A multi-lidar redundant perception system and control method for unmanned mine cars

Abstract

This invention relates to the field of environmental perception for unmanned vehicles, and discloses a multi-LiDAR redundant perception system and control method for unmanned mining trucks. The method simultaneously acquires the original point clouds and received signal strength indication values ​​of multiple isomorphic LiDARs within a control cycle, as well as the temperature and dust concentration of the external environment. It quantifies the real-time health of each LiDAR through a built-in evaluation model and performs dual degradation judgment based on the time-domain derivative of the health score. When a radar is determined to have a degradation trend, the system simultaneously initiates predictive active weight reduction and spatial field-of-view compensation in both the time and spatial domains. Furthermore, the method nonlinearly maps dust concentration to an environmental adjustment factor to dynamically modulate the fusion weights and introduces a hard isolation mechanism to eliminate extremely degraded sources, outputting a globally fused point cloud. This effectively solves the trajectory jump problem caused by the performance degradation of perception sources under harsh operating conditions, achieving high-confidence data focusing and smooth switching, and improving the driving safety of unmanned mining trucks in extreme environments.

Description

Technical Field

[0001] This invention relates to the field of environmental perception for unmanned vehicles, and in particular to a multi-LiDAR redundant perception system and control method for unmanned mining trucks. Background Technology

[0002] As autonomous driving technology continues to penetrate vertical industries, large driverless mining trucks are increasingly being used in enclosed or semi-enclosed environments such as open-pit mines and bulk cargo ports. The working environment in mining areas is extremely complex and harsh, with vehicles constantly facing challenges such as dust, mud and water splashes, severe bumps, and significant temperature differences between day and night. Under these extreme conditions, environmental perception systems based on single sensors are prone to failure due to physical obstruction or temperature drift of components, leading to blind spots and even safety accidents.

[0003] To address the issue of insufficient reliability of single sensors, existing unmanned mining trucks typically employ a multi-sensor redundant perception architecture, deploying multiple LiDARs at different locations on the vehicle for data fusion. However, traditional multi-LiDAR data fusion methods generally adopt a static fixed weight allocation strategy. Firstly, existing technologies suffer from severe spatial weight passivation defects. In real harsh environments, the spatial distribution of interference factors such as dust, sandstorms, rain, and snow is usually uneven, and the degree of radar performance degradation varies significantly at different physical installation locations. Traditional static weight systems or simple dynamic adjustment mechanisms cannot establish a sufficient confidence gap between high-quality perception data and contaminated perception data during the transition period of gradual environmental changes, resulting in the global perception point cloud being contaminated by data from some degraded sensors.

[0004] Secondly, existing technologies suffer from a passive time-domain lag defect in their fault-tolerance mechanisms. Traditional systems often rely on underlying hardware fault codes or complete signal loss before triggering the switching of sensing sources or the redistribution of weights. This feedback-based passive response mechanism causes the system to continuously output biased sensing data during the transition period from sensor performance degradation to eventual confirmation of a fault and removal. When the system finally completes the switch, it often leads to a step fluctuation in the output fused sensing results, which in turn causes the vehicle's trajectory to change or to brake suddenly, seriously affecting driving smoothness and safety. Summary of the Invention

[0005] The main objective of this invention is to provide a multi-lidar redundant sensing system and control method for unmanned mining trucks, which solves the problem of trajectory smoothness when switching sensing sources under harsh working conditions, as well as the problem of high-confidence focusing under the performance degradation of multi-source isomorphic sensors.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a multi-laser radar redundancy sensing system and control method for unmanned mining trucks, the method comprising the following steps: S1. In each control cycle, the central domain controller synchronously acquires the raw point cloud data set collected by multiple homogeneous lidars deployed at different locations on the mining truck, the received signal strength indication value (RSSI) of the internal optoelectronic devices of the lidar, and the real-time data of environmental dust concentration and environmental temperature collected by the environmental status sensor components through the vehicle CAN bus. Each lidar unit outputs raw point cloud data at a set frequency. Each point cloud data point contains the three-dimensional spatial coordinates of the target object and intensity information characterizing its reflection characteristics. The environmental sensor component continuously outputs dust concentration values ​​and ambient temperature at a preset frequency, characterizing the degree of air quality deterioration at the work site. S2. The central domain controller has a built-in multi-dimensional health assessment model. This assessment model performs calculations periodically to quantify the current real-time working efficiency of each lidar, eliminating the limitations of a single evaluation index in complex mining environments. S3. During each control cycle, the system monitors the dynamic evolution characteristics of the health score in the time domain dimension and performs a dual health degradation judgment based on the time domain derivative monitoring. S4. When a radar is determined to be deteriorating, the system simultaneously initiates predictive active weight reduction and spatial field-of-view filling in both the time and space domains. S5. The dust concentration is nonlinearly mapped to an environmental adjustment factor through a hyperbolic tangent function. The Softmax weight sensitivity is dynamically modulated to focus on high-quality radar data. When performing high-frequency weighted fusion, a hard isolation threshold is introduced to remove extremely degraded sources, and a high-confidence global fused point cloud matrix is ​​output.

[0007] S6. The processed clean point cloud data is pushed to the downstream autonomous driving software in real time.

[0008] In the preferred embodiment, step S2 specifically includes the following steps: S21. Effective point cloud ratio extraction For any raw point cloud data output by a lidar, the random sample consensus algorithm is used to perform ground plane fitting and segmentation of the point cloud, removing ground point clouds belonging to the roadbed; subsequently, the ratio of the remaining effective non-ground point cloud quantity to the total point cloud quantity is calculated using the following formula: (1); in, For the first The effective point cloud ratio of each lidar; This represents the number of non-ground point clouds remaining after removing ground points using the RANSAC algorithm. This represents the total number of raw point clouds output by the radar during the current acquisition period. As a core factor characterizing the integrity of space exploration, it is input into the subsequent comprehensive score calculation; S22, Point Cloud Noise Level Assessment On the flat surface of the mining area, severe noise interference caused by water mist refraction on the radar dome surface, crosstalk of internal optoelectronic devices, or mechanical vibration leads to an excessively high elevation standard deviation, indicating abnormal dispersion and jitter in the point cloud. For the extracted effective non-ground point cloud data, the coordinate set of all data points in the vertical direction is extracted, and its elevation standard deviation is calculated using the following formula: (2); in, For the first The standard deviation of the point cloud elevation of each lidar is used to quantify the noise level; For the first The radar number The height coordinates of one valid non-ground point; This is the average height coordinate of all valid non-ground points of the radar. This will be input as a penalty item in the subsequent comprehensive score calculation; S23. Signal strength temperature compensation based on anti-false judgment mechanism In open-pit mines, vehicles often face sudden increases in ambient temperature due to intense sunlight. The core receiving components inside the lidar system experience physical gain attenuation due to high temperatures, leading to a decrease in the Received Signal Strength Indication (RSSI). Physical compensation is necessary to prevent the system from mistakenly interpreting this as hardware damage such as severe dust accumulation on the radome surface, thus triggering a deweighting mechanism. Therefore, a pre-calibrated temperature compensation coefficient is introduced to correct the original signal strength, preventing system-level logical misjudgments. The calculation formula is as follows: (3); in, The signal strength value after compensation by the anti-false alarm mechanism; For the first The raw received signal strength indication value output from the bottom layer of the lidar; This is a temperature compensation coefficient pre-calibrated based on the physical attenuation characteristics of optoelectronic devices; The ambient absolute temperature is collected in real time. The preset reference temperature for standard equipment operation; Used to characterize the actual optical transmission capability and input into subsequent scoring calculations; S24. Calculate the overall health score. The central domain controller uses the extracted effective point cloud proportion, noise level, and compensated signal strength to calculate the quantitative health score of each radar in the current period through a preset weighted model. To highlight the contribution of the radar's core spatial detection capability to the evaluation system, the weight coefficient corresponding to the effective point cloud proportion in the model is configured to be greater than the weight coefficients corresponding to the point cloud noise level and received signal strength. The calculation formula is as follows: (4); in, For the first The comprehensive health score of each lidar in the current cycle, normalized to... Within the range; A fixed weight constant representing the contribution of each feature; This is the preset threshold for the maximum tolerance noise standard deviation; This is the calibrated value of the maximum signal strength under ideal conditions; This health score sequence It will be sent to the degradation prediction and weight dynamic allocation steps.

[0009] In the preferred embodiment, in step S3: Dual degradation assessment includes short-term drop detection and trend-based degradation detection, wherein: Short-term drop detection is used to determine whether the score of the current period has dropped sharply beyond a preset threshold compared to the previous period. Trend decay detection involves constructing a sliding time window of a specific length and calculating the first-order time derivative of the health score. The calculation formula is as follows: (5); in, , For the first The first time derivative of the health score of a LiDAR device within the current sliding window; Rate your current health status. A health score is given at the start of the sliding window. This is the set total length of the sliding time window.

[0010] When the first-order time derivative is continuously lower than the set negative warning threshold, the system determines that the target radar has deteriorated and triggers the predictive weighting strategy in step S4; if it is not triggered, the normal working flow is maintained and the process jumps to step 5.

[0011] In the preferred embodiment, step S4 specifically includes the following steps: S41, Time-Domain Linear Weighting The central domain controller sends attenuation commands to the underlying fusion weight matrix, implementing a gradual linear attenuation strategy for the fusion weights of faulty radars with a fixed time period and fixed step size. The calculation formula is as follows: (6); in, To implement the downweighting of the faulty radar at the current moment The weight allocation value; This is the weight value of the radar in the previous downweighting cycle; The pre-set single-cycle weight decay step size; By forcibly lowering the weight variables, the pollution of the global point cloud map by poor data is gradually eliminated; S42, Airspace Dynamic Blind Spot Filling Hardware Linkage In the parallel thread executing weight decay, the system issues instructions to perform hot switching and physical space blinding; First, within a specified time limit, redundant backup radars that are in a dormant or low-power state are forcibly woken up to complete the warm-up. At the same time, the effective sensing field of view lost by the faulty radar due to performance degradation is calculated, and a yaw angle compensation command is generated to drive the adjacent healthy isomorphic radars of the faulty radar to appropriately deflect their field of view toward the faulty area and implement physical field of view reconstruction.

[0012] In the preferred embodiment, to achieve dynamic spatial blind spot filling in step S42, the system of the present invention supports dual hardware implementation paths: Firstly, a mechanical gimbal is installed at the bottom of the adjacent health radar to receive commands from the domain controller, which changes the overall field of view orientation through physical rotation. Secondly, preferably, the radar used in the system is an automotive-grade solid-state or semi-solid-state lidar that supports dynamic configuration of the region of interest. The central domain controller issues low-level electronic scanning control commands and, without changing the physical installation location, adjusts the timing of the laser emission array through software to change the three-dimensional distribution of energy in the effective detection field of view in order to cope with the harsh working conditions of severe vibration of mining trucks.

[0013] In the preferred embodiment, step S5 specifically includes the following steps: S51, Environmental Factor Modulation Because the attenuation characteristics of PM10 particulate matter suspended in the air to near-infrared laser pulse beams exhibit a nonlinear physical attenuation law that deteriorates rapidly after reaching a certain concentration, this invention uses the hyperbolic tangent function to mathematically model and fit this physical phenomenon, and introduces an environmental adjustment factor characterizing the severity of the environment. The calculation formula is: (7); in, The environmental regulation factor is generated through dynamic calculation, and its value increases non-linearly and smoothly with the increase of dust concentration. The measured value of dust concentration at the current work site is collected in real time by environmental sensors. This is a dust baseline reference value based on repeated field measurements and calibrations of unmanned mining trucks under normal dust conditions. The sensitivity calibration parameter for the rate of change of the control function curve; This is the preset amplitude amplification factor; When the ambient air is clean Approaching 1; when in a high-concentration dust storm environment, It smoothly transitions with the hyperbolic tangent curve and converges to a higher extremum; S52 and Softmax nonlinear exponential fusion Modulated environmental factors As an exponential adjustment coefficient, it is introduced into the Softmax probability allocation model to calculate the relative weight of each radar. The calculation formula is as follows: (8); in, For the first time in the current control cycle The dynamic fusion weights are ultimately assigned to each lidar; This refers to the environmental regulation factor obtained from the dynamic mapping of dust concentration; A comprehensive health score; This represents the total number of radars currently in operation within the system. This calculation shows that when the concentration of ambient dust suddenly increases... During amplification, high-health radars will receive significantly increased weights due to nonlinear amplification effects; this enables the system to automatically and rapidly concentrate confidence on undamaged radars when the environment deteriorates, thereby breaking through the insensitivity bottleneck of traditional weights. S53, weighted data fusion with underlying hard isolation as a backup After acquiring the dynamically refreshed fusion weights of each radar, the central domain controller performs spatial registration and weighted summation of the underlying point cloud data at an extremely high frequency to generate global environmental perception data to support downstream decision-making. To prevent extremely degraded anomalous data from polluting the global point cloud, a hard blocking fallback mechanism based on a discard threshold is introduced on top of soft weighted fusion. The calculation formula is as follows: (9); in, The final output of the system is a 3D global environment fusion point cloud matrix; For the first Dynamic weights of individual radars; For the first Raw point cloud data of one radar; A pre-set safe discard hard threshold for the system; During the summation traversal, the domain controller checks the weight variables. If determined If the data transmission through that channel is blocked and discarded, global environmental awareness data will be generated to support downstream decision-making.

[0014] In the preferred embodiment, the output of step S6 includes: It includes a list of obstacles with 3D coordinates, motion vector velocity, and detection confidence; a 3D grid map of passable areas; and a system status diagnostic message containing the health status of each radar and its assigned weight variables.

[0015] In a preferred embodiment, the present invention further provides a multi-LiDAR redundant sensing system for unmanned mining trucks, used to execute the above method. This hardware system includes a surround-view isomorphic LiDAR array, an environmental status sensor assembly, a communication bus, and a central domain controller, wherein: The surround-view isomorphic lidar array uses multiple automotive-grade solid-state or semi-solid-state lidars that support dynamic configuration of regions of interest (ROI), deployed around the mining truck to form 360° coverage without blind spots. In order to achieve dynamic airspace blind spot filling, the lidar array supports two hardware execution paths: a mechanical gimbal at the bottom that receives domain controller commands and a three-dimensional distribution of effective detection field of view energy that is changed without changing the physical installation position, through underlying electronic scanning control commands. The environmental condition sensor assembly is deployed on the exterior of the vehicle and includes a high-precision PM2.5 detector and temperature and humidity sensors to capture external environmental parameters that can lead to sensor degradation. The communication bus adopts a high-reliability vehicle CAN bus or vehicle Ethernet to connect all sensors and the central computing unit, meeting the high real-time requirements of the fusion cycle. The central domain controller has a built-in high-performance computing chip and memory. The memory stores computer programs, and when the processor executes the programs, it can quickly and repeatedly call the complete control logic of the multi-LiDAR redundant sensing method used for unmanned mining trucks.

[0016] This invention provides a multi-lidar redundant perception system and control method for unmanned mining trucks. Compared with existing multi-sensor perception fusion technologies, this solution reduces system-level fault response time by approximately 84% due to predictive linear weighting technology based on the first derivative of health status. Even under extreme conditions of sudden severe failure of a single radar, it successfully controls the vehicle's trajectory tracking deviation during the perception switching transition period to within 5cm, avoiding emergency braking and trajectory jumps caused by perception interruption. Simultaneously, through the nonlinear exponential amplification of the Softmax sensitivity of environmental factors, the system can instantaneously increase the weight of healthy radars by more than 36% in the early stages of dust intrusion, reaching a maximum weight of PM10. In extreme dust environments, the side obstacle detection rate of this invention is reduced by about 77% compared to the traditional weighted scheme, achieving the highest quality data focusing within the isomorphic sensor; and the temperature-compensated anti-misjudgment mechanism built in the underlying logic effectively smooths out RSSI data drift caused by the drastic temperature difference between day and night in the mining area, preventing the system from fatally misjudging the signal attenuation caused by high temperature as radar dust accumulation or damage, and improving the stability of the redundant perception and control system under all-weather operating conditions. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of a multi-liquid radar redundant sensing and control method for unmanned mining trucks according to the present invention; Figure 2 This is a schematic diagram of a multi-laser radar redundancy sensing system for unmanned mining trucks according to the present invention; Figure 3 This is a health score surface diagram in the multi-liquid radar redundant sensing system and control method for unmanned mining trucks of the present invention; Detailed Implementation To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The following embodiments are carried out in a simulated and real vehicle test environment based on real industrial application scenarios, aiming to demonstrate the engineering implementation logic and technical advantages of the multi-LiDAR redundant sensing system and control method of this invention through specific engineering parameters, algorithm details, and experimental data.

[0018] Example 1 The physical scenario of this embodiment is located in the unmanned mining truck transportation operation area of ​​a large open-pit coal mine, which has typical extreme and harsh characteristics. The temperature difference throughout the year ranges from -30℃ in winter to +50℃ under the scorching sun in summer. The mining trucks often experience severe mechanical vibrations when traveling on unpaved roads, and due to the frequent crushing by heavy vehicles, the operation site is shrouded in high concentrations of PM10 dust year-round, with PM10 concentrations consistently exceeding normal levels. Extreme dust storms can reach above; The hardware configuration of the sensing system in this embodiment is as follows: The surround-view isomorphic lidar array uses four Hesai AT128 automotive-grade semi-solid-state lidars. This model of lidar supports electronic scanning and dynamic configuration of internal regions of interest (ROIs). They are deployed in front of, behind, and on the left and right sides of the mining truck, with a uniform physical installation height of 3.5 meters and a depression angle of 15 degrees, forming 360-degree spatial coverage. The environmental condition sensor components are Siemens high-precision explosion-proof dust detector and PT1000 high-precision temperature and humidity sensor. The central domain controller is equipped with an NVIDIA Orin-X main control chip, and the sensors are interconnected with the domain controller via a 1000M automotive Ethernet and CAN-FD communication bus. To overcome multicenter bias, the system first uses the Iterative Closest Point (ICP) algorithm to complete the joint calibration of multiple radars within the calibration interval, and obtains a high-precision extrinsic parameter matrix to strictly offset the systematic spatial coordinate deviation caused by vehicle heterogeneity, installation tolerances and long-term compression deformation of the suspension system.

[0019] like Figure 1-3 As shown, a multi-liquid radar redundant sensing system and control method for unmanned mining trucks are disclosed. The method includes the following steps: In the preferred embodiment, in step S1: the central domain controller, with a set control cycle of 50ms corresponding to a system processing frequency of 20Hz, balances the spatial integrity of point cloud data with the computing power consumption of edge computing units, and synchronously acquires multi-source heterogeneous data through the vehicle communication bus; in the time dimension, the system implements microsecond-level hardware clock synchronization for all acquisition nodes based on the IEEE 1588 Precise Time Protocol (PTP); if, during the acquisition process, individual radar timestamps deviate by more than 5ms due to network jitter, the system will call the built-in uniform kinematic extrapolation algorithm to perform pose compensation for the delayed point cloud frames, ensuring spatiotemporal consistency; the acquired real-time data stream specifically includes the raw point cloud data set output by four Hesai AT128 LiDARs at a physical scanning frequency of 10Hz, the received signal strength indication (RSSI) value fed back by the underlying photodetector, and the real-time environmental dust concentration and environmental temperature values ​​transmitted back at a high frequency of 50Hz by the environmental status sensor.

[0020] In the preferred embodiment, in step S2: the system calls the multi-dimensional health assessment model running on the Orin-X chip GPU unit to perform multi-dimensional in-depth quantification of the real-time working performance of each LiDAR within the current 50ms control cycle; the assessment model processes the data from four radars simultaneously through a parallel computing pipeline to eliminate the limitations of relying solely on hardware alarm codes for judgment.

[0021] In the preferred scheme, in step S21: for any single frame of raw point cloud data output by a lidar, the system first applies voxelized grid filtering to downsample and remove redundant overlapping points, and then calls the Random Sample Consensus (RANSAC) algorithm to perform ground plane fitting and segmentation; In engineering practice, considering the undulating characteristics of unpaved roads in mining areas, the distance tolerance threshold of the RANSAC algorithm is set to 0.3 meters, and the maximum number of iterations is limited to 50. After the ground point removal is completed, the system calculates the effective point cloud ratio according to formula (1) and caches it into the system register as the core factor characterizing the integrity of spatial exploration.

[0022] In the preferred scheme, in step S22: the system further performs point cloud noise level assessment on the extracted effective non-ground point cloud data; because dust in the mining area or water accumulation on the radar dome can easily cause diffuse reflection and refraction of the laser beam, generating a large number of noise points suspended in mid-air, the system extracts the coordinate set of all non-ground data points in the vertical height direction and calculates its elevation standard deviation according to formula (2); on a normal, flat mining road surface, the elevation distribution of effective obstacles has a strong structure; but when there is severe water mist or dust interference, the elevation standard deviation will show disordered abnormal dispersion. After quantification by formula (2), if the calculated standard deviation approaches or even exceeds the set maximum tolerance noise standard deviation threshold, in this embodiment, combined with the typical ore volume distribution, the maximum noise standard deviation threshold is calibrated to 1 meter, then the system determines that the current sensing channel is severely polluted by noise and uses the quantified value as the core penalty item for subsequent calculation.

[0023] In the preferred embodiment, in step S23: the system performs signal strength temperature compensation based on the anti-misjudgment mechanism; this embodiment corrects the device degradation caused by the sudden rise in ambient temperature due to summer sun exposure in open-pit mines. The silicon-based avalanche photodiode used inside the Hesai AT128 will have a significantly reduced carrier collision ionization rate at high temperatures, resulting in physical gain attenuation. If no intervention is made, the system will misjudge it as severe dust accumulation on the radome. Therefore, the system reads the temperature value returned by the PT1000 sensor and corrects it using formula (3); the preset reference temperature for standard operation of this device is calibrated to 25℃, and based on the calibration of the physical attenuation characteristics of the APD device, the temperature compensation coefficient is precisely set to... .

[0024] In the preferred scheme, in step S24: the central domain controller summarizes all the extracted feature factors and calculates the comprehensive health score using formula (4); in terms of the allocation logic of weight parameters, to ensure the obstacle avoidance safety of the autonomous driving chassis, the model prioritizes the integrity of spatial detection. Therefore, the weight constants in formula (4) are configured as an asymmetric structure, that is, the weight constant representing the effective point cloud ratio is set to 0.4, and the weight constant representing the elevation standard deviation and the compensated signal strength is set to 0.3. For example, when the effective point cloud ratio of the right radar is 0.8 and the noise standard deviation is 0.2 meters, that is, relative to the maximum threshold of 1.0 meters, the penalty term is The ratio of the compensated RSSI to its calibrated maximum value is 0.9. Substituting this into formula (4), the current health score is 0.83. This normalized score is recorded by the system with a timestamp and used for subsequent trend prediction.

[0025] In the preferred scheme, in step S3: the system enables dual degradation judgment in parallel in the time domain dimension. In the short-term sudden drop detection subroutine, the system compares the health scores of adjacent cycles in 50ms increments. The preset sudden drop threshold is set to 0.15. When the score drops below this threshold, the system will trigger a hardware cleaning command and an instant weight reduction mechanism. In the trend decay detection subroutine, a sliding time window of 500ms is constructed, and the difference between the current time and the starting time of the sliding window 500ms ago is continuously extracted. The first time derivative of the health score is obtained by formula (5). When the time derivative is lower than the set negative warning threshold for three consecutive cycles, the health score is further analyzed. When it is determined that the target radar has experienced irreversible physical degradation, such as the slow accumulation of dust on the mirror surface, step S4 is initiated.

[0026] In the preferred embodiment, in step S4: when a specific radar degradation trigger command transmitted by S3 is received, the system simultaneously initiates predictive active intervention in both the time and spatial domains.

[0027] In the preferred scheme, in step S41: the central domain controller sends a high-priority attenuation command to the fusion weight matrix in the underlying memory to implement a progressive linear attenuation strategy for the faulty radar; in order to avoid the instantaneous gap in the point cloud map caused by hard switching and the tracking divergence of the vehicle control system, the single-cycle weight attenuation step size is set to 0.2 and the attenuation execution cycle is 100ms according to formula (6), and its weight will be smoothly transitioned to complete isolation within a certain period of time; during this period, the influence factor of inferior point cloud data in the global coordinate system is gradually diluted through the forced intervention of mathematical mechanism.

[0028] In the preferred embodiment, in step S42: within the delay of executing the aforementioned weight attenuation thread, the airspace dynamic blind spot filling process is activated; within 30ms of determining the main radar degradation, the system wakes up the roof-mounted redundant high-beam backup radar in low-power sleep mode and completes laser tube preheating; simultaneously, the system calculates the blind zone volume lost by the faulty radar due to attenuation and generates a yaw angle compensation matrix; for the Hesai AT128 semi-solid-state lidar used in this embodiment, path 2 in the invention description is adopted for reconstruction: The domain controller sends an electronic scanning timing reconfiguration command to the health proximity radar via the vehicle Ethernet. Without physical rotation of the mechanical pan-tilt unit, the internal transmission chipset of the health proximity radar adjusts the delay to deflect and concentrate the originally uniformly distributed laser energy array 5 degrees toward the fault side. Through three-dimensional energy distribution reshaping at the pure software level, the cross-coverage compensation for the fault blind zone is completed within 10ms, preventing the mechanical failure risk that traditional mechanical pan-tilt units are prone to damage during severe shaking in the mining area.

[0029] In the preferred scheme, in step S5: for other radar clusters that have not triggered the weight reduction penalty and are in normal working flow, the system performs adaptive weight allocation and performs hard isolation fallback at the fusion end.

[0030] In the preferred scheme, in step S51: the system maps the collected physical dust concentration to a mathematical hyperparameter; in the engineering logic, the Mie scattering effect of PM10 dust on 905nm wavelength near-infrared laser has obvious saturation characteristics, that is, the attenuation is slow when the concentration is low, it attenuates rapidly after reaching a certain concentration, and then the attenuation rate gradually tends to flatten at extremely high concentrations; in order to perfectly adapt to this physical attenuation law, the hyperbolic tangent function is selected for mathematical modeling through formula (7), and the dust reference value is set as This represents the good air quality after water spraying to suppress dust in the mining area. The sensitivity calibration parameter is set to 30, and the amplitude amplification factor is set to 0.5. When dust storms occur at the work site, and the dust concentration rises to [a certain value], [the text continues with further information about dust concentration levels]. When the value is substituted into formula (7) for calculation, the output value is approximately equal to 1. At this time, the environmental adjustment factor is dynamically generated and amplified, and the physical quantity of severe weather is converted into the sensitivity index in the subsequent control algorithm.

[0031] In the preferred scheme, in step S52: the system introduces the high-value environmental adjustment factor as an exponential coefficient into the Softmax probability allocation model according to formula (8); assuming that the forward radar is less affected by dust due to the leeward side, the health score is 0.9; while the left radar is on the windward side and is more affected by dust, the health score is 0.5. Under the conventional arithmetic mean, the difference in confidence between the two is not significant; however, under the environmental factor modulation formula (8) introduced in this invention, the numerator of the forward radar becomes 3.86, and the numerator of the left radar becomes 2.12. After normalization, the forward radar obtains an absolute high weight of about 0.64, while the left radar is suppressed to 0.35; so that when the environment deteriorates, the system will rapidly concentrate the confidence on the high-quality radar data that is undamaged or less damaged, breaking through the insensitivity bottleneck of traditional linear weights.

[0032] In the preferred scheme, in step S53: the central domain controller performs spatial registration and weighted summation of the underlying point cloud data at a high frequency. Through formula (9), the system has a safe discard hard threshold during the traversal process of performing matrix summation. In this embodiment, it is set to 0.05. The underlying memory controller checks the dynamic fusion weight of each channel one by one. When the weight of a certain channel is lower than 0.05 after weight reduction, the channel is blocked to physically isolate the ghost point cloud caused by the degradation source. Finally, a clean and high-confidence three-dimensional global environment fusion point cloud matrix is ​​output.

[0033] In the preferred scheme, in step S6: the system pushes the processed global fused point cloud data to the autonomous driving planning and control middleware inside the mining truck in real time; the output data structure package includes the three-dimensional coordinates, motion vector velocity, obstacle list with detection confidence detected by the PointPillars deep learning network, and a three-dimensional grid map of the passable area after rasterization. To match the high-dimensional operation and maintenance needs of modern industry, the system also simultaneously outputs an interpretable system status diagnostic message. This interpretable report is generated in JSON format and records the feature contribution degree and its value that affect the weight determination within the current control cycle, for example: "The right radar is..." Time weight reduced to The cause was traced back to a surge in the standard deviation of elevation over a continuous 500ms, which led to the first time derivative reaching a certain value. "Triggering penalties" provides precise data support for predictive maintenance of equipment in mine dispatch centers.

[0034] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A multi-liquid radar redundant sensing and control method for unmanned mining trucks, characterized in that: The method includes the following steps: S1. During the control cycle, simultaneously acquire the raw point cloud data set collected by multiple isomorphic lidars, the received signal strength indication value, and the environmental dust concentration and ambient temperature collected by the environmental status sensor components. S2. The built-in multi-dimensional health assessment model quantifies the current working efficiency of each lidar. S3. Monitor the dynamic evolution characteristics of health scores in the time domain dimension and perform a dual health degradation judgment based on time domain derivative monitoring; S4. When it is determined that any lidar has a degradation trend, predictive active weight reduction and spatial field of view blind spot filling are simultaneously activated in both the time domain and the spatial domain. S5. The environmental dust concentration is nonlinearly mapped to an environmental adjustment factor, the weight sensitivity is dynamically modulated to focus on high-quality radar data, and a hard isolation threshold is introduced to remove extremely degraded sources when performing weighted fusion, and a global fused point cloud matrix is ​​output. S6. Push the processed global fusion point cloud data to the downstream autonomous driving software in real time.

2. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. Perform ground plane fitting and segmentation on the raw point cloud data output by any lidar to remove ground point clouds, and calculate the ratio of the remaining effective non-ground point cloud quantity to the total point cloud quantity to obtain the effective point cloud ratio. ; S22. For the extracted valid non-ground point cloud data, extract the coordinate set of all data points in the vertical height direction, and calculate its elevation standard deviation. ; S23. Introduce a temperature compensation coefficient pre-calibrated based on a physical mechanism to correct the original signal strength, obtaining the signal strength value after compensation by the anti-false judgment mechanism. ; S24. Extract the effective point cloud ratio Standard deviation of elevation and the compensated signal strength value The quantitative health score of each LiDAR in the current period is obtained by calculating using a preset weighted model. .

3. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 2, characterized in that, In step S24: Calculate the quantitative health score of each LiDAR in the current period. At that time, the weighted model corresponds to the effective point cloud ratio The weighting coefficients are configured to be greater than the standard deviation of elevation. and the compensated signal strength value The weighting coefficients.

4. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 1, characterized in that, Step S3, which performs a dual health degradation assessment based on time-domain derivative monitoring, specifically includes short-term drop detection and trend decay detection, wherein: In short-term sudden drop detection, it is determined whether the health score of the current control cycle has dropped beyond the preset threshold compared with the previous control cycle; In trend decay detection, a sliding time window of a specific length is constructed. The difference between the health score at the current moment and the health score at the beginning of the sliding time window is calculated, and the first time derivative of the health score is obtained. When the first time derivative When the target radar continuously falls below the set negative warning threshold, it is determined that the corresponding target radar has deteriorated, triggering predictive active downweighting.

5. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 1, characterized in that, Step S4 Specifically, the following steps are included: S41. Issue a decay command to the underlying fusion weight matrix to implement a gradual linear decay strategy with a fixed time period and fixed step size for the fusion weights of faulty radars showing a degradation trend, and obtain the weight allocation value of the faulty radar at the current time after the weight reduction is performed. ; S42. In the parallel thread that performs weight decay, wake up the redundant backup radar that is in a dormant or low-power state to complete the preheating; calculate the effective sensing field of view lost by the faulty radar and generate a yaw angle compensation command to drive the adjacent healthy isomorphic radar of the faulty radar to deflect its field of view toward the faulty area to reconstruct the physical field of view.

6. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 5, characterized in that, In step S42, the physical field of view is reconstructed through the following path 1 and / or path 2: Path 1 changes the overall field of view orientation by physically rotating a mechanical gimbal located at the bottom of the adjacent health radar. Path 2 adjusts the timing of the laser emission array by issuing electronic scanning control commands, thereby changing the three-dimensional distribution of energy in the effective detection field of view without changing the physical position.

7. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Using the hyperbolic tangent function, a mathematical model is constructed to represent the nonlinear physical attenuation law of particulate matter in the air, and an environmental adjustment factor is dynamically generated. ; S52, Environmental regulation factors Introduced as an exponential adjustment coefficient into the probability allocation model, it is combined with the quantified health score to calculate the dynamic fusion weight allocated to each radar within the current control cycle. ; S53, Obtaining the dynamic fusion weights of each radar Then, spatial registration and weighted summation of the underlying point cloud data are performed, and dynamic fusion weights are determined. Less than the safe discard hard threshold Data transmission is interrupted and discarded, and a 3D global environment fusion point cloud matrix is ​​output. .

8. The multi-liquid radar redundant sensing and control method for unmanned mining trucks according to claim 1, characterized in that, The processed global fused point cloud data output in step S6 includes: It includes a list of obstacles with 3D coordinates, motion vector velocity, and detection confidence; a 3D grid map of passable areas; and a system status diagnostic message containing the health status of each radar and its assigned weight variables.

9. A multi-liquid radar redundancy sensing system for unmanned mining trucks, characterized in that, For performing the control method as described in any one of claims 1 to 8, the system comprises: A surround-view isomorphic lidar array is deployed around the mining truck to form all-round coverage, and supports receiving commands to achieve dynamic airspace blind spot filling. Environmental condition sensor components are deployed on the vehicle body to capture external environmental parameters that cause perception degradation. The communication bus is used to connect all sensors to the central domain controller; The central domain controller has a built-in processing chip and memory. The memory stores a computer program, and when the processor executes the program, it can cyclically call the control logic of a multi-LiDAR redundant sensing control method for unmanned mining trucks.

10. A multi-liquid radar redundancy sensing system for unmanned mining trucks according to claim 9, characterized in that: The surround-view isomorphic lidar array uses multiple automotive-grade solid-state or semi-solid-state lidars that support dynamic configuration of the region of interest. The environmental condition sensor assembly includes a dust detector and a temperature and humidity sensor; The central domain controller issues low-level electronic scanning control commands and uses software to adjust the timing of the laser emission array, thereby changing the three-dimensional distribution of energy in the effective detection field of view to cope with the harsh working conditions of severe vibration of mining trucks.

Images