Path planning method, device, equipment and storage medium for omnidirectional AGV

Abstract

The present application relates to the technical field of AGV path planning, and discloses a path planning method, apparatus, equipment and storage medium for an omnidirectional AGV, the method comprising: obtaining a dynamic environment state matrix and environment change rate data within a stereo garage; calculating an instantaneous queue length based on the task arrival rate and AGV processing capacity data within the stereo garage operation cycle, and obtaining instantaneous task queue parameters and queue dissipation prediction time; performing composite trigger condition judgment on AGV position deviation and environmental variables, and obtaining a dynamic event trigger signal and task priority parameters; inputting the dynamic event trigger signal and task priority parameters into an instantaneous task allocation model for solution, and obtaining an optimal task allocation scheme and a corresponding target path set. The present application can accurately identify and quantify the task queues formed in a short period of time during the operation of the stereo garage, thereby improving the robustness and adaptability of path planning.

Description

Technical Field

[0001] The present application relates to the technical field of AGV path planning, and in particular to a path planning method, device, equipment and storage medium for an omnidirectional AGV. Background Art

[0002] Stereoscopic garages are widely used as a key solution to parking difficulties. Omnidirectional AGVs, due to their flexible motion characteristics, have become a key piece of equipment for vehicle handling in these garages. However, traditional AGV path planning methods are primarily designed for static or quasi-static environments, using pre-calculated fixed paths for navigation. Faced with the complex and dynamic environment of stereoscopic garages, these methods often fail to respond promptly to environmental changes, resulting in delayed path planning and inefficient execution. This makes it difficult to meet the high-frequency, high-density vehicle handling requirements of modern stereoscopic garages.

[0003] The internal environment of a multi-story parking garage is highly dynamic. Factors such as the interference caused by multiple AGVs operating simultaneously, real-time changes in parking space status, the sudden appearance of obstacles, and the random arrival of tasks often create transient task queues within a short period of time, resulting in significant queuing delays. Existing path planning methods, which mostly perform static optimization based on average queue length, ignore the significant impact of transient queueing on system performance. They are unable to accurately predict and handle these task queues that form and dissipate within a short period of time, resulting in significant deviations between path planning decisions and the actual operating environment. Summary of the Invention

[0004] The present application provides a path planning method, device, equipment and storage medium for an omnidirectional AGV. The present application can accurately identify and quantify the task queues formed in a short period of time during the operation of a stereo garage, thereby improving the robustness and adaptability of path planning.

[0005] A first aspect of the present application provides a path planning method for an omnidirectional AGV, the path planning method for an omnidirectional AGV comprising:

[0006] Obtain the dynamic environment state matrix and environment change rate data in the stereo garage;

[0007] Calculating the instantaneous queue length based on the task arrival rate and AGV processing capacity data within the operation cycle of the stereo garage according to the dynamic environment state matrix and the environment change rate data, and obtaining instantaneous task queue parameters and queue dissipation prediction time;

[0008] Based on the instantaneous task queue parameters and the queue dissipation prediction time, the composite trigger condition judgment of the AGV position deviation and the environmental variables is performed to obtain a dynamic event trigger signal and a task priority parameter;

[0009] The dynamic event trigger signal and the task priority parameter are input into the instantaneous task allocation model for solution to obtain an optimal task allocation solution and a corresponding target path set.

[0010] In combination with the first aspect, in a first implementation of the first aspect of the present invention, obtaining the dynamic environment state matrix and environment change rate data in the stereoscopic parking garage includes:

[0011] Real-time data collection is performed on the lidar sensors, ultrasonic sensors, and visual sensors of each parking space on each floor of the stereo garage to obtain obstacle location information, AGV real-time location information, parking space occupancy status information, and channel accessibility information;

[0012] Constructing a dynamic environment state matrix based on the obstacle position information, the AGV real-time position information, the parking space occupancy status information, and the channel passability information;

[0013] The state difference calculation within the continuous time window is performed based on the dynamic environment state matrix to obtain environment change rate data.

[0014] In combination with the first aspect, in a second implementation of the first aspect of the present invention, the instantaneous queue length is calculated based on the task arrival rate and AGV processing capacity data within the stereo parking garage operation cycle according to the dynamic environment state matrix and the environment change rate data to obtain the instantaneous task queue parameters and queue dissipation prediction time, including:

[0015] Based on the dynamic environment state matrix and the environment change rate data, statistics are collected on the handling task flow within the operation cycle of the stereo garage to obtain task arrival rate and AGV processing capacity data;

[0016] Calculating the maximum instantaneous queue length based on the task arrival rate and the AGV processing capacity data to obtain instantaneous task queue length data and a queue delay weight coefficient;

[0017] Calculating task demand fluctuation intensity based on the instantaneous task queue length data and the queue delay weight coefficient to generate instantaneous task queue parameters;

[0018] A queue dissipation time prediction calculation is performed based on the instantaneous task queue parameters to obtain a queue dissipation prediction time.

[0019] In combination with the first aspect, in a third implementation of the first aspect of the present invention, the calculation of the task demand fluctuation intensity based on the instantaneous task queue length data and the queue delay weight coefficient to generate the instantaneous task queue parameters includes:

[0020] Performing instantaneous queuing effect quantitative calculation on the instantaneous task queue length data and the queue delay weight coefficient to obtain queuing effect quantitative data;

[0021] Based on the queuing effect quantitative data, historical task arrival rate sequence analysis is performed to obtain the task arrival rate average value and standard deviation data;

[0022] Calculating the task demand fluctuation intensity variance based on the task arrival rate average value and the standard deviation data to obtain a comparison result between the task demand fluctuation intensity data and the fluctuation threshold;

[0023] An instantaneous queuing effect is determined based on the comparison result between the task demand fluctuation intensity data and the fluctuation threshold, and an instantaneous task queue parameter is generated.

[0024] In combination with the first aspect, in a fourth implementation of the first aspect of the present invention, the composite trigger condition judgment of the AGV position deviation and the environmental variables based on the instantaneous task queue parameters and the queue dissipation prediction time is performed to obtain a dynamic event trigger signal and a task priority parameter, including:

[0025] Performing Euclidean distance calculation on the actual position of the AGV and the planned position to obtain the AGV position deviation, and performing weighted summation of environmental variables based on the instantaneous task queue parameters to obtain environmental variables;

[0026] Dynamically adjust the adaptive threshold according to the ratio of the queue dissipation prediction time to the operation cycle time to obtain a trigger threshold adjustment coefficient;

[0027] Comparing and judging the composite trigger conditions based on the AGV position deviation, the environmental variables, and the trigger threshold adjustment coefficient to obtain a dynamic event trigger signal;

[0028] Priority evaluation is performed on the AGV position deviation, the environmental variables, and the task urgency data to obtain a task priority parameter.

[0029] In combination with the first aspect, in a fifth implementation of the first aspect of the present invention, inputting the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solving to obtain the optimal task allocation solution and the corresponding target path set includes:

[0030] Based on the dynamic event trigger signal, a first fixed-point problem of garage road network loading is constructed and AGV flow distribution is solved to obtain a flow distribution vector; based on the task priority parameter, a second fixed-point problem of garage road network loading is constructed and task allocation vector is solved to obtain a task allocation vector;

[0031] Calculating the path selection probability of the traffic distribution vector and the task allocation vector using an instantaneous task allocation model to obtain task selection probability distribution data;

[0032] The task selection probability distribution data is input into the variational inequality solver for cost mapping calculation, and path decoding and task allocation mapping calculation are performed under the constraints of the feasible solution set to obtain the optimal task allocation solution and the corresponding target path set.

[0033] In combination with the first aspect, in a sixth implementation of the first aspect of the present invention, performing path selection probability calculation on the traffic distribution vector and the task allocation vector using an instantaneous task allocation model to obtain task selection probability distribution data includes:

[0034] Inputting the traffic distribution vector and the task allocation vector into an instantaneous task allocation model, and calculating the path utility function value of each candidate path by using the path travel time, instantaneous queuing delay and task completion benefit to obtain path utility data;

[0035] Determine the probability distribution adjustment coefficient according to the degree of decision uncertainty of the AGV handling task in the stereo garage;

[0036] performing an exponential transformation calculation based on the path utility data and the probability distribution adjustment coefficient to obtain an exponential utility value of each candidate path;

[0037] The normalized probability operation is performed on the indexed utility value of each candidate path to obtain the task selection probability distribution data.

[0038] A second aspect of the present application provides a path planning device for an omnidirectional AGV, the path planning device for the omnidirectional AGV comprising:

[0039] An acquisition module is used to obtain the dynamic environment state matrix and environment change rate data in the stereo garage;

[0040] A calculation module is used to calculate the instantaneous queue length based on the task arrival rate and AGV processing capacity data within the operation cycle of the three-dimensional parking garage according to the dynamic environment state matrix and the environment change rate data, and obtain the instantaneous task queue parameters and queue dissipation prediction time;

[0041] A judgment module is used to perform a composite trigger condition judgment on the AGV position deviation and the environmental variables based on the instantaneous task queue parameters and the queue dissipation prediction time, and obtain a dynamic event trigger signal and a task priority parameter;

[0042] The solution module is used to input the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solution to obtain the optimal task allocation solution and the corresponding target path set.

[0043] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned omnidirectional AGV path planning method.

[0044] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned omnidirectional AGV path planning method.

[0045] Compared with the existing technology, the present application has the following beneficial effects: by constructing a dynamic environment state matrix and calculating the environment change rate data, it can accurately perceive the changes in multi-dimensional environmental information such as obstacle position, AGV status, parking space occupancy, etc. in the stereo garage in real time, and significantly improve the adaptability to complex dynamic environments compared with the traditional static map method. The establishment of an instantaneous queue length calculation mechanism and a queue dissipation prediction algorithm can accurately identify and quantify the task queues formed in a short period of time during the operation of the stereo garage, overcoming the limitation of the existing technology that only considers the average queue, and realizing the accurate modeling and processing of the instantaneous queuing effect. A composite trigger condition judgment mechanism based on AGV position deviation and environmental variables is designed to start path replanning only when necessary, avoiding the continuous calculation mode of the traditional method, and significantly reducing the consumption of computing resources while ensuring the timeliness of the system response. The variational inequality solving framework combined with the instantaneous task allocation model can simultaneously consider multiple decision dimensions such as AGV flow distribution, task allocation and path selection, and realize the global coordinated optimization of the multi-AGV system in the stereo garage. By dynamically adjusting the trigger threshold based on the queue dissipation prediction time and calculating the probability distribution based on the path utility function, the system can adaptively adjust key parameters according to the actual operating conditions, thereby improving the robustness and adaptability of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] The structures, proportions, sizes, etc. depicted in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not intended to limit the conditions under which the present invention can be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportional relationships, or adjustments in size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and objectives that can be achieved by the present invention.

[0048] Figure 1 Schematic diagram of the process of the path planning method of the omnidirectional AGV provided by the embodiment of the present invention;

[0049] Figure 2 1 is a schematic block diagram of the structure of a path planning device for an omnidirectional AGV provided in an embodiment of the present invention;

[0050] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0053] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0054] It should be further understood that the term "and / or" used in this specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In one embodiment of the present application, an embodiment of the path planning method for an omnidirectional AGV includes:

[0055] Step 100: Obtain the dynamic environment state matrix and environment change rate data in the stereo garage;

[0056] It is understandable that the execution subject of the present application can be a path planning device of an omnidirectional AGV, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0057] Specifically, a multi-source, heterogeneous perception system is constructed. LIDAR, ultrasonic, and visual sensors are deployed at key locations within parking spaces and aisles, leveraging the multi-layered structure of the multi-story parking garage. This enables high-frequency, low-latency data collection of key environmental elements. LIDAR uses precise ranging capabilities to capture the spatial distribution and boundary contours of obstacles in the garage in real time. Ultrasonic sensors assist in detecting the presence of obstacles at close range and at the end of aisles. Visual sensors identify occupied parking spaces and the position and posture of AGVs in different lanes, enhancing structured scene understanding through image segmentation and object detection algorithms. The collected raw sensor data is preprocessed by an edge computing unit to eliminate noise and data drift, and then uniformly converted to a standard spatial coordinate system for cross-sensor fusion. Based on this, an algorithm module maps four types of information—obstacle location, real-time AGV position, parking space occupancy status, and aisle accessibility—into a standard state vector. This constructs the current environmental state matrix, which describes the accessibility and operational status of the entire multi-story parking garage at the current time. To capture the dynamic changes in the environment, the state matrix is recorded in a sliding time window, and the difference between the state matrices at two consecutive moments is calculated to form rate-of-change data. The difference operation uses an element-by-element norm calculation strategy to quantify the magnitude of the state change at each state monitoring point. If some components in the rate-of-change data are significantly larger than other regions, it indicates that a structural dynamic change has occurred in that area, such as the addition of obstacles, parking space status changes, or AGV trajectory deviations. This information is used as a pre-judgment for subsequent path adjustments and task reallocation. The environmental state matrix and its corresponding rate-of-change data together constitute the core perception data input layer of the path planning system.

[0058] Step 200: Calculate the instantaneous queue length based on the task arrival rate and AGV processing capacity data within the stereo garage operation cycle according to the dynamic environment state matrix and the environment change rate data, and obtain the instantaneous task queue parameters and queue dissipation prediction time;

[0059] Specifically, based on a dynamic environmental state matrix and environmental change rate data, the task flow in each area of the parking garage during its operation cycle is continuously counted. Combined with the real-time position of the AGV and its historical records of completed tasks in each time window, the task request frequency and execution capacity of different channels and work areas within each time period are identified, thereby calculating the temporal distribution characteristics of the task arrival rate and AGV processing capacity. Furthermore, the maximum queue length is estimated based on the instantaneous difference between the task arrival rate and the AGV processing capacity. The queue delay weight coefficient is extracted based on historical task accumulation. This weight reflects the impact of system delay on overall operation efficiency at different queue lengths. Based on the statistically obtained instantaneous queue length data and the corresponding queue delay weight coefficient, the intensity of task demand fluctuations is quantitatively assessed. A sliding window mechanism is used to analyze the degree of dramatic fluctuations in task flow over time, generating instantaneous task queue parameters that reflect the system's current load state and processing stability. This dynamically characterizes the risk level of task congestion at different time points. Combining core indicators such as the current queue length, historical environmental status change trends, and task demand fluctuation intensity, a queue dissipation time prediction model is constructed to simulate the time period for the natural dissipation of the task queue under the current system operation status. Based on this, the queue dissipation prediction time is output to determine whether the current task accumulation is a short-term burst or will exist for a long time.

[0060] Based on existing environmental state analysis, the instantaneous task queue length data and queue delay weight coefficients are jointly calculated to quantitatively represent the queuing effect. A queuing effect quantification model is constructed, combining the instantaneous task queue lengths within each time period with the corresponding delay impact. This queuing effect quantification data reflects the severity of task congestion. This data reflects the current task backlog and its potential impact on overall operational efficiency. Based on this queuing effect quantification data, a statistical analysis of historical task arrival rate series is performed. By tracing the task flow fluctuation trend over a period of time, the mean and standard deviation of this series are extracted to characterize the task generation pattern and fluctuation amplitude of the garage task system within a typical operating cycle. The variance of task demand fluctuation intensity is calculated based on the mean and standard deviation of the task arrival rate data to quantify the degree of deviation of the current task arrival pattern from the historical fluctuation range, generating task demand fluctuation intensity data representing the current fluctuation intensity. This task demand fluctuation intensity data is compared with a preset fluctuation threshold. By determining whether it exceeds the threshold, the presence of sudden and significant task load changes is identified, thus completing the determination of the instantaneous queuing effect. If the current fluctuation intensity significantly exceeds the threshold, the current state will be marked as a high-risk or task congestion state requiring intervention, and instantaneous task queue parameters containing comprehensive information such as queue length, delay impact, and fluctuation degree will be generated.

[0061] Step 300: Based on the instantaneous task queue parameters and the queue dissipation prediction time, the composite trigger condition judgment is performed on the AGV position deviation and the environmental variables to obtain the dynamic event trigger signal and the task priority parameter;

[0062] It should be noted that a composite event perception mechanism that integrates spatial deviation and system load status is constructed. During the specific implementation process, the actual operation trajectory of the AGV is continuously obtained and compared with its corresponding path planning scheme in real time. By calculating the Euclidean distance between the actual position of the AGV and the planned position, the spatial position deviation value of the AGV at the current moment is obtained. This deviation value reflects whether the AGV has experienced path deviation or delay due to sudden obstacles, channel blockage or task overlap. At the same time, the queue effect data, task fluctuation intensity index and queue congestion level information in the instantaneous task queue parameters are used as weight factors to perform a weighted summation of the variables in the environment that affect the execution stability of the AGV, forming an environmental variable indicator that comprehensively describes the dynamic complexity of the current operation area. This environmental variable not only reflects the current task density, but also reflects the potential traffic risk of the area where the AGV is located. To achieve adaptive adjustment of trigger conditions under different load conditions, a trigger threshold adjustment coefficient is dynamically calculated based on the ratio between the currently predicted queue dissipation time and the standard operation cycle time. This coefficient is used to scale the original deviation threshold in real time. When the system is in a high-load, long-queue state, the adjustment coefficient will automatically decrease, making the system more sensitive to small deviations. Conversely, when the queue is about to dissipate naturally, the system tolerance is relatively improved to avoid unnecessary frequent triggering. The AGV position deviation, environmental variables, and trigger threshold adjustment coefficient are jointly input into the trigger condition judgment module, and a composite comparison operation is performed to determine whether the dynamic event conditions for triggering path replanning or task reallocation have been met. If so, a dynamic event trigger signal is immediately output, providing a decision-making start instruction for the downstream module. In order to reasonably guide the direction of resource allocation, a comprehensive evaluation is conducted based on the current AGV position deviation, environmental variables, and urgency data related to task scheduling, and a task priority parameter is constructed. This parameter is used to quantify the response urgency and resource allocation weight of the task, and to sort multiple AGVs so that AGVs in high-urgency and high-congestion risk areas are given priority in path correction and task adjustment support, thereby ensuring that the omnidirectional AGV system can still maintain the stability of operating order and scheduling efficiency in a multi-source disturbance environment.

[0063] Step 400: Input the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solution to obtain the optimal task allocation solution and the corresponding target path set.

[0064] Specifically, a loading model for the lane network within a multi-story parking garage is constructed based on local environmental disturbance points marked by dynamic event trigger signals. Based on this, a first fixed-point problem is proposed to solve the AGV traffic flow distribution vector under current environmental constraints. This vector reflects the number of AGVs and flow intensity that each lane can carry at the path execution level. The solution must consider factors such as channel congestion, accessibility boundaries, and historical path usage frequency. A second fixed-point problem is constructed using task priority parameters to match different levels of handling tasks to currently available AGV resources. The task allocation vector is then solved by combining task location, task urgency, and execution time window to determine the task unit that each AGV should prioritize at a specific moment. After obtaining these two core vectors, the traffic flow distribution vector and task allocation vector are input into the instantaneous task allocation model. The path selection probability module evaluates the AGV's tendency to select each candidate path. Factors such as historical execution success rate, path distance cost, and current traffic risk are incorporated into the calculation process. Task selection probability data is output as a probability distribution, reflecting the system's preference for each path scheduling solution. The task selection probability distribution data is fed into a variational inequality solver for cost mapping. This module analyzes the optimal task allocation solution for the current system state by establishing a comprehensive cost mapping relationship that includes path selection cost, queue interference cost, turning loss, and traffic conflicts. During the solution process, the system performs path decoding and task mapping, ensuring that the resulting path set is structurally connected, unique in terms of tasks, and balanced in terms of load. Ultimately, the optimal task allocation solution and target path set are obtained, which can be directly distributed to the AGV system under the current garage state.

[0065] The traffic distribution vector and task allocation vector are input into the instantaneous task allocation model, which incorporates multi-factor path attribute information to achieve a comprehensive utility evaluation of candidate paths. During this process, a numerical calculation of the path utility function is performed for each possible path. This utility function consists of three parts: path travel time, instantaneous queue delay, and system benefit after task completion. Among them, path travel time reflects the actual running time required for the AGV to travel from the current position to the task point via the path. The instantaneous queue delay is adjusted according to the current path congestion status and is used to measure the execution waiting time that the path may face. The task completion benefit represents the positive contribution of the task to the overall task completion rate, task level requirement satisfaction, or work efficiency after execution. After the three items are combined, the path utility data for each candidate path is formed, which serves as the basic score for subsequent probabilistic reasoning. To adapt to the uncertainties inherent in task allocation in diverse operational scenarios, the decision-making uncertainty in path selection for AGV transport tasks in a multi-story parking garage is estimated by considering factors such as the current task complexity, the intensity of competition between AGVs, task urgency, and the frequency of dynamic path changes. This uncertainty is then used to determine a probability distribution adjustment coefficient, which directly influences the distribution of each path's selection probability. After determining this adjustment coefficient, an exponential transformation is performed on the path utility data. Each path's utility value is then exponentially mapped to the adjustment coefficient scale, generating utility data with exponentially nonlinear characteristics. This amplifies the advantages of optimal paths and reduces the selection probability of suboptimal paths, resulting in a more decision-biased final path distribution while maintaining randomness. To ensure the standardization and mathematical validity of the output probabilities, the exponentially weighted values of all candidate paths are normalized. The exponential weights of all paths are summed, and each path's exponential value is then divided by the sum to obtain the task selection probability distribution data that conforms to the probability distribution definition.

[0066] In the embodiment of the present application, by constructing a dynamic environment state matrix and calculating the environment change rate data, it is possible to accurately perceive the changes in multi-dimensional environmental information such as the position of obstacles, AGV status, parking space occupancy, etc. in the stereo garage in real time. Compared with the traditional static map method, it significantly improves the adaptability to complex dynamic environments. The establishment of an instantaneous queue length calculation mechanism and a queue dissipation prediction algorithm can accurately identify and quantify the task queues formed in a short period of time during the operation of the stereo garage, overcome the limitation of the existing technology that only considers the average queue, and realize the accurate modeling and processing of the instantaneous queuing effect. A composite trigger condition judgment mechanism based on AGV position deviation and environmental variables is designed to start path replanning only when necessary, avoiding the continuous calculation mode of the traditional method, and significantly reducing the consumption of computing resources while ensuring the timeliness of system response. The variational inequality solving framework combined with the instantaneous task allocation model can simultaneously consider multiple decision dimensions such as AGV flow distribution, task allocation and path selection, and realize the global coordinated optimization of the multi-AGV system in the stereo garage. By dynamically adjusting the trigger threshold based on the queue dissipation prediction time and calculating the probability distribution based on the path utility function, the system can adaptively adjust key parameters according to the actual operating conditions, thereby improving the robustness and adaptability of path planning.

[0067] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0068] Real-time data collection is performed on the lidar sensors, ultrasonic sensors, and visual sensors of each parking space on each floor of the stereo garage to obtain obstacle location information, AGV real-time location information, parking space occupancy status information, and channel accessibility information;

[0069] Construct a dynamic environment state matrix based on obstacle location information, AGV real-time location information, parking space occupancy status information, and channel accessibility information;

[0070] The state difference within the continuous time window is calculated based on the dynamic environment state matrix to obtain the environment change rate data.

[0071] Specifically, targeted sensor clusters are deployed on each floor of the multi-story parking garage, enabling each monitoring unit to not only perceive static scene states but also respond to dynamic events at high frequency. LiDAR sensors construct a three-dimensional spatial structure, actively scanning the surrounding environment horizontally and vertically using high-precision laser beams. This data captures spatial information such as obstacle boundaries, AGV outlines, and wall structures within the garage. This data offers high ranging accuracy and strong resistance to light interference, making it suitable for capturing the location and dynamic changes of obstacle outlines. Meanwhile, ultrasonic sensors are used for close-range blind spot detection, particularly in narrow aisles, around AGVs, or at the edges of parking spaces, where LiDAR is limited by obstructions. Ultrasonic sensors provide stable distance data, making them suitable for identifying localized obstacle formations and sudden interference sources. Vision sensors utilize embedded image recognition algorithms to capture video frame data and, using deep learning models such as YOLO or Mask R-CNN, identify semantic information such as parking space occupancy status, AGV type, line tracking, and light recognition. This enables the system to possess multi-scale image recognition capabilities at the parking space, vehicle, and aisle levels. Data from these three sensor types undergo fusion preprocessing at the edge computing node, including time synchronization, coordinate transformation, data filtering, and spatial alignment. This process maps data of varying precision and dimensionality into a unified structured data model, forming a standardized input. Based on this preprocessed multi-source data, spatial reconstruction and semantic decoding are performed on obstacle location information, AGV real-time location information, parking space occupancy status information, and channel accessibility information. Obstacle location information is converted into a visually enclosed region within the obstacle area using point cloud clustering and boundary tracking algorithms. Real-time AGV location information is estimated by integrating IMU, odometry, and visual positioning methods to determine the global coordinates, orientation, and speed of each AGV at a specific point in time. Parking space occupancy status information is generated by an image recognition model that continuously detects parking space outlines and coverage areas in the video stream, determines the presence of vehicles, and generates occupancy marker data. Channel accessibility is determined by combining these three types of data: if an obstacle, a stalled AGV, or insufficient lighting is detected within the channel, the channel is marked as impassable; otherwise, it is considered open. This information is quantized into a state vector with a consistent encoding structure in a spatial coordinate system. This state vector collection represents the dynamic environmental state matrix of the parking garage at a specific moment. Each row or column represents a monitoring unit, such as a parking space, an intersection, or a passageway. The vector dimensions include multi-dimensional information such as obstacle occupancy rate, vehicle position offset, state update frequency, and passable signs. This state matrix is continuously updated over time as events such as vehicles entering and exiting the garage, AGVs performing tasks, and changes in passageway conditions occur. To achieve systemic awareness of dynamic environmental changes, state differences within continuous time windows are calculated based on this dynamic environmental state matrix.The system caches the state matrices of several consecutive moments in a time series database, constructing a sequence of state trajectories. At each new moment, it extracts the difference between the current matrix and the matrix at the previous moment or multiple historical moments. Using vector-level or matrix-level interpolation operations, it extracts the magnitude of the state change at each monitoring point. This interpolation process is not limited to numerical calculations but also includes switching between state classifications, such as when a parking space changes from occupied to vacant or a passage changes from unobstructed to blocked. These semantic changes are also recorded through interpolation. To improve the robustness of change detection, a sliding average filter and an outlier rejection mechanism are introduced to prevent misjudgments due to single-point false alarms or sudden interference. Through these interpolation operations, a continuous rate of change data set is formed, which serves as the core content of the environmental change rate data and reflects the dynamic fluctuation intensity and spatial distribution pattern of the stereo parking garage per unit time.

[0072] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0073] Based on the dynamic environment state matrix and environment change rate data, the handling task flow in the stereo garage operation cycle is counted to obtain the task arrival rate and AGV processing capacity data;

[0074] The maximum instantaneous queue length is calculated based on the task arrival rate and AGV processing capacity data to obtain the instantaneous task queue length data and queue delay weight coefficient;

[0075] Based on the instantaneous task queue length data and queue delay weight coefficient, the task demand fluctuation intensity is calculated to generate the instantaneous task queue parameters;

[0076] The queue dissipation time prediction calculation is performed based on the instantaneous task queue parameters to obtain the queue dissipation prediction time.

[0077] Specifically, task event trigger points are extracted from the constructed environmental state matrix, and the number of new transport requests generated per unit time within a specific operation cycle is counted. These task events originate from user-initiated vehicle pickup or parking instructions, system-issued automatic adjustment instructions, or task inheritance between AGVs. Each task is reflected in the state matrix as a change in vehicle occupancy, aisle state switching, or a sudden change in the AGV's dispatch trajectory. Combined with environmental change rate data, these tasks are determined to be sudden demands caused by environmental disturbances or normal, periodic task loads. Based on this, statistics are then partitioned by time window to obtain the task arrival rate, a fundamental metric reflecting the intensity of task generation. Furthermore, by analyzing the actual number of tasks completed by AGVs within the same operation cycle, average response time, available operating time, and path reachability, combined with the task types and historical performance of each AGV, a dynamic modeling of AGV processing capacity is constructed, generating data on AGV processing capacity corresponding to the task arrival rate. This processing capacity is not only constrained by the number of AGVs but also by path accessibility, traffic density, load capacity, and dispatch system efficiency. Therefore, the system integrates all dynamic environmental indicators to perform a multi-factor calculation, ensuring that processing capacity data is predictable and instructive for actual dispatch. The maximum instantaneous queue length is calculated based on task arrival rate and AGV processing capacity data to determine whether task density exceeds the system's processing threshold. By comparing the difference between the current task generation rate per unit time and the AGV processing capacity, the maximum instantaneous queue length the system can handle under the current configuration and environment is estimated. This queue length is further correlated with historical delays, the number of stranded AGVs, and task waiting times. The delay cost of each unit of task accumulation is calculated, resulting in a queue delay weight coefficient. This coefficient reflects the negative impact of task queuing on overall system performance, with higher weights indicating a more significant impact on system delay. Based on this data, the intensity of task demand fluctuations is calculated. This process continuously monitors the changing trend of queue length within a sliding time window and, combined with the queue delay weight coefficient, amplifies or suppresses fluctuations, generating quantifiable task demand fluctuation intensity data. This fluctuation intensity is used to describe changes in the number of tasks, and it also takes into account the uneven processing caused by environmental changes and AGV congestion, thereby more realistically reflecting the stability and adaptability of the garage task system. On this basis, the system generates instantaneous task queue parameters. These parameters integrate multiple factors such as the current queue size, queue penalty weight, and demand fluctuation intensity, and serve as key inputs for subsequent scheduling priority determination, path adjustment, and load shifting. Using the obtained instantaneous task queue parameters, combined with the current AGV distribution, the number of feasible paths, and environmental state trends, a queue dissipation time prediction model is established to simulate the natural decay trajectory of the existing task queue state without introducing new intervention measures.The prediction calculation takes into account multiple influencing factors such as the average processing rate of AGV, channel smoothness, task processing order and task urgency distribution. By deducing the system's ability and speed to automatically process tasks in the future time period, the time required to completely clear the estimated task backlog is calculated, thereby forming a queue dissipation prediction time.

[0078] In a specific embodiment, the step of performing task demand fluctuation intensity calculation based on the instantaneous task queue length data and the queue delay weight coefficient to generate the instantaneous task queue parameters may specifically include the following steps:

[0079] The instantaneous queue effect is quantitatively calculated based on the instantaneous task queue length data and queue delay weight coefficient to obtain the queue effect quantitative data;

[0080] Based on the queuing effect quantitative data, historical task arrival rate sequence analysis was performed to obtain the average and standard deviation data of task arrival rate;

[0081] The variance of task demand fluctuation intensity is calculated based on the mean value and standard deviation data of task arrival rate, and the comparison results of task demand fluctuation intensity data and fluctuation threshold are obtained;

[0082] The instantaneous queuing effect is determined based on the comparison results of task demand fluctuation intensity data and fluctuation threshold, and the instantaneous task queue parameters are generated.

[0083] Specifically, the instantaneous queuing effect is quantified using instantaneous task queue length data and queue delay weight coefficients. The AGV task queue length is obtained during each time sampling period. This queue length reflects the system's backlog of uncompleted tasks. The delay weight coefficient is derived from previously statistically analyzing the impact of task completion delays on overall performance. A higher weight indicates a more significant impact of queuing on overall system efficiency. The queue length is multiplied by the delay weight to generate quantified queuing effect data. Based on this quantified queuing effect data, a historical task arrival rate series analysis is conducted to obtain a stable reference value that serves as a baseline for assessing volatility. This analysis traces back task trigger records within a specific time range and counts the number of task arrivals in each time slice, forming a time series data set. The mean and standard deviation of this set are then calculated to determine the central tendency and dispersion of the task arrival rate, respectively. The mean represents the average load level of the system under stable operating conditions, while the standard deviation reflects the variability in the frequency of task generation. These values are important indicators for determining whether the system is experiencing periodic fluctuations, abnormal surges, or sudden load changes. The variance of task demand fluctuation intensity is calculated based on the mean and standard deviation of task arrival rates to quantify the degree of difference between the current task system operating state and the long-term average. This fluctuation intensity is calculated by squaring the deviation between the actual task arrival rate and the historical mean within the current time period and normalizing it with the standard deviation ratio to obtain a numerical representation that truly reflects the severity of the system's current load fluctuation. The inclusion of this fluctuation intensity data can reveal whether the system is experiencing a short-term shock or a long-term anomaly, allowing for the development of appropriate response strategies. After obtaining the task demand fluctuation intensity data, it is compared with a preset fluctuation threshold. The fluctuation threshold is determined by the system's maximum acceptable load variation range, scheduling fault tolerance, and response delay tolerance during historical operations, and serves as the demarcation point for determining whether the system requires rescheduling or path reconstruction. The final assessment of the current instantaneous queuing effect is based on the comparison between the task demand fluctuation intensity data and the threshold. If the fluctuation intensity exceeds the threshold, it indicates a serious imbalance between the current task generation and the system's processing capacity. The system then marks the current state as abnormally loaded or high-risk queuing and generates new instantaneous task queue parameters accordingly. The parameters include instantaneous queue length, delay impact coefficient, fluctuation intensity value, judgment label and subsequent trigger priority.

[0084] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0085] The Euclidean distance between the actual position and the planned position of the AGV is calculated to obtain the AGV position deviation, and the environmental variables are weighted summed based on the instantaneous task queue parameters to obtain the environmental variables;

[0086] According to the ratio of queue dissipation prediction time to operation cycle time, the adaptive threshold is dynamically adjusted to obtain the trigger threshold adjustment coefficient;

[0087] Compare and judge the composite trigger conditions based on AGV position deviation, environmental variables and trigger threshold adjustment coefficient to obtain dynamic event trigger signals;

[0088] The AGV position deviation, environmental variables and task urgency data are prioritized to obtain the task priority parameters.

[0089] Specifically, a dynamic event triggering mechanism is constructed, focusing on monitoring AGV execution deviations and incorporating environmental changes, system load, and task urgency as criteria. This mechanism is ensured to have comprehensive awareness of operational deviations, traffic complexity, and scheduling pressure. The Euclidean distance between the AGV's actual and planned positions is calculated to determine the spatial deviation of the AGV's current position relative to the ideal path. A larger position deviation indicates a greater degree of interference, obstruction, or path instability during task execution, a significant precursor to decreased system scheduling efficiency or increased risk of path failure. Furthermore, environmental factors are weighted and aggregated based on the current instantaneous task queue parameters to construct an environmental variable that reflects the system's external pressure and environmental interference. This process uses factors such as instantaneous task queue length, task arrival density, channel impedance, and queue delay weight as basic inputs. A weighted summation model is used to integrate these factors into a single environmental complexity index. This environmental variable comprehensively characterizes the degree of traffic congestion, resource competition intensity, and task conflict density within the current AGV's path or region. Because environmental variables are continuously influenced by dynamic task fluctuations and AGV location distribution, the calculation process must be periodically updated to ensure that each trigger decision is based on the latest data. To adapt the trigger mechanism to varying work paces and load intensities, a dynamic threshold adjustment mechanism is introduced based on the ratio between the predicted queue clearance time and the standard work cycle time. If the system predicts that the current task backlog will resolve naturally in the short term, there is no need for immediate path reconstruction or task reallocation. Therefore, the system automatically relaxes the trigger threshold to avoid unnecessary scheduling operations caused by minor deviations. Conversely, if the forecast indicates that task backlog persists or even worsens, the system lowers the trigger threshold, increasing sensitivity to operational deviations and environmental anomalies, allowing for early intervention and scheduling adjustments. This adaptive threshold adjustment process dynamically generates a trigger threshold adjustment coefficient based on the ratio between the queue clearance time and the work cycle time output by the prediction model. This coefficient is then used to modify the standard for subsequent judgment conditions. The AGV position deviation, environmental variables, and trigger threshold adjustment coefficients are input into the composite condition comparison module, which performs a multi-dimensional fusion judgment. The specific logic is to perform a weighted superposition of the position deviation value and the environmental variables, and then compare it with the adjusted dynamic trigger threshold. If the weighted result exceeds the threshold, the current system operation is determined to have deviated from the normal scheduling trajectory, and a dynamic event trigger signal must be immediately generated. This signal is input into the task allocation model or path replanning module to initiate the system response chain. To implement the resource priority scheduling mechanism, the calculated AGV position deviation, environmental variables, and the urgency data of each task are jointly analyzed to construct a task priority assessment model. In this model, task urgency is determined by multiple factors, including the interval between the task request time and the current time, the importance of the target vehicle, the user's reservation time limit, or the garage in and out flow control strategy.The system uniformly quantifies the three types of input data and assigns different evaluation weights, performs multi-factor linear or nonlinear combinations, and forms numerical task priority parameters.

[0090] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0091] Based on the dynamic event trigger signal, the first fixed-point problem of garage road network loading is constructed and the AGV flow distribution is solved to obtain the flow distribution vector. Based on the task priority parameter, the second fixed-point problem of garage road network loading is constructed and the task allocation vector is solved to obtain the task allocation vector.

[0092] The path selection probability is calculated for the traffic distribution vector and the task allocation vector using the instantaneous task allocation model to obtain the task selection probability distribution data.

[0093] The task selection probability distribution data is input into the variational inequality solver for cost mapping calculation. At the same time, path decoding and task allocation mapping calculation are performed under the constraints of the feasible solution set to obtain the optimal task allocation plan and the corresponding target path set.

[0094] Specifically, based on the dynamic event trigger signal, the key areas or key nodes in the current three-dimensional parking garage network that require path replanning or task adjustment are identified, and based on this, a structured model of the garage road network is constructed. This road network model transforms the current AGV spatial distribution, the set of accessible paths, the channel status, the parking space nodes, and the path connectivity into a network graph structure, where each edge represents an actual channel, each node represents a task target or an AGV state transition point, and the edge weight represents complex information such as travel cost, delay risk, or traffic pressure. When the dynamic event trigger signal is determined to be valid by the system, the system constructs the first fixed-point problem based on the current trigger position and the scope of influence, thereby analyzing the AGV traffic intensity carried by each channel in the current garage network and forming a dynamic mapping of the traffic load in the local area. To this end, the system loads the current AGV's location, destination, and traffic status information into a graphical model. The distribution of AGVs along all possible paths is used as a variable. By setting constraints on path load, accessibility, circulation, and congestion, a problem is established to solve the AGV traffic distribution problem under the current state. Using iterative approximation or a network balancing algorithm, the current traffic distribution vector for each channel is obtained. This vector characterizes the resource utilization level of the AGV on each path and serves as an important basis for subsequent task scheduling weights and path conflict determination. Based on the obtained traffic distribution vector, a second fixed-point problem is constructed based on the task priority parameter. This problem takes the urgency of the task demand, the resource contention status, and the availability of the AGV as inputs, and aims to maximize the matching degree of task assignments to solve the optimal task assignment vector. In the modeling process, each task is considered a resource unit to be allocated, and each AGV is a task executor. The task priority parameter serves as a weight coefficient for each AGV-task pair, expressing the degree of matching in the current environment. The system constructs feasible assignments based on the edges between AGV nodes and task nodes in the graph structure and performs optimization calculations while satisfying traffic constraints and task uniqueness. This results in a task assignment vector, which describes which AGV is most likely to execute each task and along which path. The traffic distribution vector and task assignment vector are input into the instantaneous task assignment model for inferential calculation of path selection probabilities. Based on the utility value of each candidate path, this model constructs a path utility function that combines path travel time, current traffic volume, delay risk, and the system benefit from task completion. Taking into account the relationship between task assignments, an exponential mapping strategy is used to calculate the selection probability of each path under each task-AGV assignment combination. This probability reflects the system's preference for scenarios with multiple paths and intense resource competition under the current environmental conditions and is a key control variable in nondeterministic scheduling systems. The system uses normalization to constrain the path selection probabilities to probability distributions that sum to one. The task selection probability distributions are then input into a variational inequality solver for cost mapping.This solver constructs a system cost mapping relationship, combining factors such as energy consumption, time consumption, blocking risk, queuing effects, and resource contention along the task execution path into a unified cost function. This transforms the task allocation problem into an optimal solution problem subject to inequality constraints. The variational inequality solution allows the system to find the optimal scheduling state under nonlinear and multi-constrained conditions, avoiding the failure or computational inefficiency of traditional linear optimization methods in complex environments. During this process, the system simultaneously loads a set of feasible solutions—the complete set of all legal path combinations and task-AGV matching combinations—and performs path decoding within this set, assigning numbers and bindings to specific paths based on selection probabilities. Task mapping is then performed based on the task assignment vector and the path encoding data, mathematically encapsulating the mapping from task requirements to path resources. The system outputs an optimal task allocation solution and a corresponding target path set, while satisfying all path access constraints, task allocation constraints, and load balancing constraints. The optimal task allocation solution describes which tasks each AGV should undertake, while the target path set contains the physical path sequence and control instructions corresponding to each task, which the scheduling system then issues to the AGV controller for execution.

[0095] In a specific embodiment, the execution step of calculating the path selection probability of the traffic distribution vector and the task allocation vector using the instantaneous task allocation model to obtain task selection probability distribution data may specifically include the following steps:

[0096] The traffic distribution vector and task allocation vector are input into the instantaneous task allocation model, and the path utility function value of each candidate path is calculated by the path travel time, instantaneous queue delay and task completion benefit to obtain the path utility data;

[0097] Determine the probability distribution adjustment coefficient according to the degree of decision uncertainty of the AGV handling task in the stereo garage;

[0098] Based on the path utility data and the probability distribution adjustment coefficient, an exponential transformation calculation is performed to obtain the indexed utility value of each candidate path;

[0099] The normalized probability operation is performed on the indexed utility value of each candidate path to obtain the task selection probability distribution data.

[0100] Specifically, the AGV traffic distribution vector and task allocation vector are used as input, representing the AGV usage density per unit time in each lane of the current parking garage and the execution resource allocation relationship bound to each task, respectively. Upon receiving these two vectors, the instantaneous task allocation model performs a comprehensive evaluation of each candidate path in the set of optional paths. The evaluation focuses on three dimensions: path travel time, which is derived from a comprehensive prediction of historical path execution time, current traffic saturation, and local path topology. A shorter path indicates a shorter time required for an AGV to travel from its current location to its target task point. Second, instantaneous queue delay, which is estimated in real time based on the number of AGVs on the current path, available lane width, the number of conflicting nodes, and a queue depth model. It provides quantitative compensation for unexpected events such as waiting, deceleration, and re-planning that AGVs may encounter during execution. Third, task completion benefit, which is derived from the system target gain after task completion as specified in the task priority parameter, such as its contribution to garage entry and exit efficiency, reduction in user waiting time, or improvement in system balance. A higher value indicates a greater impact of the task corresponding to the path on the global system state. The three indicators are weighted and combined to form the path utility function value, or path utility data. A numerical scoring system is established for each path based on the current environment and scheduling status, reflecting the path's overall attractiveness in scheduling selection. The higher the utility value, the better the path's overall performance in terms of time efficiency, smooth traffic flow, and system benefits. Considering the uncertainty factors in the actual scheduling process, such as multiple path options, multi-task competition, and multi-AGV concurrency, a probability distribution mechanism is introduced to express path selection preferences in a non-deterministic form. To this end, the decision uncertainty of AGV handling tasks in a stereoscopic parking garage is evaluated, and a probability distribution adjustment coefficient is determined accordingly. This coefficient serves as an adjustment factor for the steepness of the path selection probability curve, which can control the scheduling system's trade-off between "approaching the optimal path" and "preserving the randomness of multiple paths." If uncertainty is high, such as the presence of a large number of temporary tasks, frequent changes in path status, or uncontrollable AGV status, the adjustment coefficient should be set low to maintain the dispersion and redundancy of path selection. Conversely, if the system is stable, paths are unobstructed, and the scheduling objectives are clear, the adjustment coefficient should be set high to increase the preference for the optimal path. To determine the adjustment coefficient, an exponential transformation is performed on the aforementioned path utility data. Each path's utility value is used as the input to an exponential function, with the path utility value as the independent variable. The exponential function is scaled based on the adjustment coefficient, thereby amplifying the advantages of high-utility paths and suppressing the probability weights of low-utility paths. This transformation results in a set of indexed utility values that reflect the scheduling attractiveness of each path under the control of the adjustment factor. The indexed utility values of all candidate paths are normalized.Sum up all the path index values, and then divide the indexed value of each path by the sum to ensure that the sum of the selection probabilities of all paths is one, and obtain the task selection probability distribution data that meets the probability definition.

[0101] In this embodiment, the task selection probability distribution data is input into the variational inequality solver for cost mapping calculation, and path decoding and task allocation mapping calculation are performed under the constraints of the feasible solution set to obtain the optimal task allocation scheme and the corresponding target path set, including: constructing a variational inequality mathematical expression based on the task selection probability distribution data, establishing the variational inequality problem by finding the optimal solution vector that satisfies the inequality constraints, and obtaining the variational inequality constraints; performing a mapping function calculation on the path cost including the instantaneous queuing effect according to the variational inequality constraints to obtain the cost mapping function value; performing an intersection operation on the cost mapping function value and the path constraints of the stereo parking garage to obtain the boundary range of the feasible solution set; performing iterative calculation based on the boundary range of the feasible solution set using an adaptive step-size continuous weighted average algorithm, updating the next round of iterative solution through a weighted combination of the current solution vector and the historical solution vector to obtain a converged solution vector; performing path decoding and task allocation mapping calculation on the converged solution vector to obtain the optimal task allocation scheme and the corresponding target path set.

[0102] The above describes the path planning method of the omnidirectional AGV in the embodiment of the present application. The following describes the path planning device of the omnidirectional AGV in the embodiment of the present application. Figure 2 In one embodiment of the present application, a path planning device for an omnidirectional AGV includes:

[0103] An acquisition module 11 is used to obtain the dynamic environment state matrix and environment change rate data in the stereo garage;

[0104] The calculation module 12 is used to calculate the instantaneous queue length based on the task arrival rate and AGV processing capacity data within the operation cycle of the stereo garage according to the dynamic environment state matrix and the environment change rate data, and obtain the instantaneous task queue parameters and queue dissipation prediction time;

[0105] The judgment module 13 is used to judge the composite trigger conditions of the AGV position deviation and the environmental variables based on the instantaneous task queue parameters and the queue dissipation prediction time, and obtain the dynamic event trigger signal and the task priority parameter;

[0106] The solving module 14 is used to input the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solving, so as to obtain the optimal task allocation solution and the corresponding target path set.

[0107] Through the collaborative efforts of these components, by constructing a dynamic environmental state matrix and calculating environmental change rate data, the system can accurately perceive changes in multi-dimensional environmental information within the parking garage, including obstacle locations, AGV status, and parking space occupancy, in real time. Compared to traditional static mapping methods, this approach significantly improves its adaptability to complex dynamic environments. A mechanism for calculating instantaneous queue length and a queue dissipation prediction algorithm are established to accurately identify and quantify task queues that form within a short period of time during parking garage operations. This overcomes the limitation of existing technologies that only consider average queues, enabling precise modeling and processing of instantaneous queue effects. A composite trigger condition judgment mechanism based on AGV position deviations and environmental variables is designed to initiate path replanning only when necessary, avoiding the continuous calculation mode of traditional methods. This significantly reduces computational resource consumption while ensuring timely system response. A variational inequality solution framework combined with an instantaneous task allocation model simultaneously considers multiple decision-making dimensions, including AGV flow distribution, task allocation, and path selection, achieving global coordinated optimization of a multi-AGV system in a parking garage. By dynamically adjusting the trigger threshold based on the queue dissipation prediction time and calculating the probability distribution based on the path utility function, the system can adaptively adjust key parameters according to the actual operating conditions, thereby improving the robustness and adaptability of path planning.

[0108] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0109] The non-volatile storage medium may store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 may execute any of the above-mentioned path planning methods for the omnidirectional AGV.

[0110] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0111] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned omnidirectional AGV path planning methods.

[0112] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device 300 involved in the solution of the present application. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0113] It should be understood that the processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0114] It should be noted that, those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned omnidirectional AGV path planning method, and will not be repeated here.

[0115] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the path planning method for the omnidirectional AGV provided in the embodiment of the present application.

[0116] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash memory card, etc., equipped with the electronic device 300.

[0117] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0118] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0119] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A path planning method for an omnidirectional AGV, characterized in that: include: Obtain the dynamic environment state matrix and environment change rate data in the stereo garage; The instantaneous queue length is calculated based on the task arrival rate and AGV processing capacity data within the operation cycle of the stereoscopic parking garage according to the dynamic environment state matrix and the environment change rate data, and the instantaneous task queue parameters and queue dissipation prediction time are obtained; specifically comprising: statistics are collected on the handling task flow within the operation cycle of the stereoscopic parking garage based on the dynamic environment state matrix and the environment change rate data to obtain the task arrival rate and AGV processing capacity data; the maximum instantaneous queue length is calculated based on the task arrival rate and the AGV processing capacity data to obtain the instantaneous task queue length data and the queue delay weight coefficient; the task demand fluctuation intensity is calculated based on the instantaneous task queue length data and the queue delay weight coefficient to generate the instantaneous task queue parameters; the queue dissipation time is predicted and calculated based on the instantaneous task queue parameters to obtain the queue dissipation prediction time; Based on the instantaneous task queue parameters and the queue dissipation prediction time, the composite trigger condition judgment of the AGV position deviation and the environmental variables is performed to obtain a dynamic event trigger signal and a task priority parameter; The dynamic event trigger signal and the task priority parameter are input into the instantaneous task allocation model for solution to obtain an optimal task allocation solution and a corresponding target path set.

2. The path planning method for an omnidirectional AGV according to claim 1, characterized in that: The method of obtaining the dynamic environment state matrix and environment change rate data in the stereo garage includes: Real-time data collection is performed on the lidar sensors, ultrasonic sensors, and visual sensors of each parking space on each floor of the stereo garage to obtain obstacle location information, AGV real-time location information, parking space occupancy status information, and channel accessibility information; Constructing a dynamic environment state matrix based on the obstacle position information, the AGV real-time position information, the parking space occupancy status information, and the channel passability information; The state difference calculation within the continuous time window is performed based on the dynamic environment state matrix to obtain environment change rate data.

3. The path planning method for an omnidirectional AGV according to claim 1, characterized in that: The calculating of task demand fluctuation intensity based on the instantaneous task queue length data and the queue delay weight coefficient to generate instantaneous task queue parameters includes: Performing instantaneous queuing effect quantitative calculation on the instantaneous task queue length data and the queue delay weight coefficient to obtain queuing effect quantitative data; Based on the queuing effect quantitative data, historical task arrival rate sequence analysis is performed to obtain the task arrival rate average value and standard deviation data; Calculating the task demand fluctuation intensity variance based on the task arrival rate average value and the standard deviation data to obtain a comparison result between the task demand fluctuation intensity data and the fluctuation threshold; An instantaneous queuing effect is determined based on the comparison result between the task demand fluctuation intensity data and the fluctuation threshold, and an instantaneous task queue parameter is generated.

4. The path planning method for an omnidirectional AGV according to claim 1, characterized in that: The composite trigger condition judgment of the AGV position deviation and environmental variables based on the instantaneous task queue parameters and the queue dissipation prediction time to obtain a dynamic event trigger signal and a task priority parameter includes: Performing Euclidean distance calculation on the actual position of the AGV and the planned position to obtain the AGV position deviation, and performing weighted summation of environmental variables based on the instantaneous task queue parameters to obtain environmental variables; Dynamically adjust the adaptive threshold according to the ratio of the queue dissipation prediction time to the operation cycle time to obtain a trigger threshold adjustment coefficient; Comparing and judging the composite trigger conditions based on the AGV position deviation, the environmental variables, and the trigger threshold adjustment coefficient to obtain a dynamic event trigger signal; Priority evaluation is performed on the AGV position deviation, the environmental variables, and the task urgency data to obtain a task priority parameter.

5. The path planning method for an omnidirectional AGV according to claim 1, characterized in that: The step of inputting the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solving to obtain the optimal task allocation solution and the corresponding target path set includes: Based on the dynamic event trigger signal, a first fixed-point problem of garage road network loading is constructed and AGV flow distribution is solved to obtain a flow distribution vector; based on the task priority parameter, a second fixed-point problem of garage road network loading is constructed and task allocation vector is solved to obtain a task allocation vector; Calculating the path selection probability of the traffic distribution vector and the task allocation vector using an instantaneous task allocation model to obtain task selection probability distribution data; The task selection probability distribution data is input into the variational inequality solver for cost mapping calculation, and path decoding and task allocation mapping calculation are performed under the constraints of the feasible solution set to obtain the optimal task allocation solution and the corresponding target path set.

6. The path planning method for an omnidirectional AGV according to claim 5, characterized in that: The calculating the path selection probability of the traffic distribution vector and the task allocation vector by using the instantaneous task allocation model to obtain task selection probability distribution data includes: Inputting the traffic distribution vector and the task allocation vector into an instantaneous task allocation model, and calculating the path utility function value of each candidate path by using the path travel time, instantaneous queuing delay and task completion benefit to obtain path utility data; Determine the probability distribution adjustment coefficient according to the degree of decision uncertainty of the AGV handling task in the stereo garage; performing an exponential transformation calculation based on the path utility data and the probability distribution adjustment coefficient to obtain an exponential utility value of each candidate path; The normalized probability operation is performed on the indexed utility value of each candidate path to obtain the task selection probability distribution data.

7. A path planning device for an omnidirectional AGV, characterized in that: A method for executing a path planning method for an omnidirectional AGV according to any one of claims 1 to 6, wherein the path planning device for the omnidirectional AGV comprises: An acquisition module is used to obtain the dynamic environment state matrix and environment change rate data in the stereo garage; A calculation module is used to calculate the instantaneous queue length based on the task arrival rate and AGV processing capacity data within the operation cycle of the three-dimensional parking garage according to the dynamic environment state matrix and the environment change rate data, and obtain the instantaneous task queue parameters and queue dissipation prediction time; A judgment module is used to perform a composite trigger condition judgment on the AGV position deviation and the environmental variables based on the instantaneous task queue parameters and the queue dissipation prediction time, and obtain a dynamic event trigger signal and a task priority parameter; The solution module is used to input the dynamic event trigger signal and the task priority parameter into the instantaneous task allocation model for solution to obtain the optimal task allocation solution and the corresponding target path set.

8. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the path planning method for the omnidirectional AGV according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the path planning method for the omnidirectional AGV according to any one of claims 1 to 6 is implemented.

Images