model under test based on these characteristics. The updated control model under test refers to the control model obtained after updating it according to the defect characteristics.

[0092] When the SIL platform obtains the updated control model under test, it can pull the algorithm image from the image repository as the updated control model under test, or it can directly load the updated control model file from the algorithm library as the updated control model under test.

[0093] The SIL platform obtains the updated control model under test through the continuous integration pipeline of the DevOps platform. The DevOps platform compiles and performs unit tests on the updated control code, packages it into a Docker image and pushes it to the image repository. The SIL platform pulls the algorithm image from the image repository, thus completing the acquisition of the updated control model under test.

[0094] The SIL platform obtains updated control models under test based on defect characteristics, enabling defect repair to be automatically carried over to the next simulation job, thus improving the efficiency of algorithm iteration.

[0095] After acquiring the updated control model under test (DUT), the SIL platform re-executes the simulation operation on the simulated vehicle under test based on the updated DUT and spatial scenario model, obtaining regression test results. The regression test results refer to the results obtained after re-executing the simulation operation on the simulated vehicle under test; these results are used to verify whether the defects have been repaired.

[0096] The SIL platform re-executes the simulation operation on the vehicle under test and confirms whether the defect features have been repaired through regression test results, thereby completing a complete closed-loop iteration of problem discovery, scenario reproduction, algorithm iteration, and regression verification.

[0097] The SIL platform transforms abnormal data from mining truck operation into structured vehicle data and spatial scene models, further generating simulated vehicle entities and constructing a simulation environment. It forms a software-in-the-loop simulation closed loop through sensor physical models, control models under test, and mining truck dynamics models. Then, it locates defect characteristics through quantitative evaluation and obtains updated control models under test, and re-executes the simulation operation to obtain regression test results.

[0098] The simulation testing method for unmanned mining trucks proposed in this invention enables real vehicle anomaly data to be quickly transformed into reproducible virtual simulation scenarios, avoiding the costs and risks associated with repeated road tests, reducing the difficulty in reproducing extreme and dangerous working conditions, improving the verification efficiency and iteration speed of unmanned mining algorithms, and forming a complete data closed loop integrating problem discovery, scenario reproduction, algorithm iteration, and regression verification.

[0099] like Figure 2 As shown, optionally, in some embodiments, based on operating condition characteristics, abnormal data is parsed to generate structured vehicle data and spatial scene models, including:

[0100] Step 202: Convert the abnormal data to generate standard format data;

[0101] Step 204: Extract vehicle information from standard format data to generate structured vehicle data;

[0102] Step 206: Extract scene information from standard format data and construct a spatial scene model.

[0103] In the above embodiment, the SIL platform first performs format conversion on the abnormal data to generate standard format data.

[0104] Format conversion refers to converting the original record format of abnormal data into a unified format that can be processed later. Standard format data refers to data with a unified data format obtained after format conversion.

[0105] Abnormal data is stored on disk in the original log format at the mining site. The data recording formats generated by each sensor and vehicle control unit are inconsistent. The SIL platform converts the inconsistent abnormal data into standard format data, providing a unified data foundation for subsequent extraction of vehicle information and scene information.

[0106] After generating standard format data, the SIL platform extracts vehicle information from it to generate structured vehicle data. This vehicle information refers to details specific to the mining vehicles within the standard format data. In other words, the SIL platform extracts vehicle information from the standard format data, separating vehicle-related information to form structured vehicle data.

[0107] After extracting vehicle information, the SIL platform extracts scene information from standard format data to construct a spatial scene model. Scene information refers to information related to the mining area environment within the standard format data. In other words, the SIL platform extracts scene information from the standard format data, separating scene-dimensional information from the standard format data to construct a spatial scene model.

[0108] In some embodiments, when the SIL platform performs format conversion on abnormal data, it can convert the original record format used by the abnormal data into a standard format. Specifically, the abnormal data is stored on disk in rosbag (Robot Operating System bag) format, and the SIL platform can convert the rosbag format to standard format data.

[0109] In some embodiments, format conversion can also be achieved through timestamp alignment, where the SIL platform aligns the data records generated by each sensor and each vehicle control unit with timestamps as the reference. In other embodiments, format conversion can also be achieved through multi-source sensor message parsing, where the SIL platform parses the messages generated by each sensor and organizes the messages from different sources into standard format data.

[0110] When the SIL platform extracts vehicle information from standard format data, the vehicle information may include vehicle dynamics state sequences and steering control parameters. The vehicle dynamics state sequence refers to a data sequence formed by arranging the dynamic state quantities of the mining vehicle at each moment, such as speed, acceleration, and yaw rate, in chronological order. The steering control parameters refer to parameters related to the steering control of the mining vehicle.

[0111] The SIL platform can extract historical dynamic state sequences and steering control parameters of mining vehicles from standard format data, and use them as vehicle dimension content for generating structured vehicle data.

[0112] In some embodiments, when the SIL platform extracts scene information from standard format data, the scene information may include obstacle spatial distribution and slope-elevation topology. Obstacle spatial distribution refers to the location distribution of various obstacles in the three-dimensional space within the mining area scene, while slope-elevation topology refers to information about the mining area terrain in terms of slope, elevation (altitude), and spatial topological structure.

[0113] The SIL platform can extract spatial distribution information of obstacles in mining areas and slope, elevation and topological information of mining terrain from standard format data, and use them as scene dimension content for constructing spatial scene models.

[0114] The SIL platform transforms inconsistent abnormal data into structured vehicle data and spatial scene models by converting formats, extracting vehicle information, and extracting scene information. This improves the efficiency of converting abnormal data into simulation scenes and avoids the problem of data being unusable in simulation due to inconsistent data formats. It also facilitates the separate generation of structured vehicle data and spatial scene models.

[0115] like Figure 3As shown, optionally, in some embodiments, before constructing the simulation environment based on the simulated vehicle entity and spatial scene model, the simulation testing method for the unmanned mining truck further includes:

[0116] Step 302: Obtain test constraint parameters, including time period parameters, vehicle density parameters, and vehicle heavy load parameters;

[0117] Step 304: Generate simulation execution configuration based on test constraint parameters and spatial scene model.

[0118] In the above embodiments, before constructing the simulation environment based on the simulated vehicle entity and the spatial scene model, the SIL platform first obtains the test constraint parameters, and then generates the simulation execution configuration based on the test constraint parameters and the spatial scene model, providing a configuration basis for subsequently controlling the simulated vehicle under test to perform simulation operations based on control commands and the simulation execution configuration.

[0119] Among them, the test constraint parameters refer to the set of parameters used to constrain the execution mode of the simulation operation. The test constraint parameters include the time period parameter, the vehicle density parameter, and the vehicle load parameter. The time period parameter refers to the parameter used to set the length of the time period for the execution of the simulation operation. The vehicle density parameter refers to the parameter used to set the vehicle distribution density in the simulation scenario. The vehicle load parameter refers to the parameter used to set the load state of the vehicle to be simulated.

[0120] In some embodiments, the SIL platform can obtain test constraint parameters through a scheduling cloud platform; in other embodiments, the SIL platform can also obtain test constraint parameters through a local configuration interface.

[0121] The scheduling cloud platform sends test constraint parameters to the SIL platform via the Ethernet API interface, and the SIL platform receives the test constraint parameters sent by the scheduling cloud platform.

[0122] After acquiring the test constraint parameters, the SIL platform generates a simulation execution configuration based on these parameters and the spatial scene model. The SIL platform combines the test constraint parameters with the spatial scene model, ensuring that the simulation execution configuration simultaneously reflects both the constraints set by the test constraint parameters and the mining area terrain information described by the spatial scene model.

[0123] In some embodiments, when generating a simulation execution configuration, the SIL platform can generate a simulation execution configuration that meets the constraints based on the test constraint parameters and the slope and curvature information of the spatial scene model. The slope information describes the steepness of the terrain undulations, and the curvature information describes the curvature of the road direction.

[0124] By acquiring test constraint parameters and generating simulation execution configurations, the SIL platform makes the execution of simulation jobs configurable and reproducible, thereby improving the consistency of simulation job execution.

[0125] like Figure 4 As shown, optionally, in some embodiments, a simulation environment is constructed based on the simulated vehicle entity and the spatial scene model, including:

[0126] Step 402: Generate a 3D road network based on the spatial scene model;

[0127] Step 404: Map each simulated vehicle entity to the corresponding coordinate node in the 3D road network to generate the simulation environment.

[0128] In the above embodiments, when the SIL platform constructs the simulation environment based on the simulated vehicle entities and the spatial scene model, it is further limited to first generating a three-dimensional road network based on the spatial scene model, and then mapping each simulated vehicle entity to the corresponding coordinate node in the three-dimensional road network, thereby generating the simulation environment.

[0129] Among them, the three-dimensional road network refers to the three-dimensional network structure used to describe the spatial morphology of roads in the mining area, and the corresponding coordinate node in the three-dimensional road network refers to the three-dimensional coordinate point used to locate the spatial position of the simulated vehicle entity in the three-dimensional road network.

[0130] The SIL platform generates a three-dimensional road network based on a spatial scene model. The three-dimensional road network carries the spatial orientation, slope undulation, and road surface adhesion coefficient of the mining area roads.

[0131] Among them, spatial orientation refers to the direction of extension of the mining area road in three-dimensional space; slope undulation refers to the changes in the slope of the mining area road, which is used to describe the longitudinal slope undulation of the road; and road surface adhesion coefficient refers to the adhesion coefficient between the tires of mining vehicles and the road surface of the mining area, which is used to characterize the friction adhesion performance between the tires and the road surface.

[0132] In some embodiments, when the SIL platform generates a three-dimensional road network, it can extract the road centerline and road boundary from the road geometry information described by the spatial scene model, and construct a three-dimensional road network along the road centerline. Each coordinate node in the three-dimensional road network corresponds to a spatial location on the mining area road.

[0133] After generating the 3D road network, the SIL platform maps each simulated vehicle entity to the corresponding coordinate node in the 3D road network to generate the simulation environment.

[0134] The SIL platform maps the interactive vehicle and the simulated vehicle under test to corresponding coordinate nodes in the 3D road network, so that each simulated vehicle entity has a definite spatial position in the 3D road network, thereby generating a simulation environment that includes the interactive vehicle and the simulated vehicle under test.

[0135] In some embodiments, when the SIL platform maps each simulated vehicle entity to a corresponding coordinate node, it can assign each simulated vehicle entity to the corresponding coordinate node by coordinate matching; in other embodiments, when the SIL platform maps each simulated vehicle entity to a corresponding coordinate node, it can also deploy each simulated vehicle entity to the corresponding coordinate node by spatial interpolation.

[0136] The SIL platform generates a 3D road network and maps each simulated vehicle entity to its corresponding coordinate node, making the spatial position of each simulated vehicle entity in the simulation environment correspond to the spatial position of the mining area road. This improves the consistency between the simulation environment and the real mining area space, avoids the problem of the position setting of each simulated vehicle entity being out of sync with the mining area road space, and also facilitates the completion of multi-entity collaborative interactive simulation between the environmental interaction vehicle and the simulated vehicle under test under the same spatial reference.

[0137] Optionally, in some embodiments, the sensor physical model includes a ray tracing model and / or an optical rendering model, and the virtual perception data includes 3D point cloud data and simulation image data. In the simulation environment, the sensor physical model is invoked to generate virtual perception data of the simulated vehicle under test, including: in the simulation environment, invoking the ray tracing model to generate 3D point cloud data; and in the simulation environment, invoking the optical rendering model to generate simulation image data.

[0138] In the above embodiments, when the SIL platform calls the sensor physical model in the simulation environment to generate virtual perception data of the simulated vehicle under test, it further defines the sensor physical model as including ray tracing model and / or optical rendering model, and the virtual perception data as including three-dimensional point cloud data and simulation image data.

[0139] Among them, ray tracing model refers to a model that generates perceptual data by simulating the propagation path of light, optical rendering model refers to a model that generates perceptual data by simulating the optical imaging process, 3D point cloud data refers to perceptual data composed of a large number of 3D spatial points, and simulated image data refers to image data obtained by simulating camera imaging.

[0140] The SIL platform calls a ray tracing model in the simulation environment to generate 3D point cloud data. In some examples, when the SIL platform calls a ray tracing model to generate 3D point cloud data, the ray tracing model can correspond to a 128-line lidar mounted on a mining vehicle. The ray tracing model generates 3D point cloud data reflecting the spatial distribution of obstacles based on spatial geometric occlusion relationships by simulating the propagation and reflection process of the laser beam in the mining environment.

[0141] The SIL platform calls the ray tracing model to generate 3D point cloud data, simulating the ranging process of lidar, so that the 3D point cloud data can reflect the real perception of the mining area environment by the sensors of mining vehicles.

[0142] The SIL platform calls optical rendering models in the simulation environment to generate simulated image data. In some examples, when the SIL platform calls optical rendering models to generate simulated image data, the optical rendering model can correspond to the fisheye camera mounted on a mining vehicle. The optical rendering model generates simulated image data reflecting the mining scene based on the light attenuation relationship by simulating the lighting and imaging process of the mining environment.

[0143] The SIL platform calls the optical rendering model to generate simulated image data, simulating the imaging process of the camera, so that the simulated image data can reflect the camera's real imaging of the mining area scene.

[0144] In some embodiments, when the SIL platform generates 3D point cloud data, it can generate 3D point cloud data through spatial geometric occlusion calculation; in other embodiments, the SIL platform can also generate 3D point cloud data through depth cache sampling.

[0145] In some embodiments, when the SIL platform generates simulated image data, it can generate simulated image data through illumination attenuation simulation; in other embodiments, when the SIL platform generates simulated image data, it can also generate simulated image data through rasterization rendering.

[0146] The SIL platform generates 3D point cloud data and simulated image data through ray tracing models and optical rendering models, respectively, enabling virtual sensing data to simultaneously possess both laser point cloud and visual image sensing content. This improves the completeness and realism of virtual sensing data in perceiving the mining environment, allowing the control model under test to complete software-in-the-loop verification under conditions close to real mining area perception.

[0147] Optionally, in some embodiments, a mining truck dynamics model is invoked, which, in response to control commands, generates operating status data of the simulated vehicle under test. This includes: invoking the mining truck dynamics model, which generates operating status data based on control commands and at least one of the following factors: the mining truck's full-load mass parameters, unloaded mass parameters, tire side slip characteristics, and brake pressure delay parameters.

[0148] In the above embodiments, when the SIL platform calls the mine car dynamics model, it further limits the mine car dynamics model to generate operating status data based on control commands and at least one of the following factors: the mine car's full-load mass parameters, unloaded mass parameters, tire side slip characteristics, and brake air pressure delay parameters.

[0149] Among them, the fully loaded mass parameter refers to the mass parameter of the mine car when it is fully loaded, and the unloaded mass parameter refers to the mass parameter of the mine car when it is unloaded. Tire lateral slip characteristic refers to the characteristic of the mine car tires to deform laterally under the action of lateral force, and brake pressure delay parameter refers to the parameter that there is a delay in the establishment of brake pressure in the mine car's air pressure braking system.

[0150] The SIL platform calls the mining truck dynamics model, which responds to control commands and calculates the motion response of the simulated vehicle under test based on at least one of the following factors: fully loaded mass parameters, unloaded mass parameters, tire side slip characteristics, and brake pressure delay parameters, generating the operating status data of the simulated vehicle under test.

[0151] In some embodiments, when the mining truck dynamics model generates operating status data, the fully loaded mass parameter and the unloaded mass parameter respectively reflect the mass difference of the mining truck under fully loaded and unloaded conditions, the tire lateral slip characteristics can be reflected as nonlinear tire lateral slip force characteristics, and the brake air pressure delay parameter can be reflected as the air pressure brake pipeline response time delay parameter.

[0152] In some embodiments, the mine car dynamics model can generate operating status data through real-time iterative calculation based on control commands and full-load mass parameters, unloaded mass parameters, tire lateral slip characteristics, and brake pressure delay parameters. The operating status data may include mine car speed, yaw rate, and longitudinal slip ratio.

[0153] In some embodiments, the mining truck dynamics model can also generate operational status data through numerical integration.

[0154] The SIL platform generates operating status data based on control commands and parameters of full load, unloaded mass, tire lateral slip characteristics, and brake pressure delay through a mining truck dynamics model. This allows the operating status data to reflect the actual motion response of the mining truck under different load and braking conditions, improving the consistency between the operating status data and the actual vehicle behavior, and also enhancing the reliability of the motion simulation results of the vehicle under test.

[0155] like Figure 5 As shown, optionally, in some embodiments, the defect features of the control model under test are located based on the multidimensional evaluation matrix, including:

[0156] Step 502: Determine the abnormal time series interval that triggers the preset risk threshold based on the multidimensional evaluation matrix;

[0157] Step 504: Based on the abnormal time interval, extract the target perception data and target control instructions corresponding to the abnormal time interval.

[0158] Step 506: Generate a problem scenario model based on the target perception data and target control commands;

[0159] Step 508: Reproduce the abnormal situation based on the problem scenario model;

[0160] Step 510: Based on the abnormal situation, locate the defect characteristics.

[0161] In the above embodiments, when the SIL platform locates the defect features of the control model under test based on the multidimensional evaluation matrix, it is further defined as the process of determining the abnormal time interval that triggers the preset risk threshold, intercepting the target perception data and target control commands, generating a problem scenario model, reproducing the abnormal situation, and locating the defect features based on the abnormal situation.

[0162] Among them, the preset risk threshold refers to the threshold used to determine whether the risk exceeds the standard, the abnormal time series interval refers to the time interval in the multi-dimensional evaluation matrix that triggers the preset risk threshold, the target perception data refers to the perception data corresponding to the abnormal time series interval, the target control instruction refers to the control instruction corresponding to the abnormal time series interval, the problem scenario model refers to the model used to describe the problem scenario, and the abnormal situation refers to the abnormal performance situation reproduced according to the problem scenario model.

[0163] The SIL platform determines the abnormal time series intervals that trigger preset risk thresholds based on a multi-dimensional evaluation matrix. In other words, the SIL platform searches the multi-dimensional evaluation matrix for evaluation items that trigger preset risk thresholds, and identifies the time periods corresponding to these items as the abnormal time series intervals.

[0164] The SIL platform extracts target perception data and target control commands corresponding to abnormal time intervals based on these intervals. In other words, the SIL platform extracts perception data and control commands within the time period covered by the abnormal time interval from the data recorded during the simulation process to obtain the target perception data and target control commands.

[0165] The SIL platform generates problem scenario models based on target perception data and target control commands. In other words, the SIL platform organizes target perception data and target control commands into replayable problem scenario models, and can also reconstruct target perception data and target control commands into abnormal operating condition scenario slice models.

[0166] The SIL platform reproduces abnormal situations based on the problem scenario model. In some examples, when reproducing abnormal situations based on the problem scenario model, the SIL platform can replay the target perception data and target control commands to the Log2World (Log to World) module, and then reproduce the abnormal situations through the Log2World module.

[0167] The SIL platform locates defect characteristics based on abnormal conditions. In other words, when the SIL platform locates defect characteristics based on abnormal conditions, it can locate defect characteristics by analyzing the abnormal conditions. The defect characteristics can be reflected in the model hyperparameter defect characteristics of the control model under test.

[0168] The SIL platform improves the targeting of defect localization by determining the abnormal time interval, capturing target perception data and target control commands, generating problem scenario models, reproducing abnormal conditions, and locating defect features based on abnormal conditions. This limits the defect localization process to the time period in which the problem occurs, avoids blindly searching for defects in the full simulation data, and avoids repeatedly executing the complete simulation to reproduce occasional failures.

[0169] Optionally, in some embodiments, obtaining an updated control model under test based on defect characteristics includes: calling a large language model, modifying the control model under test according to the defect characteristics, and obtaining the updated control model under test.

[0170] In the above embodiments, when the SIL platform obtains the updated control model under test based on defect characteristics, it is further limited to calling a large language model, modifying the control model under test according to the defect characteristics, and obtaining the updated control model under test. Here, a large language model (LLM) refers to an artificial intelligence model with a large number of parameters that can generate code or adjust content based on input.

[0171] The SIL platform calls the large language model, inputs the defect features into the large language model, the large language model generates the modification content of the control model under test based on the defect features, the SIL platform modifies the control model under test according to the modification content, and obtains the updated control model under test.

[0172] In some embodiments, when the SIL platform calls a large language model to modify the control model under test, the large language model can be a pre-trained code-generated large language model, which can generate the modification content of the control model under test based on defect characteristics; in other embodiments, the large language model can also be a published code-generated large language model, which can generate the modification content of the control model under test based on defect characteristics.

[0173] In some embodiments, modifications to the control model under test may include adjusting model structure parameters, loss function penalty terms, network weights, and control logic.

[0174] In some embodiments, the SIL platform can also update the network weights of the control model under test through a large language model; in other embodiments, the SIL platform can also reconstruct the control logic of the control model under test through a large language model.

[0175] The SIL platform automates the defect repair process by calling a large language model to modify the control model under test based on defect characteristics. This improves the efficiency of algorithm iteration, avoids the time overhead of manual code modification, and enhances the connection efficiency between defect repair and simulation verification.

[0176] like Figure 6 As shown, optionally, in some embodiments, the preset quantitative evaluation indicators include minimum safe distance, collision time, average traffic speed, and emergency braking distance. Based on these preset quantitative evaluation indicators, the virtual perception data and operational status data are evaluated to generate a multi-dimensional evaluation matrix, including:

[0177] Step 602: Evaluate the virtual sensing data based on the minimum safe distance to obtain the minimum safe distance evaluation value;

[0178] Step 604: Evaluate the virtual sensing data and the operating status data based on the collision time to obtain the collision time evaluation value;

[0179] Step 606: Evaluate the operating status data based on the average traffic speed to obtain the average traffic speed evaluation value;

[0180] Step 608: Evaluate the operating status data based on the emergency braking distance to obtain the emergency braking distance evaluation value;

[0181] Step 610: Generate a multi-dimensional evaluation matrix based on the minimum safe distance assessment value, collision time assessment value, average traffic speed assessment value, and emergency braking distance assessment value.

[0182] In the above embodiments, when the SIL platform evaluates the virtual sensing data and operational status data according to preset quantitative evaluation indicators, the preset quantitative evaluation indicators are further defined as including minimum safe distance, collision time, average traffic speed, and emergency braking distance. The virtual sensing data and operational status data are evaluated based on the minimum safe distance, collision time, average traffic speed, and emergency braking distance respectively, resulting in minimum safe distance evaluation values, collision time evaluation values, average traffic speed evaluation values, and emergency braking distance evaluation values. A multi-dimensional evaluation matrix is ​​then generated based on these evaluation values.

[0183] Among them, minimum safe distance refers to the minimum distance between the simulated vehicle under test and obstacles or environmental interaction vehicles; collision time refers to the estimated time required for the simulated vehicle under test to collide with an obstacle while it is in its current motion state; average traffic speed refers to the average speed of the simulated vehicle under test during the simulation operation; emergency braking distance refers to the distance traveled by the simulated vehicle under test from triggering braking to coming to a complete stop; and evaluation value refers to the value obtained by evaluating the virtual perception data and operating status data based on quantitative evaluation indicators.

[0184] The SIL platform evaluates virtual sensing data based on the minimum safe distance to obtain a minimum safe distance assessment value. In some examples, when evaluating virtual sensing data based on the minimum safe distance, the SIL platform can determine the spatial boundary between obstacles and the simulated vehicle under test from the virtual sensing data, and calculate the minimum safe distance assessment value based on the spatial boundary.

[0185] The SIL platform evaluates virtual sensing data and operational status data based on the collision time to obtain a collision time assessment value. In some examples, when evaluating virtual sensing data and operational status data based on collision time, the SIL platform can combine the relative distance in the virtual sensing data and the relative acceleration in the operational status data to calculate the collision time assessment value.

[0186] The SIL platform evaluates operational status data based on average traffic speed to obtain an average traffic speed assessment value. In some examples, when evaluating operational status data based on average traffic speed, the SIL platform can statistically analyze the speed time-series curves in the operational status data to calculate the average traffic speed assessment value.

[0187] The SIL platform evaluates operational status data based on emergency braking distance to obtain an emergency braking distance assessment value. In some examples, when evaluating operational status data based on emergency braking distance, the SIL platform can calculate the spatial displacement between the braking control command trigger point and the vehicle speed zeroing point to obtain the emergency braking distance assessment value.

[0188] The SIL platform generates a multi-dimensional evaluation matrix based on minimum safe distance assessment values, collision time assessment values, average traffic speed assessment values, and emergency braking distance assessment values. In other words, the SIL platform generates this multi-dimensional evaluation matrix by aggregating these assessment values ​​to form a multi-dimensional evaluation matrix that assesses simulation results from multiple dimensions.

[0189] The SIL platform evaluates virtual perception data and operational status data from four dimensions: minimum safe distance, collision time, average traffic speed, and emergency braking distance, and generates a multi-dimensional evaluation matrix. This expands the evaluation of simulation results from a single dimension to multiple dimensions, improving the comprehensiveness of the evaluation and the accuracy of subsequent defect location.

[0190] In some embodiments, the simulation testing method for unmanned mining trucks is executed by a vehicle-cloud integrated simulation testing platform. The simulation testing platform adopts a four-layer architecture, including a computing resource layer, a simulation system layer, a scheduling system layer, and a test closed-loop layer, which facilitates independent development, independent deployment, and independent upgrades.

[0191] The computing resource layer is based on GPU / high-performance computing clusters and Kubernetes containerization platform, providing elastic container services, high-speed caching services, object storage and GPU computing power pools, supporting distributed simulation and multi-GPU parallel rendering, and realizing elastic scaling of computing power and resource isolation.

[0192] The simulation system layer integrates the OpenDrive / OpenScenario editor, 3D mine scene generation, high-precision dynamic model of heavy truck, and physical sensor simulation, supporting one-click generation of mine scenes and real-time Web3D preview.

[0193] The scheduling system layer enables concurrent simulation of multiple master vehicles and distributed rendering of multiple GPUs through containerized task orchestration. It provides scheduling protocols, V2X (Vehicle to Everything) protocols, algorithm access interfaces, timing control capabilities, and perception and localization algorithms. It receives instructions from the scheduling cloud platform to achieve real-time simulation, long-term single-vehicle simulation, long-term multi-vehicle simulation, and dynamic vehicle joining and leaving, ensuring stable operation of vehicle-cloud collaboration and large-scale clusters. The testing closed-loop layer provides automatic generation of test reports, test replay analysis, regression testing and CI / CD (Continuous Integration / Continuous Delivery) linkage, and hazardous scenario write-back to the scenario library, forming a closed loop of continuous enrichment of the scenario library and continuous algorithm evolution.

[0194] In some embodiments, after completing the simulation test, the SIL platform continuously writes back the effective hazardous scenarios or risky working conditions generated during the simulation test to the scenario library and the high-precision map library, so that the scenario library and the high-precision map library are continuously enriched as the simulation test continues, providing reusable scenarios for subsequent simulation tests.

[0195] In some embodiments, the SIL platform advances the simulation according to a unified simulation clock step during simulation execution and synchronizes the vehicle-cloud timing, enabling each simulated vehicle entity and the cloud platform to operate collaboratively under a unified timing reference.

[0196] In some embodiments, the SIL platform also simulates mining truck driving, excavator operation, and V2X communication interaction, enabling the simulation environment to recreate the real-world scenario of multi-vehicle collaborative operation in a mining area.

[0197] In some embodiments, the simulation scene may include obstacle categories such as mining trucks, cones, human obstacles, cattle, horses, sheep, and water pits.

[0198] In some embodiments, testers can create simulation scenarios through a scenario editor, define vehicle paths, oncoming traffic triggering conditions and avoidance logic, and store the created simulation scenarios in a scenario library for easy retrieval and reuse.

[0199] In some embodiments, the simulation testing platform can support the simultaneous operation of dozens of mining trucks, breaking through the limitations of traditional simulation software that operates on a single machine and a single truck.

[0200] In some embodiments, the SIL platform allocates corresponding simulation containers, GPUs, and CPUs to each mining truck participating in the simulation, completes port mapping and network communication, and starts a distributed simulation cluster.

[0201] In some embodiments, the SIL platform performs format conversion on the abnormal data, performs scene mining and cleaning on the converted data, and stores the scene-mined and cleaned data in the scene library in the OpenScenario standard format.

[0202] In some embodiments, the SIL platform generates editable scenario files when reproducing abnormal conditions, which can be edited and reused later. In some embodiments, testers can also import scenarios converted from Log2World from the data closed-loop platform for simulation testing.

[0203] In some embodiments, the regression test result can be a comparison result obtained by comparing the simulation result corresponding to the updated control model under test with the simulation result corresponding to the control model under test, which is used to verify whether the defect features have been repaired.

[0204] A second aspect of the present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the simulation testing method for the unmanned mining truck in any of the above embodiments. Therefore, this storage medium possesses all the beneficial effects of the simulation testing method for the unmanned mining truck in any of the above embodiments, which will not be elaborated further here.

[0205] Storage media include read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0206] like Figure 7 As shown, the present invention also proposes an electronic device 70, including a memory 702, a processor 704, and a computer program stored in the memory 702 and executable on the processor 704. When the processor 704 executes the computer program, it implements the steps of the simulation testing method for the unmanned mining truck in any of the above embodiments. Therefore, the electronic device 70 possesses all the beneficial effects of the simulation testing method for the unmanned mining truck in any of the above embodiments, which will not be elaborated further here.

[0207] like Figure 8 As shown, this invention also proposes a simulation testing system 80 for an unmanned mining truck, comprising: a first acquisition module 802, used to acquire abnormal data and a control model under test during the operation of the mining truck; a parsing module 804, used to parse the abnormal data based on operating condition characteristics to generate structured vehicle data and a spatial scene model; a first generation module 806, used to generate a simulation vehicle entity based on the structured vehicle data, the simulation vehicle entity including an environmental interaction vehicle and a simulation vehicle under test; an environment construction module 808, used to construct a simulation environment based on the simulation vehicle entity and the spatial scene model; a second generation module 810, used to call the sensor physical model in the simulation environment to generate virtual perception data of the simulation vehicle under test; and an output module 812, used to input the virtual perception data into the control model under test and output control data. The system comprises the following modules: a control command module 814, an execution module 814, and a generation module 816. The third generation module 816 calls the mine car dynamics model, which, in response to the control command, generates the operating status data of the simulated vehicle under test. The fourth generation module 818 evaluates the virtual perception data and operating status data according to preset quantitative evaluation indicators, generating a multi-dimensional evaluation matrix. The positioning module 820 locates the defect features of the control model under test based on the multi-dimensional evaluation matrix. The second acquisition module 822 acquires the updated control model under test based on the defect features. The regression test module 824 re-executes the simulation operation on the simulated vehicle under test based on the updated control model under test and the spatial scene model, obtaining regression test results.

[0208] The present invention also proposes a simulation test system 80 for unmanned mining trucks. Each module in the simulation test system 80 corresponds one-to-one with each step of the above-mentioned simulation test method for unmanned mining trucks, and is used to implement all the steps of the above-mentioned simulation test method for unmanned mining trucks.

[0209] Since the above-mentioned unmanned mining truck simulation test system 80 adopts the same technical means as the above-mentioned unmanned mining truck simulation test method, the above system has the same beneficial effects as the above-mentioned unmanned mining truck simulation test method, which will not be elaborated here.

[0210] In the claims, description, and accompanying drawings of this invention, the term "plural" refers to two or more. Unless otherwise explicitly defined, the terms "upper," "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and simplifying the descriptive process, and are not intended to indicate or imply that the device or element referred to must have the described specific orientation, or be constructed and operated in a specific orientation. Therefore, these descriptions should not be construed as limiting the invention. The terms "connected," "installed," "fixed," etc., should be interpreted broadly. For example, "connected" can be a fixed connection between multiple objects, a detachable connection between multiple objects, or an integral connection; it can be a direct connection between multiple objects or an indirect connection between multiple objects through an intermediate medium. For those skilled in the art, the specific meaning of the above terms in this invention can be understood based on the specific circumstances described above.

[0211] In the claims, description, and accompanying drawings of this invention, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In the claims, description, and accompanying drawings of this invention, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0212] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A simulation testing method for an unmanned mining truck, characterized in that, include: Acquire abnormal data and control models under test during the operation of mining trucks; Based on the operating condition characteristics, the abnormal data is analyzed to generate structured vehicle data and spatial scene models; Based on the structured vehicle data, a simulated vehicle entity is generated, which includes an environmental interaction vehicle and a simulated vehicle under test. Based on the simulated vehicle entity and the spatial scene model, a simulation environment is constructed; In the simulation environment, the sensor physical model is invoked to generate virtual perception data of the simulated vehicle under test; The virtual sensing data is input into the control model under test, and control commands are output. Based on the control commands and simulation execution configuration, the vehicle under test is controlled to perform simulation operations; The mining truck dynamics model is invoked, and the mining truck dynamics model responds to the control command to generate the operating status data of the simulated vehicle under test; The virtual sensing data and the operating status data are evaluated based on preset quantitative evaluation indicators to generate a multi-dimensional evaluation matrix. Based on the multidimensional evaluation matrix, the defect characteristics of the control model under test are located; Based on the defect characteristics, obtain the updated control model to be tested; Based on the updated control model under test and the spatial scene model, the simulation operation is re-executed on the simulated vehicle under test to obtain regression test results.

2. The simulation testing method for unmanned mining trucks according to claim 1, characterized in that, The step of parsing the abnormal data based on operating condition characteristics to generate structured vehicle data and a spatial scene model includes: The abnormal data is converted to a standard format to generate standard format data; Vehicle information is extracted from the standard format data to generate the structured vehicle data; Scene information is extracted from the standard format data to construct the spatial scene model.

3. The simulation testing method for unmanned mining trucks according to claim 1, characterized in that, Before constructing the simulation environment based on the simulated vehicle entity and the spatial scene model, the simulation testing method for the unmanned mining truck further includes: Obtain test constraint parameters, including time-series period parameters, vehicle density parameters, and vehicle heavy load parameters; The simulation execution configuration is generated based on the test constraint parameters and the spatial scene model.

4. The simulation testing method for unmanned mining trucks according to claim 1, characterized in that, The step of constructing a simulation environment based on the simulated vehicle entity and the spatial scene model includes: Based on the aforementioned spatial scene model, a three-dimensional road network is generated; Each of the simulated vehicle entities is mapped to the corresponding coordinate nodes in the three-dimensional road network to generate the simulation environment.

5. The simulation testing method for unmanned mining trucks according to claim 1, characterized in that, The sensor physical model includes a ray tracing model and / or an optical rendering model; the virtual perception data includes 3D point cloud data and simulation image data; and in the simulation environment, the sensor physical model is invoked to generate virtual perception data for the simulated vehicle under test, including: In the simulation environment, the ray tracing model is invoked to generate the 3D point cloud data; In the simulation environment, the optical rendering model is invoked to generate the simulation image data.

6. The simulation testing method for unmanned mining trucks according to claim 1, characterized in that, The process of calling the mining truck dynamics model, which responds to the control command, generates the operating state data of the simulated vehicle under test, including: The mining truck dynamics model is invoked, and the mining truck dynamics model generates the operating status data based on the control command and at least one of the following factors: the full-load mass parameter, the unload mass parameter, the tire side slip characteristics, and the brake air pressure delay parameter.

7. The simulation testing method for unmanned mining trucks according to any one of claims 1 to 6, characterized in that, The step of locating the defect features of the control model under test based on the multidimensional evaluation matrix includes: Based on the multidimensional evaluation matrix, the abnormal time interval that triggers the preset risk threshold is determined; Based on the abnormal time interval, extract the target perception data corresponding to the abnormal time interval and the target control command corresponding to the abnormal time interval; Based on the target perception data and the target control commands, a problem scenario model is generated; Based on the problem scenario model, reproduce the abnormal situation; Based on the aforementioned abnormal situation, the defect characteristics are located.

8. The simulation testing method for unmanned mining trucks according to any one of claims 1 to 6, characterized in that, The step of obtaining the updated control model to be tested based on the defect characteristics includes: The large language model is invoked, and the control model under test is modified according to the defect characteristics to obtain the updated control model under test.

9. The simulation testing method for unmanned mining trucks according to any one of claims 1 to 6, characterized in that, The preset quantitative evaluation indicators include minimum safe distance, collision time, average traffic speed, and emergency braking distance. The virtual sensing data and the operational status data are evaluated based on these preset quantitative evaluation indicators to generate a multi-dimensional evaluation matrix, including: The virtual sensing data is evaluated based on the minimum safe distance to obtain a minimum safe distance evaluation value; Based on the collision time, the virtual sensing data and the operating status data are evaluated to obtain a collision time evaluation value; The operating status data is evaluated based on the average traffic speed to obtain an average traffic speed evaluation value. Based on the emergency braking distance, the operating status data is evaluated to obtain an emergency braking distance evaluation value; The multidimensional evaluation matrix is ​​generated based on the minimum safe distance assessment value, the collision time assessment value, the average traffic speed assessment value, and the emergency braking distance assessment value.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the simulation test method for the unmanned mining truck as described in any one of claims 1 to 9.

Images