AGV intelligent control method and system based on end-cloud cooperation

Abstract

The invention relates to the technical field of AGV intelligent control, and discloses an AGV intelligent control method and system based on end-cloud collaboration, and the method comprises the steps: obtaining the position, obstacle and environment data of an AGV in real time, constructing a dynamic space map, carrying out the grid processing, and when the channel width is detected to be not enough for safe passing, carrying out the grid processing of the dynamic space map; the method comprises the following steps: firstly, adopting an A * algorithm to quickly generate an alternative path, then carrying out multi-AGV cooperative conflict detection through path overlapping comparison, carrying out global optimization on a scheduling sequence and a passing priority of multiple AGVs by applying a genetic algorithm, and finally generating a conflict-free efficient scheduling instruction. According to the method, through multi-algorithm fusion and a real-time decision-making mechanism, the passing efficiency and the space resource utilization rate of multiple AGVs in a dynamic environment are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for AGVs, and in particular to an intelligent control method and system for AGVs based on end-to-cloud collaboration. Background Technology

[0002] In the fields of modern logistics and intelligent manufacturing, Automated Guided Vehicles (AGVs) are core equipment for unmanned material handling and are widely used in warehouses, factories, and other scenarios. However, with the increasing complexity of industrial scenarios, the space management and dynamic scheduling capabilities of AGV systems still need to be improved.

[0003] In one existing technology, a central dispatch system receives task instructions from an upper-level system. Based on a global map and the real-time status of all AGVs, it uses the A* algorithm to calculate an optimal path from the starting point to the target point for a specified AGV and transmits the path instructions to the AGV via a wireless communication network. The AGV's onboard controller then takes over control, with navigation sensors performing real-time positioning and path tracking while continuously monitoring the surrounding environment. The controller calculates precise speed and steering control values ​​based on the deviation between the current position and the target path, driving the motors to execute the movement. However, this method struggles to handle real-time changes, such as temporary adjustments to cargo volume or the appearance of sudden obstacles. When handling multiple AGVs working collaboratively, it often lacks the ability to respond quickly to emergencies, such as the inability to rapidly adjust paths to ensure unobstructed passageways during unexpected events.

[0004] Therefore, existing technologies suffer from the problem of congestion caused by overlapping and conflicting path planning when multiple AGVs are running collaboratively, leading to a decrease in operating efficiency. Summary of the Invention

[0005] This invention provides an intelligent control method and system for AGVs based on end-to-cloud collaboration, which solves the problem of congestion caused by overlapping and conflicting path planning when multiple AGVs are running collaboratively, resulting in decreased operating efficiency.

[0006] Firstly, to address the aforementioned technical problems, this invention provides an intelligent control method for AGVs based on end-to-cloud collaboration, comprising: Acquire real-time status data including the current position of the AGV, warehouse area data, and obstacle positions; perform path planning based on the current position of the AGV and the obstacle positions to obtain updated real-time status data containing the new path task. The warehouse area data in the updated real-time status data is divided into grids to obtain a dynamic spatial map; If the channel width in the dynamic spatial map is less than the preset channel width threshold, then alternative paths are calculated to obtain the adjusted AGV running path sequence. The expected occupied areas of other AGVs are obtained from the adjusted AGV running path sequence, and path overlap is compared to obtain the conflict detection results; Based on the conflict detection results, a genetic algorithm is used to adjust the scheduling order of multiple AGVs to obtain a priority queue; The scheduling instruction for the next AGV is obtained from the priority queue and transmitted to the corresponding AGV device to obtain an execution confirmation signal; The execution confirmation signal is transmitted to the target AGV device, and the cargo handling is completed according to the execution confirmation signal.

[0007] In one optional implementation, the step of performing path planning based on the current position of the AGV and the position of the obstacle to obtain updated real-time status data containing the new path task includes: The location distribution of goods is obtained. If the current position of the AGV deviates from the preset path, the AGV movement path is readjusted to obtain a new path task. Based on the new path task and the location of the obstacle, the A* algorithm is used to calculate the obstacle avoidance path of the AGV to obtain the optimized movement trajectory. Based on the optimized movement trajectory and the cargo location distribution, the K-means algorithm is used to cluster the cargo storage area to obtain the optimal storage location. Adjust the AGV scheduling task according to the optimal storage location to obtain updated real-time status data.

[0008] In one optional implementation, the step of dividing the warehouse area data into grids based on the updated real-time status data to obtain a dynamic spatial map includes: Obtain warehouse area data from the updated real-time status data, and divide the warehouse area data into grids to obtain gridded areas; The dynamic spatial map is obtained by calculating the channel width and corner occupancy based on the gridded area.

[0009] In one optional implementation, if the channel width in the dynamic spatial map is less than a preset channel width threshold, then alternative paths are calculated to obtain an adjusted AGV running path sequence, including: When the width of the passage in the dynamic spatial map is less than the sum of the cargo volume and the safety buffer, the map data is acquired and the passage width parameter is extracted to obtain the spatial constraint conditions. Based on the spatial constraints, the feasible paths are calculated using the A* algorithm to obtain a set of candidate paths. The candidate paths are then sorted according to path length and travel time to obtain a sorted path sequence. The Dijkstra algorithm is used to dynamically adjust the paths in the sorted path sequence to obtain the optimized path sequence. Based on the optimized path sequence, navigation planning is performed on the AGV operation instructions to obtain the AGV operation task sequence; If a dynamic obstacle is detected when the AGV operation task sequence is executed, the location data of the dynamic obstacle is obtained, and the new spatial constraints are recalculated to obtain the adjusted AGV operation path sequence.

[0010] In one optional implementation, the step of obtaining the expected occupied areas of other AGVs from the adjusted AGV running path sequence and performing path overlap comparison to obtain conflict detection results includes: Obtain the adjusted AGV running path sequence of multiple AGVs, compare whether the boundary coordinates of the expected occupied area and the emergency passage reserved area intersect, and obtain the area overlap judgment result; When the region overlap judgment result is true, the path planning data of the relevant AGV is extracted, the specific location of the conflict area is determined, and the K-means clustering algorithm is used to classify the conflict area to obtain the core location point of the conflict; when the region overlap judgment result is false, the conflict detection result is obtained directly, and the task is executed to the next step. Based on the core location of the conflict, the expected overlap of the occupied areas of each AGV is recalculated to obtain the conflict detection result.

[0011] In one optional implementation, the step of adjusting the scheduling order of multiple AGVs using a genetic algorithm based on the conflict detection results to obtain a priority queue includes: Obtain traffic area density data, and analyze the AGV movement demand in traffic areas exceeding the preset number of traffic areas based on the traffic area density data and the conflict detection results to obtain a demand priority list; Based on the aforementioned priority list of requirements, a genetic algorithm is used to adjust the scheduling order of AGVs, resulting in an optimized set of scheduling schemes. The running efficiency of the optimized scheduling scheme set is calculated to obtain an efficient scheduling scheme; When the efficient scheduling scheme meets the preset efficiency threshold, the output is a priority queue, and the final AGV scheduling order is obtained. The final AGV scheduling order is optimized according to the actual traffic conditions to obtain the priority queue.

[0012] In one optional implementation, the step of obtaining the scheduling instruction for the next AGV from the priority queue and transmitting it to the corresponding AGV device to obtain an execution confirmation signal includes: The scheduling instructions are extracted from the priority queue and the instruction data to be sent is obtained according to the priority allocation order. The target AGV device acquires the instruction data to be sent and determines whether the instruction is executed normally, thus obtaining the instruction judgment result; If the instruction judgment result is normal, then the queue data in the priority queue is updated according to the priority attribute to obtain the updated scheduling instruction sequence; The updated scheduling instruction sequence is encoded for communication to obtain an execution confirmation signal, and the above steps are repeated to update the priority queue in real time. The execution confirmation signal is transmitted to the target AGV device, and the cargo handling is completed according to the execution confirmation signal.

[0013] Secondly, the present invention provides an intelligent control system for AGV vehicles based on end-to-cloud collaboration, comprising: The data acquisition module is used to acquire real-time status data, including the current position of the AGV, warehouse area data, and obstacle positions. The path planning module is used to plan a path based on the current position of the AGV and the position of the obstacle, and obtain updated real-time status data containing the new path task. The map partitioning module is used to divide the warehouse area data into grids based on the updated real-time status data to obtain a dynamic spatial map. The path adjustment module is used to calculate alternative paths and obtain the adjusted AGV running path sequence if the channel width in the dynamic spatial map is less than a preset channel width threshold. The conflict detection module is used to obtain the expected occupied areas of other AGVs from the adjusted AGV running path sequence and perform path overlap comparison to obtain conflict detection results; The sequence optimization module is used to adjust the scheduling order of multiple AGVs based on the conflict detection results using a genetic algorithm to obtain a priority queue; The instruction acquisition module is used to acquire the scheduling instruction for the next AGV from the priority queue and transmit it to the corresponding AGV device to obtain an execution confirmation signal; The signal transmission module is used to transmit the execution confirmation signal to the target AGV device and complete the cargo handling according to the execution confirmation signal.

[0014] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the intelligent control method for AGV vehicles based on end-to-cloud collaboration as described above.

[0015] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the intelligent control method for AGV vehicles based on end-to-cloud collaboration as described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs a dynamic spatial map by acquiring AGV location, environment and obstacle data, and uses a grid method for environmental modeling. It combines the A* algorithm for local path replanning, and uses multi-vehicle path overlap analysis to achieve conflict detection. It uses a genetic algorithm to dynamically optimize the multi-AGV scheduling queue and form collaborative control instructions. This solves the problem of congestion caused by overlapping path planning when multiple AGVs are running collaboratively in the prior art, which leads to a decrease in operating efficiency.

[0017] (2) This invention constructs a dynamic spatial map through real-time data fusion, providing an environmental perception basis for path planning. Based on the gridded modeling method, the system can quantitatively evaluate the passage capacity. When spatial constraints are not met, the A* algorithm is triggered to optimize the local path. To further solve the multi-AGV coordination problem, a path spatiotemporal overlap detection mechanism is introduced, and a genetic algorithm is used to globally optimize the multi-vehicle scheduling sequence, ultimately forming an efficient control scheme that combines distributed decision-making and centralized scheduling.

[0018] (3) This invention improves environmental adaptability by constructing a dynamic map, optimizes path planning and scheduling efficiency by adopting a hybrid strategy of A* and genetic algorithm, and reduces the probability of traffic congestion by conflict detection and priority scheduling mechanism, thereby improving the operating efficiency and space utilization of multi-AGV system in complex scenarios. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the intelligent control method for AGV vehicles based on end-to-cloud collaboration provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the intelligent control system structure of AGV vehicles based on end-to-cloud collaboration provided in the second embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 The first embodiment of the present invention provides an intelligent control method for AGVs based on end-to-cloud collaboration, including the following steps: S11, acquire real-time status data including the current position of the AGV, warehouse area data, and obstacle positions, perform path planning based on the current position of the AGV and the obstacle positions, and obtain updated real-time status data including the new path task; S12, perform grid division based on the warehouse area data in the updated real-time status data to obtain a dynamic spatial map; S13, If the channel width in the dynamic space map is less than the preset channel width threshold, then calculate the alternative paths to obtain the adjusted AGV running path sequence. S14, Obtain the expected occupied areas of other AGVs from the adjusted AGV running path sequence and perform path overlap comparison to obtain the conflict detection result; S15, Based on the conflict detection results, a genetic algorithm is used to adjust the scheduling order of multiple AGVs to obtain a priority queue; S16, Obtain the scheduling instruction for the next AGV from the priority queue and transmit it to the corresponding AGV device to obtain an execution confirmation signal; S17, the execution confirmation signal is transmitted to the target AGV device and the cargo handling is completed according to the execution confirmation signal.

[0022] In step S11, the step of performing path planning based on the current position of the AGV and the position of the obstacle to obtain updated real-time status data containing the new path task includes: The location distribution of goods is obtained. If the current position of the AGV deviates from the preset path, the AGV movement path is readjusted to obtain a new path task. Based on the new path task and the location of the obstacle, the A* algorithm is used to calculate the obstacle avoidance path of the AGV to obtain the optimized movement trajectory. Based on the optimized movement trajectory and the cargo location distribution, the K-means algorithm is used to cluster the cargo storage area to obtain the optimal storage location. Adjust the AGV scheduling task according to the optimal storage location to obtain updated real-time status data.

[0023] First, a sensor network is used to collect real-time data on the AGV's current position, cargo volume, and obstacle positions. Kalman filtering is then used to fuse this data into warehouse space status data. The sensor network consists of LiDAR, ultrasonic sensors, and cameras, deployed at key nodes in the warehouse. The AGV's current position is obtained through LiDAR scanning with centimeter-level accuracy; cargo volume is measured using a 3D vision sensor; and obstacle positions are detected collaboratively by ultrasonic sensors and cameras. Based on the AGV's motion state, its next state is predicted, and the error covariance is estimated. The AGV position coordinates acquired by LiDAR, the cargo volume measurement from the 3D vision sensor, and the obstacle position data acquired by ultrasonic sensors and cameras are input into a Kalman filter according to sensor type, constructing a centralized extended Kalman filter model with the AGV pose [X, Y, θ]^T as the state vector. The coordinate data provided by LiDAR, the cargo volume data provided by the vision sensor, and the obstacle coordinate data provided by ultrasonic sensors / cameras are transformed using a coordinate system and then input as observation vectors into the centralized extended Kalman filter model. The noise covariance matrix of each sensor needs to be pre-determined through offline calibration experiments, and its weights are dynamically adjusted according to the signal-to-noise ratio (SNR) during the filtering process. Specifically, the initial noise covariance matrix is ​​obtained by calculating the variance matrix of the error values ​​from 100 repeated measurements of the sensors in a static environment. The SNR is defined as the reciprocal of the variance of the measured values, and the weight adjustment formula is wi = 1 / σi^2. 2 , where σi^ 2 It is the noise variance of sensor i.

[0024] AGV path planning consists of two levels: global planning and local replanning. First, global planning is performed. When the AGV receives a new task or deviates significantly from the path (>2 meters), Dijkstra's algorithm is used to calculate a globally optimal path. Then, local replanning is performed. As the AGV travels along the globally optimal path, onboard sensors continuously detect the environment. If an obstacle is detected occupying a grid on the globally optimal path (e.g., an obstacle is detected at (10,23), local replanning is immediately triggered. The grid point on the global path closest to the obstacle and without obstacles is selected as the regression point. Using the A* algorithm, starting from the current position (10,22), the algorithm bypasses the obstacle and searches for a local path to a later, obstacle-free point (10,25) on the globally optimal path. The heuristic function h(n) for the A* algorithm uses Euclidean distance. Finally, path execution is performed. The AGV prioritizes executing the local path. After successfully navigating the obstacle area, it automatically switches back to the subsequent path of the globally optimal path, thus achieving the global objective.

[0025] It should be noted that after the A* algorithm generates the path, kinematic trajectory optimization is performed to ensure that the path conforms to the AGV's minimum turning radius (e.g., 1.5 meters in this embodiment). It's worth noting that the A* algorithm, as a path planning algorithm, improves pathfinding efficiency. Combining actual and estimated movement costs, it uses Euclidean distance as a heuristic function to dynamically prioritize and expand the optimal path nodes most likely to lead to the target. In AGV applications, the A* algorithm can quickly generate smooth, kinematically constrained shortest paths, while naturally supporting dynamic obstacle avoidance. When the environment changes (such as the addition of obstacles), the algorithm can locally replan without global recalculation, reducing computational resource consumption. Its output path can be directly used for AGV motion control, reducing subsequent trajectory optimization steps and improving real-time performance and operational efficiency, making it particularly suitable for collaborative scheduling scenarios of large-scale AGV clusters.

[0026] Then, based on the optimized movement trajectory and cargo volume data, the warehouse space occupancy is determined, and the K-means algorithm is used to cluster the cargo storage areas according to the cargo location distribution to obtain the optimal storage location. For example, data on 100 items in the warehouse is preprocessed, with each data point containing the X and Y coordinates of the item's location and its volume attribute. The number of clusters to be clustered is set to K=3, i.e., 3 main storage areas, and three cluster centers are randomly initialized. Iterative calculations are then performed. The volume weight is set to 0.6 to prioritize the grouping of items with similar volumes, thereby reducing the frequency of AGV shelf adjustments; the position coordinate weights X and Y are each 0.2 to maintain the locality of clustering in space. In the first step, the Euclidean distance from each item to each center point is calculated, and items with higher similarity in volume are grouped together and given higher weights, while the position coordinates are assigned lower weights. The algorithm employs Min-Max normalization to eliminate the influence of dimensions, linearly scaling each feature value to the [0,1] interval and assigning it to the cluster containing the nearest center point. In the second step, the mean center point of each cluster is recalculated, updating its position coordinates and average volume features. This iterative process continues until the cluster assignments no longer change or the maximum number of iterations (100) is reached. The algorithm then converges, clustering the data into three main storage regions and determining the optimal storage location, such as region A near the AGV path. It's worth noting that setting the maximum number of iterations to 100 provides a sufficient safety margin, ensuring that even in adverse conditions such as complex data distribution and poor initial points, the algorithm can be forcibly terminated and output a result, thus not affecting the AGV's operating rhythm.

[0027] It should be noted that the K-means algorithm, as a widely used unsupervised machine learning method, has the core function of automatically discovering inherent clustering patterns in complex data, thereby achieving efficient grouping and structured processing of large-scale data. This capability makes it play an important role in many fields. Applying it to the warehousing and logistics scenario described in this paper, in addition to directly reducing the AGV handling distance, it brings many other beneficial effects: the algorithm replaces traditional manual experience-based division with data-driven decision-making, significantly improving the objectivity and scientific nature of storage area planning; it can adaptively re-cluster based on dynamic inbound and outbound data of goods, providing support for continuous optimization of warehouse layout; at the same time, this clustering method simplifies the complexity of inventory management, allowing similar goods to be stored together, which not only reduces the difficulty of AGV path optimization but also improves the efficiency of manual inventory counting and replenishment operations, thus achieving a synergistic improvement in warehouse space utilization and operational efficiency overall.

[0028] Finally, for dynamic obstacles such as temporarily stacked goods, task scheduling is adjusted in real time based on the optimal storage location to obtain updated real-time status data. When the sensor network detects new obstacles or the AGV detects obstacles in its path via ultrasound or cameras, the location of the dynamic obstacle is uploaded to adjust the new path. For example, if the AGV was originally scheduled to move goods to area A, but is blocked by obstacles, the task is reassigned to area B based on the clustering results of the K-means algorithm, optimizing task allocation. The division rules for areas A and B are based on the K-means clustering results, inferring the optimal K value from the dataset using the elbow rule or silhouette coefficient. Each area is defined by the physical coordinates of its cluster center point, such as the coordinates within the warehouse. By dividing the physical coordinates by the preset grid resolution and rounding, the center point coordinates are converted into a unique grid index number, corresponding to the gridded warehouse space, thus clarifying its spatial range. The data acquisition frequency of the sensor network is dynamically adjusted based on real-time task allocation data. For example, when there are 5 AGVs that need to pass through the same section within 3 minutes, resulting in a dense workload, the sensor frequency is increased from 1 second / time to 0.5 seconds / time to ensure real-time data.

[0029] In step S12, the step of dividing the warehouse area data into grids based on the updated real-time status data to obtain a dynamic spatial map includes: Obtain warehouse area data from the updated real-time status data, and divide the warehouse area data into grids to obtain gridded areas; The dynamic spatial map is obtained by calculating the channel width and corner occupancy based on the gridded area.

[0030] First, the warehouse area data is acquired from the updated real-time status data. This data is collected by a sensor network including infrared sensors, depth cameras, and pressure sensors, primarily deployed at warehouse shelves and aisle nodes. Infrared sensors detect the presence of objects in grid cells, depth cameras measure object height to determine occupancy, and pressure sensors sense ground load to confirm the location of heavy goods. For example, the warehouse is divided into a 100×100 grid, with each grid cell being 1 square meter. The infrared sensor scan detects goods in grid (20,30), the depth camera measures the goods height as 1.5 meters, and the pressure sensor confirms a load of 500 kilograms; based on these combined findings, the grid is determined to be occupied.

[0031] Subsequently, the warehouse area data is divided using a gridding method, and the occupancy of each grid cell is analyzed through spatial status data. The physical warehouse coordinates are linearly mapped to the grid index using a predefined resolution: the grid row and column number corresponding to each physical point is calculated, thus converting the continuous space into a two-dimensional grid array. The gridded area represents dividing the warehouse into multiple 1-square-meter cells, generating a map containing occupancy and vacancy information. For example, grid (25,35) with no objects and no ground load is marked as an vacant cell.

[0032] Finally, the channel width is calculated based on the continuity of adjacent free grids. The occupancy of corner areas is determined by analyzing the spatial distribution of gridded turning areas, and a dynamic spatial map is obtained by marking the map according to availability indicators. For example, if the channel requires a width of at least 3 meters, i.e., a grid of 1×1 square meters, it is calculated that three consecutive free grids can meet the width requirement. The scan found that grids (30,40) to (30,41) are free, but (30,42) is occupied, and the channel width is only 2 meters, which is lower than the preset threshold of 3 meters for safe passage, so it is marked as unusable. In addition, the turning area is a 4×4 grid group near grid (50,50). If multiple consecutive grids in this area are occupied, and the proportion is higher than the preset threshold of 50%, and the width of the consecutive grids is less than the channel width required for normal turning, it is marked as unusable. For example, 12 grids from (50,50) to (53,53) are occupied by goods, restricting the turning space, and are marked as unusable areas.

[0033] It should be noted that the dynamic spatial map is a grid-based, real-time updated environmental representation model. Its core data structure is a two-dimensional array, where each element corresponds to a grid cell and stores the state information of that grid cell, such as status labels for occupied, idle, AGV position, temporary obstacles, channel width, and turning feasibility. This map dynamically updates the grid status by continuously integrating real-time data from LiDAR and visual sensors with task scheduling instructions. For example, when temporary stacking of goods is detected, the corresponding grid is immediately marked as "occupied," thus providing accurate data for path planning, region clustering, and task scheduling.

[0034] In step S13, if the channel width in the dynamic spatial map is less than a preset channel width threshold, alternative paths are calculated to obtain an adjusted AGV running path sequence, including: When the width of the passage in the dynamic spatial map is less than the sum of the cargo volume and the safety buffer, the map data is acquired and the passage width parameter is extracted to obtain the spatial constraint conditions. Based on the spatial constraints, the feasible paths are calculated using the A* algorithm to obtain a set of candidate paths. The candidate paths are then sorted according to path length and travel time to obtain a sorted path sequence. The Dijkstra algorithm is used to dynamically adjust the paths in the sorted path sequence to obtain the optimized path sequence. Based on the optimized path sequence, navigation planning is performed on the AGV operation instructions to obtain the AGV operation task sequence; If a dynamic obstacle is detected when the AGV operation task sequence is executed, the location data of the dynamic obstacle is obtained, and the new spatial constraints are recalculated to obtain the adjusted AGV operation path sequence.

[0035] First, the preset channel width threshold refers to the sum of the cargo volume and the safety buffer zone. When the channel width is less than the sum of the cargo volume and the safety buffer zone, additional spatial constraints need to be extracted. The cargo volume can be calculated based on data obtained from the warehouse's 3D vision sensors. For example, if the warehouse grid is 100×100, each grid is 1 square meter, the cargo volume is 1.2 meters wide, and the safety buffer zones on both sides are 0.3 meters wide, then the channel needs to be at least 1.8 meters wide. Scanning revealed that grids (20,30) to (20,31) are continuously empty, with a channel width of 2 meters, satisfying the spatial constraints; however, grid (30,40) is only 1 meter wide, not meeting the requirements. These width parameters are extracted to form spatial constraints, marking unusable channels to ensure the safe passage of the AGV.

[0036] Subsequently, based on spatial constraints, the A* algorithm is used to calculate feasible paths. The AGV moves from grid (10,10) to (50,50). The A* algorithm integrates grid occupancy status and travel cost to obtain a set of candidate paths. For example, path 1 is from (10,10) to (30,30) and then to (50,50), with a length of 80 meters and an estimated travel time of 100 seconds; path 2 is from (10,10) to (40,40) and then to (50,50), with a length of 90 meters and a travel time of 110 seconds. Based on length and time, path 1 has higher priority, resulting in a sorted path sequence. The A* algorithm calculation process here is basically the same as the A* algorithm calculation process in step S11, so it will not be described again.

[0037] Next, for the sorted path sequence, Dijkstra's algorithm calculates the dynamic adjustment cost to obtain the optimized path sequence. In path 1, grid (30,30) is temporarily occupied, requiring a detour to (32,32), adding 5 meters of distance and 10 seconds of time. Dijkstra's algorithm evaluates the detour cost, updates the path sequence, and confirms that path 1 is still optimal. This method ensures that the path remains efficient in a dynamic environment. The process of Dijkstra's algorithm calculating the better path here is basically the same as the process of Dijkstra's algorithm calculating the shortest path and improving efficiency in step S11, so it will not be described again.

[0038] It's important to note that the A* algorithm uses Euclidean distance to calculate the initial path from the starting point to the target point because it can quickly generate a set of candidate paths and sort them based on path length and time. When local path adjustments are needed, Dijkstra's algorithm is more suitable because it doesn't rely on heuristic functions but systematically explores all possible directions to ensure the shortest path is found. This is especially true when the environment changes frequently, as it reliably calculates detour costs. Both the A* and Dijkstra algorithms are classic graph search methods and are inherently compatible because they share similar data structures, such as priority queues and node expansion logic, both used to find the shortest path.

[0039] Then, the navigation planning module assigns AGV running instructions based on the optimized path sequence, resulting in an AGV running task sequence. Once path 1 is selected, the task sequence is generated: the AGV moves from (10,10) along a straight line to (30,30), then turns to (50,50). The instructions include speed control and turning radius to ensure stable operation. The clear task sequence reduces AGV scheduling conflicts.

[0040] Finally, when dynamic obstacles appear, the path is adjusted in real time to obtain the adjusted AGV running path sequence. For example, if grid (30,30) is temporarily occupied by goods, the sensor detects its position and updates the spatial constraints. If only one continuous grid is currently free and the channel width is 1 meter, it does not support AGV passage and is marked. The A* algorithm recalculates the path from (10,10) to (32,32) and then to (50,50) to obtain the adjusted AGV running path sequence. If no feasible path can be replanned and replanning fails, the system will initiate a remedial mechanism: first, it will try to make the AGV wait briefly at its current position for 10 seconds before replanning again; if it still fails, it will send a signal to confirm that the task is suspended or require manual removal of the obstacle. The determination of the continuous existence of dynamic obstacles is based on time thresholds and continuous detection: if the grid is reported as "occupied" within 10 seconds of three consecutive sensor scan cycles, it is determined to be a continuous obstacle. The system will treat it as a static obstacle, update the global map constraints, and notify all AGVs to synchronize this information.

[0041] In step S14, obtaining the expected occupied areas of other AGVs from the adjusted AGV running path sequence and performing path overlap comparison to obtain conflict detection results includes: Obtain the adjusted AGV running path sequence of multiple AGVs, compare whether the boundary coordinates of the expected occupied area and the emergency passage reserved area intersect, and obtain the area overlap judgment result; When the region overlap judgment result is true, the path planning data of the relevant AGV is extracted, the specific location of the conflict area is determined, and the K-means clustering algorithm is used to classify the conflict area to obtain the core location point of the conflict; when the region overlap judgment result is false, the conflict detection result is obtained directly, and the task is executed to the next step. Based on the core location of the conflict, the expected overlap of the occupied areas of each AGV is recalculated to obtain the conflict detection results.

[0042] First, the adjusted AGV running path sequences are extracted, generating the position coordinates corresponding to each time point. The boundary coordinates of the expected occupied area and the reserved area for emergency passages are compared to determine if there is any overlap, thus obtaining the area overlap judgment result. For example, the warehouse grid is 100×100, with a grid unit of 1 square meter. AGV1 moves from grid (10,10) to (50,50) at a speed of 1 m / s, expected to arrive in 57 seconds; AGV2 moves from grid (20,20) to (60,60) at the same speed. AGV1 is recorded as being at grid (30,30) at the 28th second, and AGV2 is recorded as being at grid (35,35) at the 21st second, forming a time-location dataset.

[0043] When generating the boundary coordinates of the expected occupied area, the occupied range is calculated based on the AGV's size and safety buffer zone. For example, if an AGV is 1 meter wide with a 0.3-meter safety buffer zone on each side, and the geometric center of AGV1 is located in a 1×1 square meter grid (30,30), occupying a total width of 1.6 meters, the boundary coordinates of the occupied area are (29.7,29.7) to (30.3,30.3). This means that the area centered at (30,30) occupies 0.8 meters of width on both sides, while the grid (30,30) itself is 0.5 meters long, thus occupying 0.3 meters on each side of the grid. Similarly, the occupied area of ​​AGV2 is calculated to form a set of boundary coordinates. These coordinates are compared with the reserved area for emergency evacuation, such as [grid (40,40) to (40,50), 1 meter wide], to determine if there is any overlap, thus obtaining the area overlap judgment result. For example, AGV1's occupied area at 20 seconds does not intersect with the emergency passage, but AGV2's occupied area at 15 seconds overlaps with grid (40,40), which is determined to be an area overlap. If there is no area overlap, a conflict detection result of false is output, and execution continues.

[0044] The designated emergency evacuation area is first converted from physical coordinates to grid coordinates. For example, with a resolution of 1 meter per grid, the area from (40.0m, 40.0m) to (40.0m, 50.0m) corresponds to grid coordinates [40,40] to [40,50]. Then, while ensuring safety, it's necessary to minimize the impact on normal operations. Therefore, the width and location of the emergency passage are calculated based on the AGV size (including the safety buffer) and typical cargo volume to ensure the passage meets emergency needs without excessively occupying space. For example, based on the sum of the AGV width of 1 meter and the safety buffer of 0.6 meters, the minimum required passage width is calculated to be 1.6 meters. Therefore, a passage width of 2 meters (i.e., 2 grids) is set to provide a margin.

[0045] Subsequently, after detecting overlapping areas, the time of the conflict and the path planning data of the relevant AGVs are extracted. For example, if the conflict occurs at 15 seconds, involving AGV2, the conflict area is located at grid (40,40). K-means clustering is used to classify the conflict area and generate cluster centers. Due to the concentrated nature of the conflict, the K value is set to 1 to generate a single cluster center. For example, if the conflict area includes grid (40,40) and its neighboring grid (40,41), the cluster center is determined to be (40.5,40.5), obtaining the core location of the conflict. Data points are obtained by mapping discrete grid coordinates to their physical center continuous values ​​as (X+0.5, Y+0.5), and then the arithmetic mean of these points is calculated to determine the cluster center, such as (40.5,40.5). The K-means convergence condition is set to a threshold of 0.01 set to avoid occupying other lanes; the initial center point is usually randomly selected from the data points. This location point is used to extract the path of the affected AGV, and the path of AGV2 needs to be adjusted.

[0046] Next, when adjusting the path sequence, the AGV2 path is replanned based on the core conflict location point. The original path went from (20,20) to (60,60) and then through (40,40). Now it detours to (42,42), generating a new path sequence: (20,20) to (42,42) and then to (60,60). The adjusted path is simulated, and the expected occupied area of ​​the AGV2 is recalculated. For example, at the 15th second, it is located in grid (42,42), with boundary coordinates of (41.85,41.85) to (42.15,42.15). If there is no intersection with the emergency passage area, the conflict detection result is false.

[0047] In step S15, adjusting the scheduling order of multiple AGVs using a genetic algorithm based on the conflict detection results to obtain a priority queue includes: Obtain traffic area density data, and analyze the AGV movement demand in traffic areas exceeding the preset number of traffic areas based on the traffic area density data and the conflict detection results to obtain a demand priority list; Based on the aforementioned priority list of requirements, a genetic algorithm is used to adjust the scheduling order of AGVs, resulting in an optimized set of scheduling schemes. The running efficiency of the optimized scheduling scheme set is calculated to obtain an efficient scheduling scheme; When the efficient scheduling scheme meets the preset efficiency threshold, the output is a priority queue, and the final AGV scheduling order is obtained. The final AGV scheduling order is optimized according to the actual traffic conditions to obtain the priority queue.

[0048] First, obtaining traffic zone density data is the foundation for optimized scheduling. When the conflict detection result is false, the AGV movement demand in high-density areas is analyzed based on the traffic zone density data. Traffic zone density data is generated by statistically analyzing the AGV occupancy frequency of a grid per unit time. For example, if more than 5 AGVs pass through grid (35,35) within 3 minutes, exceeding the preset number of passages based on historical analysis for congestion, the density is high and it is marked as a high-density area. Grid (35,35) is identified as having frequent task assignments with overlapping AGV passage times, requiring priority scheduling. The demand priority list is generated based on task urgency and area density. For example, AGV1's task is to transport high-priority goods, so it is assigned priority 1; AGV2's task is routine handling, so it is assigned priority 2. When there are AGVs with the same priority but different passage times, the one with the shorter passage time is prioritized. When priority and passage time are still the same, passage is prioritized based on arrival time, resulting in the final demand priority list.

[0049] Subsequently, a genetic algorithm is used to adjust the scheduling order of AGVs according to the aforementioned priority list, resulting in an optimized set of scheduling schemes. When initializing the genetic algorithm population, chromosome encoding is used to represent the scheduling order; for example, encoding [1,2] indicates that AGV1 has priority over AGV2. The population is initialized to generate a random scheduling sequence (e.g., chromosome [1,2] indicates that AGV1 has priority over AGV2, and chromosome [2,1] indicates that AGV2 has priority over AGV1), and a fitness function is set. ,in This is a normalized time efficiency metric used to reflect the total task completion time. To quantify and assess risks such as path collisions and deadlocks using the normalized conflict index, with a particular emphasis on safety performance, the weight of the conflict index is set to w2=0.6, and the weight of time efficiency is set to w1=0.4. By comprehensively evaluating task time and conflict indicators, this weight configuration reflects a "safety first" scheduling strategy. By increasing the weight of conflict indicators, the algorithm is forced to prioritize avoiding collision risks, while retaining the time weight to avoid generating an overly conservative scheduling scheme, thus achieving a balance between safety and efficiency.

[0050] The normalization formula is as follows: A min Indicates the shortest passage time or the fewest number of conflicts, A max A represents the longest passage time or the maximum number of conflicts, and A represents the current passage time or the current number of conflicts. n This represents the normalized value.

[0051] High-fit individuals are selected using a roulette wheel (e.g., prioritizing [1,2]). Partial matching crossover (PMX) is used for gene exchange, with a crossover rate of 80% and a mutation rate of 2%. For example, crossing parent [1,2] with [2,1] may produce offspring [1,2] or [2,1]. Mutation operations are then performed, such as randomly swapping gene positions to change [1,2] to [2,1]. This process is iteratively optimized until fitness stabilizes for 20 consecutive generations with a time variation not exceeding 1 second, outputting the optimal scheduling scheme. The running efficiency of the scheme is calculated. If an efficient scheme meets a preset time threshold (e.g., total time less than 12 seconds), it is converted into a priority queue. The order is then fine-tuned based on real-time traffic conditions to generate the queue, resulting in an optimized set of scheduling schemes. The preset time threshold is obtained by analyzing historical running data and statistically analyzing the average completion time of similar tasks under normal operating conditions.

[0052] Next, after obtaining the initial scheduling scheme set, check for path conflicts. If AGV1 and AGV2 pass through grid (35,35) simultaneously, exchange priorities through the crossover operation of the genetic algorithm to generate a new code [2,1], or adjust the waiting time of AGV2 through the mutation operation to avoid conflict. The crossover and mutation operations of the genetic algorithm here are similar to those in the previous genetic algorithm, so they will not be described again.

[0053] Then, the operational efficiency of the optimized scheduling scheme set is evaluated. The transit time of each scheme in the high-density area is calculated. For example, scheme [1,2] makes the transit time of AGV1 and AGV2 in grid (35,35) 5 seconds and 7 seconds, respectively, with higher overall efficiency. If the sum of the transit times of two AGVs is the same, the shortest transit time is used to determine the efficiency. For example, in scheme [2,1], the transit time of AGV1 and AGV2 in grid (35,35) is 6 seconds, so scheme [1,2] is more efficient.

[0054] Finally, the most efficient solution is selected. If the transit time is less than the preset transit time threshold of 8 seconds, the shorter transit time has a more significant impact on overall efficiency when priorities are equal. The solution is output as a priority queue, prioritizing AGV1 and then AGV2. During task allocation, AGV1 executes the path from grid (15,15) to (55,55), while AGV2 waits briefly before executing the new path from (25,25) to (65,65). The generated task execution sequence includes specific instructions, such as AGV1 proceeding straight to (55,55) and AGV2 starting at the 2nd second. For example, the AGV operating status can be monitored in real time to dynamically adjust the demand priority queue. By detecting the actual position of AGV1 in the grid (25,25) using sensor data, if delays due to obstacles require turning and increase travel time, the startup time of AGV2 is dynamically adjusted accordingly to increase its travel time. If the sum of the travel time of AGV1 and the remaining travel time of AGV2 is significantly greater than the time required for the replanned alternative route, AGV2 will navigate to a new route to ensure smooth passage in high-density areas. This dynamic adjustment strategy effectively improves scheduling flexibility and ensures task execution efficiency.

[0055] In step S16, obtaining the scheduling instruction for the next AGV from the priority queue and transmitting it to the corresponding AGV device to obtain an execution confirmation signal includes: The scheduling instructions are extracted from the priority queue and the instruction data to be sent is obtained according to the priority allocation order. The target AGV device acquires the instruction data to be sent and determines whether the instruction is executed normally, thus obtaining the instruction judgment result; If the instruction judgment result is normal, then the queue data in the priority queue is updated according to the priority attribute to obtain the updated scheduling instruction sequence; The updated scheduling instruction sequence is encoded for communication to obtain an execution confirmation signal, and the above steps are repeated to update the priority queue in real time. The execution confirmation signal is transmitted to the target AGV device, and the cargo handling is completed according to the execution confirmation signal.

[0056] First, scheduling instructions are extracted from the priority queue and allocated according to priority. The priority queue is constructed based on task priority attributes; for example, tasks transporting urgent goods have higher priority than regular handling. For example, in a 100×100 grid, AGV1's task is to transport medical supplies, with priority 1; AGV2's task is to handle ordinary goods, with priority 2. Instructions are extracted from the queue, determining that AGV1 will prioritize the path from grid (10,10) to (50,50), generating instruction data "AGV1 proceeds straight to (50,50)", ultimately yielding the instruction data to be sent. The instruction data is then encoded for communication, using the ZigBee wireless protocol to generate data packets. The data packets contain AGV1's device identifier ID001 and target location information. A communication link matching ID001 is selected, and the data packets are wirelessly transmitted to AGV1, receiving a successful transmission status signal.

[0057] Subsequently, AGV1 receives the data packet, executes the instruction, and returns an acknowledgment signal. The instruction has four states: idle, in progress, completed, and fault. Analyzing the acknowledgment signal content, if it contains the "task execution" status, the instruction is considered normal, and the task status is updated to "in progress" until a "complete" instruction result is returned, and the AGV1's current task is removed from the priority queue. When "idle," there is no task, and the status remains unchanged. If the returned signal contains the "fault" status, the AGV is marked as faulty, and the next AGV task is assigned. If no signal is returned or a "fault" error is reported three times, a manual intervention alarm is triggered.

[0058] Next, if the instruction judgment result is normal, AGV2's task enters the top of the queue, is reordered, and the instruction "AGV2 from grid (20,20) to (60,60)" is generated. Through the queue management mechanism, the priority of AGV2's task is checked, and grid (30,30) is found to be a high-density area. The AGV2 instruction is adjusted to "start after 2 seconds" to avoid conflicts. The new instruction sequence is again encoded and sent via the ZigBee protocol, and the execution confirmation signal from AGV2 is obtained cyclically. When the instruction judgment result is abnormal, for example, if AGV1 encounters an obstacle during execution, the signal returns to a "pause" or "backtrack" state. The priority queue is dynamically adjusted, and AGV2's task is moved forward according to task priority, generating a new instruction "AGV2 start immediately." Subsequently, the path of AGV1 is readjusted, and after the path adjustment is completed, AGV1 returns to the highest priority execution state. This dynamic adjustment improves the flexibility of task allocation.

[0059] Then, the execution confirmation signals are transmitted sequentially according to priority, the completed AGV tasks are deleted, the next AGV task is brought forward, the priority queue is updated in real time, and AGVs with shorter paths are prioritized for allocation based on the task requirements of high-density areas to reduce waiting time. For example, AGV3's task is from grid (15,15) to (20,20), which is a shorter path, so its passage instruction is generated first to improve overall scheduling efficiency.

[0060] Finally, the execution confirmation signal is transmitted to the target AGV device, and the cargo handling is completed according to the execution confirmation signal.

[0061] In summary, this invention discloses an intelligent control method for AGVs based on end-to-cloud collaboration. By issuing scheduling instructions and updating the map status in real time according to the execution confirmation signal, the method ultimately achieves the technical effects of AGV obstacle avoidance and stable operation, thereby improving the level of intelligence and operational efficiency of warehousing and logistics, and providing an efficient solution for AGV scheduling in complex dynamic environments.

[0062] Reference Figure 2 The second embodiment of the present invention provides an intelligent control system for AGV vehicles based on edge-cloud collaboration, comprising: The data acquisition module is used to acquire real-time status data, including the current position of the AGV, warehouse area data, and obstacle positions. The path planning module is used to plan a path based on the current position of the AGV and the position of the obstacle, and obtain updated real-time status data containing the new path task. The map partitioning module is used to divide the warehouse area data into grids based on the updated real-time status data to obtain a dynamic spatial map. The path adjustment module is used to calculate alternative paths and obtain the adjusted AGV running path sequence if the channel width in the dynamic spatial map is less than a preset channel width threshold. The conflict detection module is used to obtain the expected occupied areas of other AGVs from the adjusted AGV running path sequence and perform path overlap comparison to obtain conflict detection results; The sequence optimization module is used to adjust the scheduling order of multiple AGVs based on the conflict detection results using a genetic algorithm to obtain a priority queue; The instruction acquisition module is used to acquire the scheduling instruction for the next AGV from the priority queue and transmit it to the corresponding AGV device to obtain an execution confirmation signal; The signal transmission module is used to transmit the execution confirmation signal to the target AGV device and complete the cargo handling according to the execution confirmation signal.

[0063] It should be noted that the intelligent control system for AGVs based on end-to-cloud collaboration provided in this embodiment of the invention is used to execute all the process steps of the intelligent control method for AGVs based on end-to-cloud collaboration in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0064] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a path planning program. When the processor executes the computer program, it implements the steps described in the various embodiments of the edge-cloud collaborative AGV intelligent control method, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0065] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0066] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0067] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0068] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0069] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0070] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0071] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An AGV trolley intelligent control method based on end-cloud cooperation, characterized in that, The application relates to a method for scheduling multiple AGVs, and the method comprises the following steps: acquiring real-time state data including the current position of an AGV, warehouse area data and obstacle position data, performing path planning according to the current position of the AGV and the obstacle position data, and obtaining updated real-time state data containing a new path task; performing grid division according to the warehouse area data in the updated real-time state data, and obtaining a dynamic space map; if the channel width in the dynamic space map is less than a preset channel width threshold, calculating an alternative path, and obtaining an adjusted AGV running path sequence; obtaining an expected occupation area of other AGVs from the adjusted AGV running path sequence and performing path overlap comparison, and obtaining a conflict detection result; adjusting the scheduling sequence of multiple AGVs according to the conflict detection result by using a genetic algorithm, and obtaining a priority queue; obtaining a scheduling instruction of a next AGV from the priority queue and transmitting the scheduling instruction to a corresponding AGV device, and obtaining an execution confirmation signal; transmitting the execution confirmation signal to a target AGV device and completing goods carrying according to the execution confirmation signal. 2.The AGV intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, The method for scheduling multiple AGVs comprises the following steps: obtaining a goods position distribution, and readjusting the moving path of the AGV when the current position of the AGV deviates from a preset path, and obtaining a new path task; calculating an AGV obstacle avoidance path by using an A* algorithm according to the new path task and the obstacle position, and obtaining an optimized moving track; performing cluster analysis on a goods storage area by using a K-means algorithm according to the optimized moving track and the goods position distribution, and obtaining an optimal storage position; adjusting the AGV scheduling task according to the optimal storage position, and obtaining updated real-time state data. 3.The AGV intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, The method for scheduling multiple AGVs comprises the following steps: obtaining the warehouse area data in the updated real-time state data, performing grid division on the warehouse area data to obtain a grid area, calculating the channel width and corner occupation situation according to the grid area, and obtaining a dynamic space map. The method for scheduling multiple AGVs comprises the following steps:

4. The AGV intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, when the channel width in the dynamic space map is less than the sum of the volume of goods and a safety buffer, obtaining map data and extracting a channel width parameter, and obtaining a space constraint condition; calculating a feasible path by using an A* algorithm according to the space constraint condition, obtaining an alternative path set, performing priority sorting on the alternative path set according to path length and passing time, and obtaining a sorted path sequence; dynamically adjusting the path of the sorted path sequence by using a Dijkstra algorithm, and obtaining an optimized path sequence; performing navigation planning on the AGV running instruction according to the optimized path sequence, and obtaining an AGV running task sequence; ​ The AGV operation task sequence is executed, if a dynamic obstacle is detected, dynamic obstacle position data is acquired, new space constraint conditions are recalculated, and an adjusted AGV operation path sequence is obtained.

5. The AGV trolley intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, The expected occupation area of other AGVs is acquired from the adjusted AGV operation path sequence, and path overlap comparison is performed to obtain a conflict detection result, including: The adjusted AGV operation path sequences of multiple AGVs are acquired, and whether there is an intersection between the boundary coordinates of the expected occupation area and the emergency passage reserved area is compared to obtain a region overlap judgment result; When the region overlap judgment result is true, the path planning data of the related AGV is extracted, the specific position of the conflict area is determined, and the K-means clustering algorithm is used to classify the conflict area to obtain a conflict core position point; when the region overlap judgment result is false, the conflict detection result is directly obtained, and the task is executed downward; According to the conflict core position point, the expected occupation area overlap of each AGV is recalculated to obtain the conflict detection result. 6.The AGV intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, The scheduling order of multiple AGVs is adjusted according to the conflict detection result using a genetic algorithm to obtain a priority queue, including: The passing region density data is acquired, and the AGV movement demand of the passing region exceeding the preset passing number is analyzed according to the passing region density data and the conflict detection result to obtain a demand priority list; The scheduling order of the AGVs is adjusted according to the demand priority list using a genetic algorithm to obtain an optimized scheduling scheme set; The running efficiency of the optimized scheduling scheme set is calculated to obtain an efficient scheduling scheme; When the efficient scheduling scheme meets the preset efficiency threshold, the output is a priority queue, and the final AGV scheduling order is obtained. The final AGV scheduling order is optimized according to the actual passing condition to obtain a priority queue.

7. The AGV trolley intelligent control method based on end-cloud cooperation according to claim 1, characterized in that, The scheduling instruction of the next AGV is acquired from the priority queue and transmitted to the corresponding AGV device to obtain an execution confirmation signal, including: The scheduling instruction is extracted from the priority queue and the priority assignment order is obtained to obtain the to-be-sent instruction data; The target AGV device acquires the to-be-sent instruction data and judges whether the instruction execution is normal to obtain an instruction judgment result; If the instruction judgment result is normal, the queue data in the priority queue is updated according to the task priority attribute to obtain an updated scheduling instruction sequence; The updated scheduling instruction sequence is communication encoded to obtain an execution confirmation signal, and the above steps are repeatedly executed to update the priority queue in real time; The execution confirmation signal is transmitted to the target AGV device, and the goods carrying is completed according to the execution confirmation signal.

8. An AGV trolley intelligent control system based on end-cloud cooperation, characterized in that, Including: The data acquisition module is used to acquire real-time state data including AGV current position, warehouse area data, and obstacle position; The path planning module is used to perform path planning according to the AGV current position and the obstacle position to obtain updated real-time state data containing a new path task; The map division module is used to perform grid division according to the warehouse area data in the updated real-time state data to obtain a dynamic space map; The path adjustment module is configured to calculate an alternative path if the channel width in the dynamic space map is less than a preset channel width threshold, and obtain an adjusted AGV running path sequence. The conflict detection module is configured to obtain an other-AGV-expected-occupancy region from the adjusted AGV running path sequence, perform path overlap comparison, and obtain a conflict detection result. The sequence optimization module is configured to adjust a scheduling sequence of multiple AGVs according to the conflict detection result by using a genetic algorithm, and obtain a priority queue. The instruction acquisition module is configured to obtain a scheduling instruction of a next AGV from the priority queue, transmit the scheduling instruction to a corresponding AGV device, and obtain an execution confirmation signal. The signal transmission module is configured to transmit the execution confirmation signal to a target AGV device, and complete cargo carrying according to the execution confirmation signal.

Images