Path selection method and system for AGV (Automatic Guided Vehicle) transportation equipment, product and medium

Abstract

The invention discloses a path selection method and system for AGV transportation equipment, a product and a medium. The method comprises the following steps: firstly, when the automated guided vehicle runs, calculating and updating vibration and slippage cost values for each unit of a grid map in real time by collecting vertical vibration and wheel-actual speed difference data, so as to construct a dynamic physical risk map; after a transportation task is received, obtaining exclusive vibration and slippage weights according to task characteristics (such as whether goods are fragile or not); evaluating a plurality of to-be-selected paths, and accumulating the length cost of each path and the weighted vibration cost and the weighted slip cost of all grids along the path to calculate the comprehensive passing cost; and selecting the path with the lowest comprehensive passing cost as a final execution path. By implementing the technical scheme provided by the invention, the adaptability of AGV path planning to complex physical road conditions is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of path planning of an autonomous driving vehicle, and in particular to a path selection method, system, product and medium of an AGV transportation device. BACKGROUND

[0002] At present, with the in-depth development of logistics automation and intelligent manufacturing, the automatic guided vehicle (AGV) as a key automated equipment connecting production, warehousing and distribution links is increasingly widely used, and is very important for improving the overall operation efficiency and intelligent level of modern industry.

[0003] In related technologies, an advanced AGV path planning method is to perform global path optimization based on a layered cost map. This method assigns a comprehensive passing cost value to each grid in the map by fusing a static map layer (such as the building structure of a factory, fixed shelves) and a dynamic obstacle layer (such as moving staff or other temporary obstacles). A path planning algorithm (such as the A* algorithm) searches on this cost map to find a path with the lowest cumulative cost value from the starting point to the ending point as the navigation route.

[0004] However, since the cost value of the cost map mainly comes from the perception of the geometric structure and obstacle distribution of the environment, this method tends to assign a uniform basic passing attribute to all areas that are identified as passable in geometry. Therefore, when the road surface of a path has degraded in physical properties, such as fine cracks on the ground due to long-term weight, uneven settlement at the joint, or local slippery due to liquid leakage, and these changes do not reach the level of geometric obstacles, the cost map is difficult to update and reflect such risks, resulting in the path planning system may continue to judge a geometrically optimal path as the best choice, ignoring the fact that its physical passing quality has significantly decreased, and there is a potential risk of causing goods to fall or vehicles to lose stability. SUMMARY

[0005] The present application provides a path selection method, system, product and medium of an AGV transportation device, for improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0006] In a first aspect of the present application, a path selection method of an AGV transportation device is provided, the method comprising: The vertical vibration data of vertical direction motion change in the driving process of the automatic guided vehicle, the first speed data representing wheel speed, and the second speed data representing actual displacement change are acquired; a grid map is established according to a preset grid size, and when the automatic guided vehicle drives to each grid, the physical state data of the corresponding grid is calculated and continuously updated; after receiving the transportation task, the vibration weight and the slip weight corresponding to the transportation task are acquired; for a plurality of candidate paths of the transportation task, the path length cost, the product of the vibration cost value and the vibration weight, and the product of the slip cost value and the slip weight of each grid through which each candidate path passes are added respectively, to obtain the comprehensive passing cost of the corresponding candidate path; in the plurality of candidate paths, the candidate path with the lowest comprehensive passing cost is selected as the final execution path.

[0007] In the above embodiment, by collecting and analyzing physical data such as vertical vibration and wheel speed difference in real time, the physical characteristics of the road surface that are ignored in traditional path planning, such as bumps and wetness, are quantified as calculable vibration cost value and slip cost value. Avoid the short board that only rely on geometric information in related technology and cannot perceive physical road condition defects. Further, by combining these physical costs reflecting the real passing risk with the task-specific weight to calculate the comprehensive passing cost, the path selection is no longer limited to the shortest distance, but can actively avoid those paths that may be more geometrically optimal but have the risk of bumps or skidding, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0008] In combination with some embodiments of the first aspect, in some embodiments, after the path length cost, the product of the vibration cost value and the vibration weight, and the product of the slip cost value and the slip weight of each grid through which each candidate path passes are added respectively for a plurality of candidate paths of the transportation task, to obtain the comprehensive passing cost of the corresponding candidate path, the method further comprises: calculating a single-point risk value of each grid in each candidate path; selecting grids with a single-point risk value exceeding a preset threshold from each candidate path as risk mutation points; calculating the difference between the single-point risk value of the risk mutation point and the risk average of the adjacent grid to obtain the isolation degree of the corresponding risk mutation point; when there is a risk mutation point in the candidate path, the comprehensive passing cost of the corresponding candidate path is corrected to the sum of the comprehensive passing cost and the product of the single-point risk value and the isolation degree of the corresponding risk mutation point.

[0009] In the above embodiment, by screening high-risk positions on the path positioning risk mutation points, the isolation degree of these mutation points is further calculated, and the suddenness of risk occurrence is quantified - an isolated risk point surrounded by a flat road has a higher isolation degree than a bump in a continuous bumpy road section. By multiplying the isolation degree with the risk value itself to correct the comprehensive travel cost, the penalty for unexpected risks is significantly amplified, forcing the path planning algorithm to avoid paths containing sudden dangers, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0010] In combination with some embodiments of the first aspect, in some embodiments, after calculating the difference between the single-point risk value of the risk mutation point and the adjacent grid risk average value to obtain the isolation degree corresponding to the risk mutation point, it further includes: When the driving direction of the automatic guided vehicle changes at the risk mutation point, the steering angle value is obtained; when the steering angle value is less than a preset first angle threshold, the isolation degree is multiplied by a preset first correction coefficient; when the steering angle value is not less than the first angle threshold and less than a preset second angle threshold, the isolation degree is multiplied by a preset second correction coefficient; when the steering angle value is not less than the second angle threshold, the isolation degree is multiplied by a preset third correction coefficient.

[0011] In the above embodiment, by capturing the steering behavior of the vehicle at the risk mutation point, and dynamically correcting the core indicator of risk assessment, isolation degree, according to the size of the steering angle. The instability risk of sharp turning at a bump or slippery point is much higher than straight-line passing, which is integrated into the algorithm. By matching a higher correction coefficient to a larger steering angle, the risk superposition effect is quantified and amplified, so as to impose a heavier penalty on such high-risk combination behaviors in path cost calculation, guiding the AGV to choose to complete the steering on the smooth road surface, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0012] In combination with some embodiments of the first aspect, in some embodiments, when the grid map is established according to the preset grid size, after the automatic guided vehicle travels to each grid, the physical state data corresponding to the grid is calculated and continuously updated, further including: The vibration substitute value and the slip substitute value of each grid in the grid map are calculated, and the adjacent vibration difference set and the adjacent slip difference set of all adjacent grids are calculated; when there is a value exceeding a preset difference threshold in the adjacent vibration difference set or the adjacent slip difference set, and the vibration substitute value or the slip substitute value of the corresponding grid is greater than that of all adjacent grids, the corresponding grid is determined as a risk source grid; two-dimensional slope information of the risk source grid and all adjacent grids is obtained; a local height map covering the corresponding grid is constructed according to the two-dimensional slope information, and a risk diffusion direction is determined based on the local height map; a plurality of risk spreading grids are determined according to the risk diffusion direction; the vibration substitute value and the slip substitute value of the risk spreading grid are multiplied by a preset risk coefficient; and in the calculation of the comprehensive passing cost of the candidate path, the vibration substitute value and the slip substitute value of the risk spreading grid are used.

[0013] In the above embodiment, firstly, the risk source grid as the root of the problem is located by analyzing the local mutation of the risk value, the risk is not regarded as an isolated point, and the diffusion direction conforming to the physical law is matched for the risk according to the physical causes (slip or vibration) of the risk and the local slope information, for example, the slip risk will spread along the slope downward. By applying a risk coefficient decreasing with distance to the subsequent grids along this direction, an isolated risk point is expanded to a continuous risk area with reasonable shape and gentle boundary on the cost map. From the modeling of point risk to physical surface risk, the path planning algorithm can foresee and smoothly bypass the entire dangerous area, rather than only avoiding a point at the last moment, finally improving the adaptability of AGV path planning to complex physical road conditions.

[0014] In combination with some embodiments of the first aspect, in some embodiments, the plurality of risk spreading grids are determined according to the risk diffusion direction, specifically comprising: The vibration substitute value or the slip substitute value of the risk source grid is obtained as a risk source intensity index; according to the risk source intensity index, the number of risk spreading grids is determined by querying a preset mapping relationship between the risk source intensity index and the number of risk spreading grids; and the plurality of adjacent grids along the risk diffusion direction and determined by the dynamic number of risk spreading are determined as the risk spreading grids.

[0015] In the above embodiment, the severity of the risk is quantified by extracting the risk source intensity index. The intensity index is associated with a preset mapping relationship, so that the distance to which the risk should spread is dynamically and proportionally determined, so that a serious slip source will generate a large avoidance area, and a slight bump will only affect the local, so that the path planning can more reasonably evaluate and weigh different degrees of danger, and finally improve the adaptability of AGV path planning to complex physical road conditions.

[0016] In some embodiments of the first aspect, in some embodiments, after the physical state data is calculated and continuously updated for the corresponding grid, the method further comprises: marking the corresponding grid with a vibration value exceeding a preset frequency threshold and a slip value greater than a preset slip starting threshold as a coupling risk grid; adding a preset coupling risk value to the comprehensive passing cost of all coupling risk grids.

[0017] In the above embodiments, by identifying the coupling risk grid of the vehicle encountering the extremely dangerous combination of bumpy road (high frequency vibration) and wet and slippery road (high slip) at the same time, a coupling risk value is added to the nonlinear amplification penalty of the dangerous level of this combined risk, reflecting the physical fact that bumpy and wet and slippery roads are more likely to cause loss of control than single risk in the real world. This mechanism forces the path planning algorithm to avoid such most dangerous areas, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0018] In some embodiments of the first aspect, in some embodiments, after receiving the transportation task, the method further comprises: When the transportation task is a multi-site task containing multiple work sites, the multi-site task is divided into multiple consecutive transportation sections according to the order of the work sites; for each transportation section, the expected load state of the automated guided vehicle when driving on the corresponding transportation section is determined in advance according to the operation instructions of the starting work site and the ending work site; the real-time vibration weight and the real-time slip weight are found in the preset load weight mapping table according to the expected load state; the comprehensive passing cost of the path with the lowest cost of each transportation section is calculated and executed independently using the real-time vibration weight and the real-time slip weight, until the entire multi-site task is completed.

[0019] In the above embodiments, by dividing the complex task into independent transportation sections and predicting the unique expected load state for each section, the physical characteristics (such as center of gravity, inertia) of the AGV are recognized to be dynamically changing throughout the task. Based on this, different vibration and slip risk weights are matched for different load states, which enables the core cost calculation of path planning to be dynamically adjusted, and the risk preference strategy to be adjusted in real time according to the task progress, so that the path selection can match the physical characteristics of the vehicle at each stage, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0020] In the second aspect, the embodiments of the present application provide a path selection system, which comprises one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code comprises computer instructions, and the one or more processors invoke the computer instructions to make the path selection system execute the method described in the first aspect and any possible implementation manner of the first aspect.

[0021] In a third aspect, the embodiments of the present application provide a computer program product comprising instructions which, when executed on a path selection system, cause the path selection system to carry out the method according to the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, the embodiments of the present application provide a computer-readable storage medium comprising instructions which, when executed on a path selection system, cause the path selection system to carry out the method according to the first aspect and any possible implementation of the first aspect.

[0023] It can be understood that the path selection system provided by the second aspect, the computer program product provided by the third aspect, and the computer storage medium provided by the fourth aspect are all used to execute the path selection method of the AGV transportation device provided by the embodiments of the present application. Therefore, the beneficial effects that can be achieved are referred to the beneficial effects in the corresponding method, which will not be described here.

[0024] The one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. The present application quantifies the physical characteristics of the road surface, such as bumps and slippery, which are ignored in traditional path planning, as calculable vibration and slip values by real-time collection and analysis of physical data such as vertical vibration and wheel speed difference. Avoid the short board that only rely on geometric information in related technology and cannot perceive the physical road condition defects. Further, by combining the physical cost reflecting the real traffic risk with the task-specific weight to calculate the comprehensive traffic cost, the path selection is no longer limited to the shortest distance, but can actively avoid those paths that may be more optimal in geometry but have the risk of bumps or slippery, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0025] 2. The present application locates high-risk positions on the path by screening risk mutation points, further calculates the isolation degree of these mutation points, and quantifies the suddenness of risk occurrence - an isolated risk point surrounded by flat road has a higher isolation degree than a bump point in a continuous bump road section. By multiplying the isolation degree with the risk value itself to correct the comprehensive traffic cost, the punishment for unexpected risks is significantly magnified, thereby forcing the path planning algorithm to avoid paths containing sudden dangers, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0026] 3. Firstly, the application locates the risk source grid as the root of the problem by analyzing the local mutation of the risk value, does not regard the risk as an isolated point, but according to the physical cause (slip or vibration) of the risk and the local slope information, matches the diffusion direction of the risk with the physical law, for example, the slip risk will spread down the slope. By applying a risk coefficient that decreases with distance to the subsequent grid along this direction, an isolated risk point is expanded to a continuous risk area with reasonable shape and gentle boundary on the cost map. From the modeling of point risk to physical surface risk, the path planning algorithm can foresee and smoothly bypass the entire dangerous area, rather than just avoid a point at the last moment, finally improving the adaptability of AGV path planning in the face of complex physical road conditions. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a flowchart of the path selection method of the AGV transportation device in the embodiment of the application; Figure 2 is another flowchart of the path selection method of the AGV transportation device in the embodiment of the application; Figure 3 is an exemplary hardware structure schematic diagram of the path selection system in the embodiment of the application. DETAILED DESCRIPTION

[0028] The terms used in the following embodiments of the application are only for the purpose of describing specific embodiments and are not intended to be limiting to the application. As used in the specification and the appended claims of the application, the singular forms "a," "an" and "the" are intended to include both singular and plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0029] Hereinafter, the terms "first", "second" are only used for the purpose of description, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the application, unless otherwise specified, the meaning of "multiple" is two or more.

[0030] In the related art, AGV path planning methods generally rely on grid-based cost maps for global optimization. This method mainly perceives the geometric layout of the environment, such as walls, shelves, fixed equipment, and dynamic obstacles, through sensors such as laser radars and vision, and converts this information into a grid map cost value. Path planning algorithms (such as A*) search for a path with the shortest geometric distance or the lowest detour cost based on this. However, the essence of this method is a geometric abstraction of the environment, which can effectively identify passable and impassable areas, but lacks the ability to perceive the physical travel quality within passable areas. For example, a bumpy road segment caused by uneven floor joints, or an area that becomes slippery due to accidental oil stains or water accumulation, has no difference in travel cost from a flat and dry road surface in the traditional cost map. Therefore, AGV may be guided to a path that is geometrically optimal (e.g., shortest), but physically has the risk of overturning, slipping, or damaging goods, which poses a hidden danger to safe and efficient material handling.

[0031] In the embodiments of the present application, in order to make up for the above-mentioned defects of relying only on geometric information and ignoring physical travel risks, the present scheme proposes a path planning method based on vehicle physical state feedback. By real-time acquisition and processing of physical data that can directly reflect road conditions, i.e., vibration cost values calculated from vertical motion changes, and slip cost values calculated from the difference between wheel speed and actual displacement speed, these dynamically updated physical risk indicators are continuously enabled as new cost dimensions for each cell of the grid map. Ultimately, when making path decisions, instead of solely pursuing the shortest distance, a path with the lowest comprehensive travel cost that combines path length, vibration risk, and slip risk is calculated by combining task-specific vibration weights and slip weights. This enables the path selection to actively avoid areas that are physically dangerous but seemingly harmless geometrically, improving the AGV's ability to perceive and adapt to complex physical road conditions.

[0032] Figure 1 is a flowchart of the path selection method for AGV transportation equipment in the embodiments of the present application, including the following steps: S101, acquiring vertical vibration data of vertical direction motion changes of the automatic guided vehicle during driving, first speed data representing wheel speed, and second speed data representing actual displacement changes.

[0033] AGV, which stands for Automated Guided Vehicle, refers to a transport vehicle equipped with electromagnetic or optical automatic guiding devices, capable of traveling along a specified guiding path, with safety protection and various transfer functions. Vertical vibration data represents the acceleration or angular velocity change of AGV in the Z-axis direction, i.e., the direction perpendicular to the ground, which directly represents the degree of jolt caused by uneven ground (such as joints, pits, and bumps). The first speed data refers to the theoretical travel speed calculated by measuring the rotational speed of the AGV drive wheel and combining the wheel radius, reflecting the speed that the motor control system expects the vehicle to achieve. The second speed data refers to the actual moving speed of the AGV vehicle body in the global coordinate system directly measured by positioning systems such as laser radar, visual odometry, or high-precision inertial measurement unit (IMU), representing the real, ground-relative motion rate of the vehicle.

[0034] Specifically, AGV synchronously collects three sets of physical quantities during travel. First, through the acceleration sensor or inertial measurement unit (IMU) installed on the AGV vehicle body or suspension system, the vehicle's vertical vibration signal is captured in real time, and the severity of this signal is directly related to the flatness of the road, which is the basis for judging the risk of jolt. Second, through the encoder (such as an optical encoder or a Hall sensor) installed on the drive motor or wheel, the rotational angular velocity of the wheel is measured, and the first speed data, i.e., the linear speed of the wheel, is calculated accordingly. Finally, the SLAM (Simultaneous Localization and Mapping) system carried by the AGV, which usually integrates laser radar (LiDAR) and IMU data, is used to obtain the real-time pose of the vehicle in the global grid map, and the second speed data, i.e., the actual motion result of the vehicle, is obtained by time differentiation of the pose. All three data streams have accurate timestamps to ensure synchronization and alignment of the data in subsequent calculations.

[0035] In some embodiments, data acquisition in this step can be achieved in various ways to adapt to different hardware configurations and accuracy requirements: optionally, a high-precision industrial-grade configuration is adopted: a high-sensitivity, high-sampling-rate (e.g., greater than 200 Hz) three-axis MEMS accelerometer is installed at the center of the AGV's load platform, specifically for collecting vertical vibration data. A high-resolution optical encoder is installed on each drive wheel to calculate the first speed data. A 16-line or 32-line laser radar is used in combination with a high-performance IMU to obtain centimeter-level positioning accuracy through a SLAM algorithm based on graph optimization, thereby calculating the second speed data with high confidence.

[0036] S102, a grid map is established according to a preset grid size, and when the automated guided vehicle travels to each grid, the physical state data is calculated and continuously updated for the corresponding grid.

[0037] The preset grid size is set according to the layout of the AGV working environment (such as a factory or a warehouse) and the accuracy requirement of path planning, determines the fineness of the divided map, and is configured artificially according to the application scenario; the grid map refers to discretizing the working environment of the AGV into a two-dimensional grid composed of a large number of square or rectangular units (i.e. grids), each grid represents a small area in the environment, and stores the attribute information of the area; the physical state data refers to a set of quantitative indicators associated with each grid unit, which are obtained by associating the vibration equivalent value, the slip equivalent value and the current position value of the automatic guided vehicle; the vibration equivalent value is a dimensionless value, which is used to represent the degree of severe shaking when the AGV passes through the grid, and is calculated by analyzing the frequency and amplitude characteristics of the vertical vibration data, the higher the value, the more uneven the road surface; the slip equivalent value is also a dimensionless value, which is used to represent the severity of the skidding or idling of the wheels on the grid surface, and is obtained by calculating the difference between the first speed data and the second speed data, the higher the value, the more slippery or lower the friction of the road surface.

[0038] Specifically, first, a grid map with a preset resolution is established according to the layout of the space where the factory or warehouse is located. When the AGV is driving, the real-time position is continuously mapped to this grid map to determine the current grid unit. For the calculation of the vibration equivalent value, the vertical vibration data in a time window (for example, 0.5 seconds) is intercepted, and the frequency spectrum is analyzed through signal processing technology (such as fast Fourier transform FFT), and the occurrence of high amplitude or specific frequency (such as the resonance frequency harmful to precision instruments) will produce a higher vibration equivalent value; or the vibration intensity is quantified by a simpler time domain statistics (such as calculating the root mean square RMS value). For the calculation of the slip equivalent value, the first speed data (wheel expected speed) and the second speed data (vehicle actual speed) at the same time are compared, and the difference or ratio (for example: (first speed-second speed) / first speed) directly reflects the slip rate, which is the slip equivalent value after normalization. The equivalent value of each grid is not fixed once the measurement is made, but a smoothing filtering strategy is adopted, such as the exponential moving average method (EMA), the formula is: new equivalent value = a* this measurement value + (1-a)* old equivalent value. Here, a is a small learning rate (such as 0.1), which enables the risk data on the map to be continuously self-corrected and improved over time and with the passage of a large number of vehicles, thereby filtering out the noise of a single measurement and approximating the true physical condition of the road section.

[0039] In some embodiments, the calculation and update of the equivalent value in this step can be realized in various ways: Optionally, frequency-domain analysis based vibration cost calculation and weighted average update: when the AGV is in a certain grid, a vertical vibration data sequence containing 256 sampling points is collected; after applying the Hanning window function to the sequence, the FFT transform is performed to obtain the power spectrum density graph; according to the energy distribution of the power spectrum in multiple preset sensitive frequency bands, a comprehensive vibration cost value is obtained by weighted summation, and the value is used to update the historical data of the grid by weighted average (instead of exponential moving average).

[0040] Optionally, time-domain statistical based vibration cost calculation and maximum value retention update: a data sequence is also collected; the peak-to-peak value or kurtosis of the sequence is directly calculated, which is very sensitive to impact vibration; the calculated statistical value is normalized to a vibration cost value through a nonlinear mapping function (such as a Sigmoid function), and a "maximum value retention" strategy is adopted for update, i.e. new cost value = MAX (this measurement value, old cost value).

[0041] It can be understood that other ways can also be used to realize the quantification of the cost and the update of the data, for example, introducing a machine learning model to directly regress the cost value according to the original vibration and slip data sequence, or using a more complex Bayesian filtering method for data update, which is not limited here.

[0042] In some embodiments, after calculating and continuously updating the conventional physical state data for the corresponding grid, special risk grids coupled by high-frequency vibration and wet road surface can also be further identified and marked, aiming to enhance the identification and avoidance ability of linearly superimposed risks.

[0043] Specifically, first, when calculating the vibration cost value in S102, an additional frequency analysis is performed on the vibration signal. Through fast Fourier transform (FFT) or other spectrum estimation algorithm, the power spectrum density of the vibration signal is obtained. If a dominant frequency peak is found in the power spectrum, and the frequency of the peak exceeds a high frequency threshold (the preset high frequency threshold is set according to the natural frequency of the vehicle suspension system and the damageable characteristics of the transported goods, which is based on the physical characteristics of the vehicle and known experience rules) preset according to the natural frequency of the vehicle suspension and the damageable characteristics of the goods, a high frequency vibration mark of Boolean type is added to the current grid. The significance of this step is to distinguish ordinary low-frequency jolt caused by road undulation from high-frequency micro-vibration caused by fine texture of the ground (such as metal non-slip plate) or high-frequency mechanical equipment, which is more destructive.

[0044] Subsequently, each grid in the map is periodically polled. For a grid, two conditions are checked simultaneously: first, whether the grid is marked with a high-frequency vibration flag; second, whether the slip cost value of the grid is greater than a preset slip initiation threshold, which represents the critical state where the road surface is no longer completely dry and starts to have slight slippage, determined by experimental data and vehicle dynamics analysis. A grid is identified as a coupling risk grid only when it satisfies both conditions simultaneously, and an additional coupling risk cost value with a large, fixed preset value is associated with its physical state data. This has the benefit of simulating a real-world non-linear physical trap: high-frequency vibrations cause rapid micro-bounces of the wheels, instantaneously destroying the static friction force basis between the tire and the ground, at which point if the ground is wet and slippery, each micro-bounce will translate into a small, irreversible slip, and after countless accumulations, it will lead to a serious loss of vehicle tracking ability, which is far more dangerous than the simple linear addition of vibration cost and slip cost. By imposing a large, punitive additional cost, the subsequent path planning algorithm can actively avoid special road sections that appear normal but harbor great risks when calculating the cost.

[0045] Through the above steps, coupling risks that cannot be captured by conventional linear models can be identified. When the AGV plans a path, even if a path is very short in geometry, but as long as it passes through a marked coupling risk grid, the comprehensive travel cost will skyrocket, avoiding the combination trap of high-frequency vibration and wet road, thereby selecting a path that may be longer but is much safer physically, enhancing the robustness and safety of path selection, and ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0046] S103、In response to receiving the transportation task, obtain the vibration weight and the slip weight corresponding to the transportation task.

[0047] The transportation task refers to a specific instruction issued by an upper-level scheduling system (such as WMS or FMS) to the AGV, usually including task ID, starting point, target point, and cargo information, etc. The vibration weight is a numerical coefficient representing the sensitivity of the current transportation cargo to jolts, and the higher the weight, the more inclined the path planning should be to avoid bumpy road sections. The slip weight is also a numerical coefficient representing the sensitivity of the current transportation cargo to slip and rollover risks, and the higher the weight, the more inclined the path planning should be to avoid wet road sections.

[0048] Specifically, when the AGV receives a new transport task, the task instruction contains a cargo type identifier (Cargo ID) or a task priority code. The on-board computing unit of the AGV or the server connected thereto pre-stores a cargo-weight mapping table, which defines the vibration weight and the slip weight corresponding to different types of cargo in detail. For example, "Class A: precision optical glass" may correspond to the weight {vibration: 0.9, slip: 0.3}, indicating that it is extremely sensitive to vibration, but as long as it does not overturn, it can tolerate slip; and "Class B: full load liquid tank" may correspond to the weight {vibration: 0.2, slip: 0.9}, indicating that it is not afraid of jolting, but due to high center of gravity and liquid sloshing, it is prone to overturning when slipping and turning. After analyzing the task, the AGV will query the two weight values found in the table according to the cargo identifier, and load them into the current planning session. For the same physical map, the AGV transporting different cargos will present different risk cost maps, realizing individualized path planning.

[0049] S104, for the multiple candidate paths of the transport task, respectively, the path length cost of each grid passed by each candidate path, the product of the vibration value and the vibration weight, and the product of the slip value and the slip weight are added up to obtain the comprehensive passing cost corresponding to the candidate path.

[0050] Wherein, the candidate path refers to a plurality of geometrically feasible path sequences between the current starting point and the target point searched by a global path planning algorithm (such as A*, Dijkstra, etc.); the path length cost refers to the geometric length of the path; the comprehensive passing cost is a final decision-making, comprehensive quantitative score, which combines the length, jolting risk and slippery risk of the path, and is the final index for evaluating the advantages and disadvantages of a path.

[0051] Specifically, first, a graph search algorithm is called to find multiple candidate paths from the start point to the end point on the grid map. For each candidate path found, a cost traversal is performed. This traversal process visits each grid cell on the path in sequence. When visiting each grid, three core attributes of the grid are read from the grid map: path length cost (usually the length of the edge or diagonal of the grid), vibration cost value, and slip cost value. Then, a cost calculation formula is applied to combine these attributes with the vibration weight and slip weight obtained in S103. A typical calculation formula is: grid traversal cost = path length cost + vibration cost value * vibration weight + slip cost value * slip weight. Finally, the traversal costs of all grids on the path are accumulated, and the comprehensive traversal cost =∑[path length cost + vibration cost value * vibration weight + slip cost value * slip weight]. By performing this calculation process for each candidate path, a list containing all candidate paths and their corresponding comprehensive traversal costs can be obtained.

[0052] In some embodiments, the cost calculation and path generation in this step can be implemented in various ways. Optionally, the cost calculation is integrated into the heuristic function of the A* algorithm: when performing the A* algorithm search, g(n) (the actual cost from the start point to the current node) in the cost function f(n) = g(n) + h(n) of the A* algorithm is redefined; g(n) is no longer simply the path length accumulation, but the comprehensive traversal cost is accumulated at each step. In this way, when the A* algorithm expands nodes, it will naturally tend to avoid grids with high vibration or slip costs (depending on the weights), thereby directly searching for the path with the lowest comprehensive cost, with higher efficiency.

[0053] It can be understood that other ways of calculating the cost can also be used, such as using a multiplication model instead of an addition model to combine the costs, or introducing more complex nonlinear functions to handle the relationship between the weights and the costs, which are not limited here.

[0054] In some embodiments of the present application, after calculating the preliminary comprehensive traversal costs of the candidate paths, potential risk mutation points in the paths can be further identified, and the total cost can be dynamically revised in combination with the isolation degree of the points and the expected turning action of the AGV at the points, to avoid traps hidden in flat road segments.

[0055] Firstly, for each grid on the candidate path, the vibration surrogate value and the slip surrogate value are read from the grid map. The single-point risk value is calculated by multiplying the vibration surrogate value and the slip surrogate value, which is an index to quantify the inherent risk level of a single grid. The design rationale is that the two risks of vibration and slip are coupled and amplified, and a road with only vibration but dry or only slip but flat is still in a controllable range. However, when the AGV experiences a transient bounce on a bumpy road, and the road is also slippery, the probability of vehicle out-of-control will increase dramatically. The multiplication operation can capture this phenomenon: only when both surrogate values are high, the result will be significantly increased, thus reflecting this compound risk scenario.

[0056] Secondly, after obtaining the single-point risk values of all grids on the path, they are compared with a pre-set risk mutation threshold, which is determined based on safety engineering and long-term empirical data, representing the critical point of risk from acceptable to vigilant. Any grid whose single-point risk value exceeds this threshold will be marked as a risk mutation point, which represents a specific grid on a path whose single-point risk value is significantly higher than the normal level, posing a potential threat, such as a sudden pothole or a patch of oil on a flat road. For each marked risk mutation point, the isolation degree is calculated, which is a quantitative index to represent the risk difference between the risk mutation point and its surrounding environment. The higher the isolation degree, the more conspicuous the risk point. The calculation first needs to determine the average risk of adjacent grids, i.e., obtaining the single-point risk values of all adjacent grids (8 grids in a typical eight-neighborhood model) around the point and calculating the arithmetic mean. Then, subtract the average from the single-point risk value of the risk mutation point itself, and the difference is the isolation degree. The concept of isolation degree is introduced to quantify the conspicuousness of risk. If a high-risk point is surrounded by other high-risk points (low isolation degree), it belongs to a known dangerous area; otherwise, if it is surrounded by very safe road segments (high isolation degree), it constitutes an unpredictable trap, posing a greater threat to safe driving.

[0057] Furthermore, the geometric sequence of the candidate path is analyzed, and for each risk mutation point, the turning angle value that the AGV needs to perform at this point is calculated through the positions of the previous grid, the current grid, and the next grid. The turning angle value is used to represent the change in the direction of the AGV when it passes through a certain grid on the path. For example, when passing through a straight line, the angle is 0 degrees, and when making a right-angle turn, the angle is 90 degrees. Then, according to this angle value, the isolation of the risk mutation point is dynamically corrected. If the turning angle is less than a preset first angle threshold (such as 15°, which is considered to be a straight pass), the isolation is multiplied by a smaller first correction coefficient. If the turning angle is between the first angle threshold and a second angle threshold (such as 45°), it is multiplied by a larger second correction coefficient. If the turning angle exceeds the second threshold (which is considered to be a sharp turn), it is multiplied by a maximum third correction coefficient. The preset first / second angle threshold and the preset first / second / third correction coefficient together form a piecewise function, which is a pre-configured system parameter. It is used to amplify the risk index of isolation at different levels according to the severity of the turning, in order to simulate the dynamic increase of risk when turning. The physical principle of this step is that turning at a risk point (especially a point with a risk of slipping) will generate centrifugal force, increasing the risk of side slipping and overturning, and the sharper the turn, the greater the risk. In this way, the risk assessment is upgraded from judging the static road surface to judging the dynamic vehicle-road interaction system.

[0058] Finally, after completing all the above analyses, the comprehensive travel cost of the candidate path identified as containing risk mutation points is recalculated. The corrected formula is: corrected cost = original cost + Σ [single-point risk value x corrected isolation]. The summation symbol Σ here means that the penalty term generated by all risk mutation points on the path needs to be added up. This penalty term is obtained by multiplying the single-point risk value by the turning-corrected isolation, achieving double weighting of risk. If a risk point is highly dangerous (high single-point risk value) and very conspicuous (high isolation), the resulting penalty term will be larger, causing the final comprehensive travel cost of the path containing such a deadly trap to be raised, thereby reducing the probability of being used in the final decision.

[0059] Through the above steps, additional penalties are imposed on paths containing isolated trap risks, especially when the AGV needs to turn at the trap. This significantly raises the final comprehensive travel cost of such paths, putting them at a disadvantage in the comparison stage of S105, and thus being naturally avoided by the path selection algorithm. Ultimately, the AGV can actively abandon pseudo-optimal paths that have low average costs but contain deadly risk points, and choose a truly safe and robust route, improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0060] S105, among the multiple candidate paths, the candidate path with the lowest comprehensive travel cost is selected as the final execution path.

[0061] Specifically, after S104 completes the comprehensive cost calculation for all candidate paths, it obtains a key-value pair set of paths and costs. All cost values ​​in this set are compared, and the path with the lowest cost is selected as the final execution path. This selected path is essentially a sequence of waypoints containing coordinates and order, which is passed to the AGV's underlying motion controller. The motion controller is then responsible for generating smooth speed and angular velocity commands based on this waypoint sequence, driving the motors and steering mechanism to guide the AGV along this path until the task is completed.

[0062] In the above embodiments, by collecting and analyzing physical data such as vertical vibration and wheel speed difference in real time, road surface physical characteristics, such as bumps and slippage, which are ignored in traditional path planning, are quantified into calculable vibration and slippage costs. This avoids the shortcomings of related technologies that rely solely on geometric information and cannot perceive physical road condition defects. Furthermore, by combining these physical costs, which reflect real traffic risks, with task-specific weights to calculate comprehensive traffic costs, path selection is no longer limited to the shortest distance but can proactively avoid geometrically better paths that have the risk of bumps or slippage, ultimately improving the adaptability of AGV path planning to complex physical road conditions.

[0063] In other embodiments of this application, when multiple risk sources are densely distributed in space to form risk clusters, the simple overlap of risk spread areas may lead to misjudgment of the overall risk of the area. Using the path selection method for AGV transportation equipment provided in this application, risk sources can be clustered first, risk clusters can be identified as overall risk areas, and then risk spread can be planned uniformly based on comprehensive intensity and macro-boundaries, thereby more accurately assessing and avoiding such complex and dangerous areas.

[0064] like Figure 2 The diagram shown is another flowchart illustrating the path selection method for AGV transportation equipment provided in this application embodiment, including the following steps: S201. Acquire vertical vibration data of vertical motion changes during the operation of the automated guided vehicle, first velocity data representing wheel rotation speed, and second velocity data representing actual displacement changes.

[0065] S202. Establish a grid map according to the preset grid size. When the automated guided vehicle travels to each grid, calculate and continuously update the physical status data for the corresponding grid.

[0066] Steps S201-S202 and Figure 1 Steps S101-S102 in the illustrated embodiment are similar and can be found in the descriptions of steps S101-S102, which will not be repeated here.

[0067] S203, calculate the vibration difference set and the slip difference set of each grid in the grid map.

[0068] wherein, the adjacent grids refer to the grids that share a side or a vertex with the current central grid, for example, the eight grids surrounding the central grid in the eight-neighborhood model; the vibration difference set and the slip difference set refer to the set of all the differences obtained by subtracting the risk value (vibration or slip) of each adjacent grid from the risk value of the central grid, which is used to depict the local gradient of the risk value of the point.

[0069] Specifically, each grid in the grid map is accessed in turn as a temporary central grid. For each central grid, the vibration value of the central grid is read first, and the vibration values of all the adjacent grids are obtained at the same time. Then, a series of subtraction operations are performed, i.e., the vibration value of the central grid is subtracted from the vibration value of each adjacent grid, so as to obtain a set containing a plurality of differences, which is defined as the vibration difference set of the central grid. The same calculation process is also applied to the slip value to generate the corresponding slip difference set. After this step, each grid in the map will be associated with two new attributes, i.e., the vibration difference set and the slip difference set.

[0070] S204, when there is a value in the vibration difference set or the slip difference set that exceeds the preset difference threshold value, and the vibration value or the slip value of the corresponding grid is greater than that of all the adjacent grids, the corresponding grid is determined as a risk source grid.

[0071] wherein, the preset difference threshold value refers to a critical value that is used to determine whether the local change of the risk value is drastic, which represents the boundary from gentle transition to risk mutation; the risk source grid refers to a special grid that is the most concentrated and most conspicuous in the local environment, which must meet two conditions at the same time: the risk value is the highest point (local peak) in the neighborhood; the risk difference with at least one neighbor is great (gradient is drastic). For example, the center point of a deep pit or a small patch of oil on a generally flat road surface can be identified as a risk source grid.

[0072] Specifically, after the difference sets of all the grids are calculated in S203, each grid in the map is traversed again, and a composite rule-based judgment logic is applied to each grid. For a grid, the following two conditions are checked at the same time: local peak condition: the vibration value of the grid is greater than the vibration values of all the adjacent grids. This condition ensures that the point is a mountain in its micro-environment.

[0073] Abrupt gradient condition: In the adjacent group of vibration difference values of the grid, there must be at least one difference value whose absolute value exceeds the preset difference threshold value, excluding the case that although it is a local highest point, the overall gradient is gentle and the danger is not great.

[0074] The above two conditions are executed on the vibration value and the slip value respectively. As long as a grid meets the above two conditions in any risk dimension at the same time, it will be finally marked as a risk source grid and the coordinates will be recorded.

[0075] In some embodiments, the screening and determination of the risk source grid can be realized in various ways: Optionally, a two-stage filtering method is adopted: in the first stage, all grids are traversed and only the local peak condition is checked. All grids that meet the condition are put into a candidate list; in the second stage, only the much smaller candidate list is traversed. For each candidate grid in the list, it is checked again whether it meets the abrupt gradient condition; finally, the remaining grid is the risk source grid.

[0076] It can be understood that the screening logic can also be expressed as a specific query statement for the spatial database, and the index and query optimizer of the database are used to efficiently complete the screening, which is not limited here.

[0077] S205, obtain the two-dimensional gradient information of the risk source grid and all adjacent grids.

[0078] The two-dimensional gradient information refers to a quantitative data describing the inclination degree of the ground represented by the grid in two orthogonal horizontal directions (for example, the X and Y axes of the map), which can usually be represented as a gradient vector containing two components (Gx, Gy), which indicates the steepest rising direction and steepness of the terrain. This information is usually obtained from high-precision sensors such as 3D laser radar.

[0079] Specifically, after at least one risk source grid is successfully identified in step S204, for each determined risk source grid, its coordinates are used as an index to query an independent terrain information layer that is spatially aligned with the cost map. After identifying the risk source grid, the coordinates of the grid are used as an index to query an independent terrain information layer (or slope map layer) that is spatially aligned with the cost map. These slope or height information is usually collected and constructed in advance by high-precision sensors such as 3D laser radar. The two-dimensional gradient information of the risk source grid itself and the gradient information of all directly adjacent grids are obtained and associated with the corresponding risk source grid to determine how the risk (especially the slip risk like liquid) spreads from the source, understand the overall trend of the surrounding micro-topography, rather than just the slope of a point.

[0080] In some embodiments, the acquisition of two-dimensional slope information can be achieved in several ways: alternatively, by querying a pre-constructed slope map layer: in the initialization or map construction phase, using the original point cloud data or elevation map, the two-dimensional slope information of each grid of the entire map is pre-calculated and stored as a separate two-dimensional slope vector array; when this step is triggered, according to the coordinates (x, y) of the risk source grid; directly access the slope array, read all the slope vector data in the 3x3 area centered on (x, y).

[0081] It can be understood that more accurate slope information can also be dynamically calculated by plane fitting on the point cloud in the area corresponding to the risk source grid directly from the original 3D point cloud data, and this is not limited here.

[0082] S206, constructing a local height map covering the corresponding grid according to the two-dimensional slope information, and determining the risk diffusion direction based on the local height map.

[0083] Among them, the local height map refers to a three-dimensional surface model that can reflect the micro-topography of a small area constructed according to the slope or height information of a risk source grid and its surrounding grids; the risk diffusion direction refers to the main path or direction predicted based on physical laws that the risk spreads from its source to the surrounding area and affects the surrounding area; the slope descending direction refers to the direction from the higher point to the lower point on the local height map, which simulates the influence of gravity on the motion trajectory of objects (especially liquids).

[0084] Specifically, after obtaining the slope information of the risk source and its neighborhood, first, a local, continuous or discrete height map is constructed according to these data to facilitate the analysis of the terrain. Then, check the reason why the grid was initially determined as a risk source (is it due to vibration risk or sliding risk), and enter a two-branch logical judgment process: Sliding risk source. If the grid is determined because there are super-threshold values in the adjacent sliding difference value group, analyze the local height map to find the steepest descending path from the risk source grid position. This direction, i.e. from high to low, is uniquely determined as the diffusion direction of the sliding risk.

[0085] Vibration risk source. If the grid is determined because the adjacent vibration difference value group is super-threshold, since a fixed road defect (such as a pit or protrusion) will pose a threat to AGVs approaching it from any direction. Therefore, the risk is considered to be non-directional and uniformly radiated in all directions. At this time, all directions from the risk source grid to all its directly adjacent grids are determined as the risk diffusion direction.

[0086] Through this intelligent differentiation processing, the prediction of the risk model is no longer blind, but has a high scene adaptability and physical reality.

[0087] In some embodiments, the determination and construction of the risk diffusion direction can be realized in various ways: optionally, a discrete height value comparison-based way: first, obtain the accurate height values of the risk source grid and its neighbors; for a sliding risk source, traverse the eight neighbors to find the neighbor with the lowest height value. Determine the direction vector (such as [1, -1]) from the center to the lowest neighbor as the diffusion direction; for a vibration risk source, directly generate a list containing eight direction vectors ([1, 0], [1, 1], [0, 1]...) as its diffusion direction.

[0088] Optionally, a gradient vector analysis-based way: for a sliding risk source, directly use the two-dimensional slope gradient vector (Gx, Gy) of the center grid obtained in S205; calculate its inverse direction vector (-Gx, -Gy), which represents the steepest slope descent direction; discretize this continuous vector, that is, find the one with the smallest angle among the eight basic direction vectors, as the final diffusion direction.

[0089] It can be understood that a more complex fluid simulation algorithm, such as a model based on a cellular automaton, can also be introduced for a sliding risk to simulate the flow path of a liquid on a complex microtopography to obtain a more refined diffusion direction graph, which is not limited here.

[0090] S207, determining a plurality of risk spread grids according to the risk diffusion direction.

[0091] Among them, the risk diffusion direction refers to the direction of one or more paths of the risk spreading from the source to the outside; the preset spread number is a parameter preset in advance, used to define the distance of risk spread, for example, if the preset number is 3, it means that the influence range of the risk will extend from the source to the length of 3 grids; the risk spread grid refers to the grid located in the determined risk diffusion direction and within the distance range defined by the preset spread number.

[0092] Specifically, taking each risk source grid determined in S204 as the starting point, for each risk source, obtain one or more risk diffusion directions assigned to it in S206. Then, for each diffusion direction, perform a ray tracing-like operation. Starting from the position of the risk source grid, move forward along the current diffusion direction by one grid distance, and mark the new grid reached as a risk spread grid. Repeat the moving and marking process until the total number of grids advanced along the direction reaches the preset spread number.

[0093] If a risk source has multiple diffusion directions (e.g. a vibration risk source), the above tracking and marking process is performed independently for each direction. After all risk sources and their diffusion directions are processed, one or more linear or radial risk propagation zones are formed on the map, and these marked grids collectively constitute the total area of risk impact.

[0094] In some embodiments, the determination of risk propagation grids can be achieved in various ways: optionally, a graph traversal algorithm is used: the grid map is regarded as an unweighted graph, and the grids are regarded as nodes; for each risk source, a temporary directed subgraph is constructed according to its diffusion direction, which only contains edges along the diffusion direction; starting from the risk source node, a breadth-first search (BFS) or depth-first search (DFS) is performed on this directed subgraph, with a maximum search depth limit, and the search depth is the preset propagation number. All traversed nodes are determined as risk propagation grids.

[0095] It can be understood that graphics techniques such as digital differential analyzer (DDA) or Bresenham line algorithm can also be used to handle non-standard, continuous-angle diffusion directions to determine the grids passed on the path, which is not limited here.

[0096] In some embodiments, after the risk diffusion direction is determined, the impact range can be further dynamically adjusted according to the risk of the risk source itself, so as to realize the differentiated modeling of different levels of risk sources, and to make the risk propagation range more consistent with physical intuition.

[0097] First, after the risk source grid is determined in S204, the reason for its determination is traced back. If the grid is marked because there is an over-threshold in its adjacent slip difference group, it is determined to be a slip-type risk source, and the slip value of the grid itself is directly obtained as the risk source intensity indicator. Otherwise, if it is marked because the adjacent vibration difference group is over-threshold, the vibration value will be used as the intensity indicator.

[0098] Secondly, the risk source intensity indicator obtained in the last step is taken as input to query a pre-configured mapping relationship to determine the dynamic risk propagation number. This mapping relationship is the core of the entire dynamic mechanism, which converts abstract risk values into specific physical ranges. This mapping relationship is usually designed to be monotonically increasing, i.e. the higher the risk source intensity, the greater the determined risk propagation number.

[0099] Finally, after the specific number of risk spread is determined, the final risk spread grid set is determined based on it. This process is similar to the original S207 step. Starting from the risk source, ray tracing is performed along one or more risk diffusion directions determined in S206, but the step size (i.e. the number of grids) of the tracing is no longer a fixed preset value, but the risk spread number just dynamically calculated. In this way, the size of the risk pollution area generated by each risk source is tailored, thereby realizing differentiated modeling of different levels of risk.

[0100] The above steps directly link the inherent intensity of the risk source with the influence range, achieving fine and differentiated modeling of risk. A small risk point (such as a shallow pit) will only generate a local and small-range avoidance zone, while a serious risk source (such as a large oil stain) will generate a large-range and high-priority no-entry zone. This risk level matching cost map correction method enables the path planning algorithm to make more accurate trade-offs, avoiding excessive avoidance of minor risks and neglecting major risks, thereby improving the adaptability of AGV path planning in complex physical road conditions.

[0101] S208, the vibration cost value and the slip cost value of the risk spread grid are multiplied by a preset risk coefficient.

[0102] The preset risk coefficient is a number multiplier greater than 1, which is used to artificially and proportionally increase the passing cost of the risk spread grid; the coefficient is not a fixed value, but decreases as the distance from the risk source grid increases. For example, the grid adjacent to the risk source may be multiplied by a higher coefficient, while the grid farther away may be multiplied by a lower coefficient, to simulate the natural phenomenon of gradual weakening of risk intensity from the center to the periphery.

[0103] Specifically, all grids marked as risk spread grids in S207 are traversed. For each risk spread grid, the distance between the current risk spread grid and the original risk source grid corresponding to it is calculated. This distance can be a simple Manhattan distance (|x1-x2|+|y1-y2|) or a more accurate Euclidean distance. According to the distance calculated in the previous step, a preset risk coefficient is determined from a preset attenuation function or lookup table. This function or table is designed such that the longer the distance, the smaller the coefficient value but always greater than 1. The original vibration cost value and slip cost value of the current risk spread grid are multiplied by the risk coefficient to obtain the corrected new values. Then the old values of the grid in the map are overwritten with the new values.

[0104] Once this process has been completed for all risk contagion grids, the originally isolated risk points have now become a risk contamination zone with a gradient and shape, and the degree of danger smoothly decreases from the center to the edge.

[0105] In some embodiments, the application of risk coefficients and cost updates can be implemented in several ways: Optionally, based on a Discrete Lookup Table (LUT): a predefined array or dictionary is used as the lookup table, with the index being distance and the value being the risk coefficient, for example, lut=[1.8, 1.5, 1.2] (index 0 is not used); for each diffuse grid cell, the Manhattan distance d between it and the source point is calculated; the coefficient coeff=lut[d] is retrieved from the lookup table, and the cost update new_cost=old_cost*coeff is performed.

[0106] Optionally, a continuous decay function-based approach is used: a mathematical formula is predefined to represent the decay relationship, such as exponential decay coeff = 1.0 + A*exp(-k*d), where A is the maximum additional coefficient, k is the decay rate, and d is the distance; for each diffuse grid cell, its Euclidean distance d to the source point is calculated; d is substituted into the formula to calculate the coeff value, and this value is used to update the cost. This approach can produce a smoother risk gradient.

[0107] Understandably, when a grid is affected by multiple risk sources at the same time, different strategies can be chosen to combine the effects, such as taking the maximum value among all calculated correction costs, or accumulating all additional risk values, without limitation here.

[0108] S209. When calculating the comprehensive passage cost of the candidate path, the vibration cost and slip cost of the modified risk spread grid are used.

[0109] Specifically, when it is necessary to calculate the comprehensive travel cost of any candidate path in the future (as in step S211), when the path planning algorithm (e.g., the A* algorithm) needs to query the vibration or slip value of a certain grid on the path, the cost map corrected by the risk contagion model is read and used.

[0110] By mandating the adoption of a new cost map, instead of just seeing isolated risk spikes, we can see shaped, gradient danger zones that spread from these spikes, allowing us to more intelligently avoid these zones with a greater safety margin.

[0111] In some embodiments, the adoption of the revised cost value can be implemented in various ways: optionally, an in-place update is adopted: when step S208 is executed, the modification is directly made on the data structure (such as a two-dimensional array) of the global cost map in the memory; when the subsequent path planning algorithm is started, the global cost map that has been modified is directly read without any special operation.

[0112] It can be understood that a more complex version control system can also be used to manage multiple revisions of the cost map, allowing switching and adopting different versions of the risk model in different scenarios, which is not limited here.

[0113] S210, after receiving the transportation task, obtaining the vibration weight and the slip weight corresponding to the transportation task.

[0114] Step S210 is similar to step S103 in the embodiment shown in the figure, and reference can be made to the description in step S103, which will not be repeated here. Figure 1 The step S103 in the embodiment shown in the figure is similar, and reference can be made to the description in step S103, which will not be repeated here.

[0115] In some embodiments, when processing a complex transportation task containing multiple work links such as picking up goods and unloading goods, the multi-site task can also be decomposed, and the risk weight of path planning can be dynamically adjusted according to the expected load state of each road section, to realize the control of risk preference in different transportation stages.

[0116] Specifically, this embodiment is aimed at a more complex transportation task containing multiple intermediate stops (for example, picking up goods from point A, transporting to point B, picking up another piece of goods from point B, and transporting to point C), and a segmented planning strategy is adopted.

[0117] When a multi-site task is received, the entire long task is decomposed into a series of continuous point-to-point transportation road sections according to the order of each work site in the task instruction. For example, a task of "A-B-C" will be decomposed into two independent transportation road sections: road section 1 (A-B) and road section 2 (B-C). This decomposition is the basis for all subsequent dynamic adjustments, because it logically splits a complex task that may change state multiple times into multiple simple tasks with constant states.

[0118] For each decomposed transportation segment, the expected load state of the AGV when driving on the segment is determined in advance. This is achieved by analyzing the operation instructions of the start and end stations of the segment. For example, for segment 1 (A-B), if the operation instruction of station A is to pick up goods, it can be explicitly inferred that the physical state of the AGV during the entire process of leaving point A to go to point B will be full load. Similarly, if the instruction of station B is to unload goods, the expected load state of the AGV will be determined as empty load when planning the next segment 2 (B-C). This step successfully converts the semantic information at the task level (such as "pick up goods" and "unload goods") into a physical state description at the vehicle level.

[0119] After determining the expected load state of a certain segment, the state is used as a query key to find the corresponding real-time vibration weight and real-time slip weight in the preset load weight mapping table. The mapping table is pre-calibrated based on vehicle dynamics and goods safety requirements. Full load state: goods safety and stability are the primary consideration, and the tolerance to jolts is extremely low. Therefore, the real-time vibration weight is set very high, and the real-time slip weight is relatively low, because the vehicle body is heavier and has stronger grip when full load, and the risk of slipping is inherently reduced; empty load state: the vehicle body is lighter and less sensitive to jolts, but it may be more prone to slipping due to excessive driving wheel torque on some road surfaces. Therefore, the real-time vibration weight is reduced, and the real-time slip weight is correspondingly increased to preferentially select paths with greater friction.

[0120] According to the order of the segments, path planning and driving are performed one by one. When planning segment 1 (A-B), the weight corresponding to the full load state is used to calculate the global comprehensive traffic cost map, and an optimal path is planned. After the AGV drives to point B and completes the operation, the path is completed. Then, switch to segment 2 (B-C), use the new weight corresponding to the empty load state to recalculate a completely new comprehensive traffic cost map, and on this basis, plan a completely new optimal path from B to C. The entire multi-station task is gradually completed in the cycle of planning-execution-replanning-re-execution.

[0121] This method decomposes the macroscopic transportation task into multiple segments closely related to the physical state of the vehicle, and assigns the most appropriate risk assessment model (i.e. weight) to each segment, so that the path decision is no longer static and unchangeable. This dynamic and segmented optimization strategy allows path planning to respond to the shift in risk preference brought about by changes in vehicle load, thereby improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0122] S211, for the plurality of candidate paths of the transportation task, respectively, the product of the path length cost of each grid passed by each candidate path, the vibration value and the vibration weight, and the product of the sliding value and the sliding weight are added up to obtain the comprehensive passing cost corresponding to the candidate path.

[0123] S212, in the plurality of candidate paths, the candidate path with the lowest comprehensive passing cost is selected as the final execution path.

[0124] Steps S211-S212 are similar to steps S104-S105 in the embodiment shown in Figure 1 The steps S104-S105 in the embodiment shown in

[0125] By using the above technical solution, the local gradient change of the risk value is analyzed, and the risk source grid which is both a local peak value and has a sharp change is identified. Further, according to the physical type (such as sliding or vibration) of the risk and the local terrain information, the diffusion direction conforming to the physical law is matched for the risk source, which makes the prediction of the risk model have a real basis. Subsequently, by applying a risk coefficient decreasing with distance to the grids on the diffusion path, the originally isolated risk point is converted into a continuous risk field with smooth gradient and reasonable shape on the cost map. This way from point risk to surface risk with physical meaning enables the path planning algorithm to foresee and bypass the entire dangerous area rather than just one point, ultimately improving the adaptability of AGV path planning in the face of complex physical road conditions.

[0126] The following introduces an exemplary path selection system 300 provided by an embodiment of the present application. Figure 3 is an exemplary hardware structure schematic diagram of the path selection system 300 provided by an embodiment of the present application.

[0127] In some embodiments, the path selection system 300 is a computer device or includes a computer device in the path selection system 300. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control ability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with other terminals or servers outside through network connection. In some embodiments, the network interface can be a wired network interface, and in some embodiments, the network interface can also be a wireless network interface. The computer program is executed by the processor to implement the method in the embodiments of the present application.

[0128] Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0129] The above-described embodiments are only used to illustrate the technical scheme of the present application, but not limit it; although the technical scheme recorded in the foregoing embodiments has been described in detail, those skilled in the art should understand that they can still modify the technical scheme recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical scheme deviate from the scope of the technical scheme of the embodiments of the present application.

[0130] In the above-described embodiments, the term "when" can be interpreted as "if" or "after" or "in response to determining" or "in response to detecting" according to the context. Similarly, the phrase "upon determining" or "if detecting (the stated condition or event)" can be interpreted as "if determining" or "in response to determining" or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)" according to the context.

[0131] In the above-described embodiments, all or part of the technical scheme can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the technical scheme can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as DVD), or semiconductor media (such as solid state disk), etc.

[0132] Those skilled in the art can understand that all or part of the processes in the above-mentioned method embodiments can be implemented by a computer program instructing relevant hardware to complete, the program can be stored in a computer readable storage medium, and the program can include the processes of the above-mentioned method embodiments when executed. The aforementioned storage medium includes ROM or random storage memory RAM, magnetic disc or optical disc and various storage code medium.

Claims

1. A path selection method for AGV transportation equipment, characterized in that, include: Acquire vertical vibration data of vertical motion changes during the operation of the automated guided vehicle, first velocity data representing wheel rotation speed, and second velocity data representing actual displacement changes; A grid map is created according to a preset grid size. When the automated guided vehicle (AGV) travels to each grid, physical state data is calculated and continuously updated for the corresponding grid. The physical state data is obtained by associating vibration cost, slip cost, and the current position value of the AGV. The vibration cost is calculated by analyzing the frequency and amplitude characteristics of the vertical vibration data. The slip cost value is obtained by calculating the difference between the first velocity data and the second velocity data; After receiving the transportation task, obtain the vibration weight and slip weight corresponding to the transportation task; For the multiple candidate paths of the transportation task, the path length cost of each grid passed through by each candidate path, the product of the vibration cost and the vibration weight, and the product of the slip cost and the slip weight are added together to obtain the comprehensive passage cost of the corresponding candidate path. Among the multiple candidate paths, the candidate path with the lowest overall travel cost is selected as the final execution path.

2. The method according to claim 1, characterized in that, After summing the path length cost, the product of the vibration cost and the vibration weight, and the product of the slip cost and the slip weight for each grid along the multiple candidate paths of the transportation task to obtain the comprehensive passage cost of the corresponding candidate path, the method further includes: Calculate the single-point risk value of each grid in each candidate path; the single-point risk value is the product of the vibration cost value and the slip cost value of the corresponding grid. Each candidate path selects grid cells whose single-point risk value exceeds a preset threshold as risk mutation points; The isolation degree of the corresponding risk mutation point is obtained by calculating the difference between the single-point risk value of the risk mutation point and the average risk value of the adjacent grids; the average risk value of the adjacent grids is the arithmetic mean of the single-point risk values ​​of all adjacent grids of the risk mutation point. When the risk mutation point exists in the candidate path, the comprehensive travel cost of the corresponding candidate path is corrected to the sum of the product of the comprehensive travel cost and the single-point risk value and isolation degree of the corresponding risk mutation point.

3. The method according to claim 2, characterized in that, After calculating the difference between the single-point risk value of the risk mutation point and the average risk value of the adjacent grid cells to obtain the isolation degree of the corresponding risk mutation point, the method further includes: When the direction of travel of the automated guided vehicle changes at a risk mutation point, the steering angle value is obtained; When the steering angle value is less than a preset first angle threshold, the isolation degree is multiplied by a preset first correction coefficient; When the steering angle value is not less than the first angle threshold and less than the preset second angle threshold, the isolation degree is multiplied by the preset second correction coefficient; When the steering angle value is not less than the second angle threshold, the isolation degree is multiplied by a preset third correction coefficient; the preset first angle threshold is less than the preset second angle threshold; the preset first correction coefficient is less than the preset second correction coefficient, and the preset second correction coefficient is less than the preset third correction coefficient.

4. The method according to claim 1, characterized in that, After establishing a grid map according to a preset grid size, and calculating and continuously updating the physical state data for each grid cell as the automated guided vehicle travels to that grid cell, the process further includes: Calculate the vibration cost and slip cost of each grid cell in the grid map with the adjacent vibration difference set and adjacent slip difference set of all adjacent grid cells; When there is a value in the adjacent vibration difference group or the adjacent slip difference group that exceeds the preset difference threshold, and the vibration cost or slip cost of the corresponding grid is greater than all the corresponding adjacent grids, the corresponding grid is identified as a risk source grid. Obtain the two-dimensional slope information of the risk source grid and all adjacent grids; A local height map covering the corresponding grid is constructed based on the two-dimensional slope information, and the risk diffusion direction is determined based on the local height map. The risk diffusion direction is determined as follows: when there is a value exceeding a preset difference threshold in the adjacent slip difference group, the slope descent direction is determined as the direction from high to low as determined by the local height map; when there is a value exceeding a preset difference threshold in the adjacent vibration difference group, the directions of all adjacent grids of the risk source grid are determined as the risk diffusion direction. Multiple risk spread grids are determined based on the risk spread direction; the risk spread grids are adjacent grids that spread a predetermined number of times along the risk spread direction; The vibration cost and slip cost of the risk propagation grid are multiplied by a preset risk coefficient, respectively; the preset risk coefficient decreases as the distance from the risk source grid increases. When calculating the overall travel cost of the candidate paths, the vibration cost and slip cost of the modified risk spread grid are used.

5. The method according to claim 4, characterized in that, The determination of multiple risk spread grids based on the risk spread direction specifically includes: The vibration cost or slip cost of the risk source grid is obtained as a risk source intensity index. The risk source intensity index is determined as follows: when there is a value in the adjacent slip difference group that exceeds a preset difference threshold, the slip cost is determined as the risk source intensity index; when there is a value in the adjacent vibration difference group that exceeds a preset difference threshold, the vibration cost is determined as the risk source intensity index. Based on the risk source intensity index, the number of risk spread grids is determined by querying the preset mapping relationship between the risk source intensity index and the number of risk spread grids; Multiple adjacent grids along the direction of risk spread, determined by the dynamic number of risk spreads, are identified as the risk spread grid.

6. The method according to claim 1, characterized in that, After calculating and continuously updating the physical state data for the corresponding grid, the method further includes: The corresponding grid cells whose vibration cost exceeds a preset frequency threshold and whose slip cost is greater than a preset slip initiation threshold are marked as coupling risk grid cells; Add a preset coupling risk cost to the total passage cost of all the coupled risk grids.

7. The method according to claim 1, characterized in that, After receiving the transportation task, the following is also included: When the transportation task is a multi-site task that includes multiple work stations, the multi-site task is decomposed into multiple consecutive transportation segments according to the order of the work stations. For each transport segment, the expected load status of the transport vehicle is pre-determined based on the operation instructions of the starting and ending work stations. Based on the expected load state, the real-time vibration weight and real-time slip weight are searched in the preset load weight mapping table. The real-time vibration weight and the real-time slip weight are used to independently calculate and execute the path with the lowest comprehensive passage cost for each of the transport segments until the entire multi-site task is completed.

8. A path selection system, characterized in that, The path selection system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the path selection system to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on the path selection system, the path selection system performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the path selection system, the path selection system performs the method as described in any one of claims 1-7.

Images