An AGV intelligent warehousing path planning method and system based on inertial navigation

Abstract

This invention discloses an AGV intelligent warehousing path planning method and system based on inertial navigation. The method first constructs a spatiotemporal path distribution matrix for AGV scheduling based on historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse. Next, a Markov chain is used to analyze this matrix in depth to predict the position distribution of each AGV within a preset time period, thereby constructing a position occupancy grid map. This map determines the conflict probability density of each grid in the target warehouse, and network transmission quality is evaluated for each grid. Finally, by combining the conflict probability density and network transmission quality of each grid, a scientifically reasonable AGV path planning scheme is determined. This invention can effectively improve the efficiency and accuracy of AGV path planning in intelligent warehousing, reduce conflict risks, optimize network transmission, and provide strong support for the efficient operation of intelligent warehousing.

Description

Technical Field

[0001] This invention relates to the field of AGV equipment control technology, and in particular to an AGV intelligent warehousing path planning method and system based on inertial navigation. Background Technology

[0002] In recent years, with the rapid development of intelligent warehousing systems, Automated Guided Vehicles (AGVs), as core logistics equipment, have seen their path planning efficiency and collaborative obstacle avoidance capabilities become crucial factors affecting warehousing operation efficiency. Traditional AGV navigation relies heavily on fixed markers such as magnetic tracks and QR codes, resulting in high deployment costs and poor path flexibility. While inertial navigation-based AGV systems achieve autonomous positioning by integrating data from sensors such as gyroscopes and accelerometers, offering advantages in environmental adaptability, the accumulation of inertial navigation errors over long-term operation can easily lead to positioning drift, and dynamic path conflicts among multiple AGVs are difficult to predict, affecting the overall reliability of the system.

[0003] In existing technologies, path planning methods mostly focus on real-time obstacle avoidance or static scheduling based on task priorities, such as using the A* algorithm or dynamic window method for local path optimization. However, these methods lack in-depth analysis of historical operational data and fail to effectively predict the risks of spatiotemporal resource competition in multi-AGV collaborative operations. Furthermore, existing systems typically assume stable communication network quality and do not consider the impact of network latency or signal interference on the transmission of scheduling instructions, leading to the inability to execute emergency obstacle avoidance instructions in a timely manner. These limitations mean that traditional solutions still suffer from high path conflict rates and insufficient scheduling fault tolerance in complex warehousing scenarios.

[0004] To address the aforementioned technical bottlenecks, this invention proposes an AGV intelligent warehousing path planning method and system based on inertial navigation. By constructing a spatiotemporal path distribution matrix to analyze AGV scheduling patterns, combining Markov chains to predict future path conflict probabilities, and quantifying the impact of network transmission quality on instruction execution, dynamic path optimization under multi-objective constraints is ultimately achieved. This method effectively solves the problems of accumulated positioning errors, delayed conflict prediction, and strong network dependence in traditional solutions, significantly improving the robustness of the AGV system and the efficiency of warehousing operations. Summary of the Invention

[0005] To address at least one of the aforementioned technical problems, this invention proposes an AGV intelligent warehousing path planning method and system based on inertial navigation.

[0006] The first aspect of this invention provides an AGV intelligent warehousing path planning method based on inertial navigation, comprising: Based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a spatiotemporal path distribution matrix for AGV scheduling system to schedule AGVs is constructed. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a preset time period is predicted, and a position occupancy grid map is constructed. The conflict probability density of each grid cell in the target warehouse is determined based on the location-occupied grid map. Evaluate the network transmission quality for each grid in the target warehouse; The path planning scheme for the AGV is determined based on the collision probability density of each grid and the network transmission quality.

[0007] In this scheme, the step of constructing a spatiotemporal path distribution matrix for AGV scheduling by the AGV scheduling system based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse is as follows: The target warehouse is divided into an M×N gridded map. Based on the timestamp, the historical inertial navigation data of each AGV for a preset data acquisition period is divided into time windows. The real-time coordinate sequence, motion speed and heading angle data of the AGV in each time window are extracted to obtain spatiotemporal dynamic trajectory data. The task scheduling instruction data within the preset data acquisition time period is parsed to extract the current task type, target shelf coordinates, and task priority parameters for each AGV, and a task attribute vector is constructed. The spatiotemporal dynamic trajectory data and the task attribute vector are spatiotemporally aligned. Based on the AGV identifier, the trajectory data and task data of the same AGV are fused into a three-dimensional tensor structure. The first dimension of the tensor is the time slice index, the second dimension is the path segment number, and the third dimension is the fused multidimensional feature vector. The multidimensional feature vector includes coordinate offset, instantaneous acceleration, task priority weight, and shelf coordinate deviation. Tensor decomposition algorithm is introduced to perform low-rank approximation on the three-dimensional tensor. The product of the core tensor and the factor matrix is ​​calculated by alternating least squares method to extract the scheduling correlation features of each path segment of AGV driving in the spatiotemporal dimension. Based on the scheduling association features, a spatiotemporal path distribution matrix is ​​constructed, where the row vectors of the matrix represent the dynamic weights of each path segment within a time slice, the column vectors represent the probability distribution of the same path segment in different time slices, and the matrix element values ​​are determined by the product of the Euclidean distance of the path segment association features and the task priority weight.

[0008] In this solution, the analysis of the spatiotemporal path distribution matrix based on Markov chains to predict the position distribution of each AGV within a preset time period and construct a position occupancy grid map specifically involves: The real-time task scheduling instruction data of each AGV in the target warehouse is obtained. The Markov chain method is introduced to map each path segment number in the spatiotemporal path distribution matrix to the state node of the Markov chain, and the real-time task scheduling instruction data is attached as the state attribute. Based on the column vector probability distribution of the spatiotemporal path distribution matrix, the basic transition probability of the path segment in different time slices is extracted. Then, according to the target shelf coordinate deviation and task priority parameters of the real-time scheduling task, the adjustment factor is calculated. The basic transition probability and the adjustment factor are weighted and fused to generate the state transition probability matrix. The current path segment of the AGV is taken as the initial state. Multiple candidate path sequences for future time slices are generated by Monte Carlo tree search. The selection weight of each candidate path is determined according to the state transition probability matrix. The AGV scheduling system predicts the scheduling path of the AGV in the future preset time period based on the selection weight. The location distribution of each AGV in the target warehouse is determined based on the scheduling prediction path within a future preset time period, and a grid map of the AGV's location occupancy at each time node within the future preset time period is constructed based on the location distribution.

[0009] In this solution, determining the conflict probability density of each grid cell in the target warehouse based on the location-occupied grid map specifically involves: Extract AGV occupancy data for each time node within a future preset time period from the location-occupied grid map, and count the number of times each grid is simultaneously occupied by multiple AGVs at different time nodes; Based on the time difference between the time node and the current time, a time decay factor is assigned to each time node, and the number of occupations is weighted and summed with the corresponding time decay factor to generate an initial conflict probability value. The intersection area of ​​each AGV's predicted path within a preset time period is detected, and the number of path intersections is extracted as a dynamic adjustment parameter to perform weighted correction on the initial conflict probability value. The corrected collision probability values ​​are spatially smoothed using a kernel density estimation algorithm to eliminate local abnormal peaks and output the collision probability density distribution for each grid cell.

[0010] In this solution, the network transmission quality assessment of each grid in the target warehouse specifically includes: Based on the conflict probability density, predict the number of AGVs traveling in each grid at each time node within a future preset time period, and obtain the signal reception location information of the AGVs by the AGV scheduling system. The wireless network transmission signal obstruction between the signal receiving location and each grid is obtained, and the interaction signal strength loss value between the AGV and the AGV scheduling system in each grid is determined based on the wireless network transmission signal obstruction. The network channel occupancy rate of each grid is determined based on the AGV driving quantity information. The network transmission quality of each grid is evaluated based on the interaction signal strength loss value and the network channel occupancy rate, and the network quality evaluation result of each grid is obtained within a preset time period.

[0011] In this scheme, the determination of the AGV's path planning scheme based on the collision probability density of each grid and the network transmission quality specifically involves: According to the conflict probability density, if the conflict probability density of each grid in the target warehouse is less than the preset conflict probability density, the current path automatically planned by the AGV scheduling system of the target warehouse will be used as the AGV scheduling planning path. If the target warehouse is stored in a grid with a conflict probability density not less than a preset conflict probability density, the grid is marked as a grid to be adjusted, the network quality assessment result of the grid to be adjusted is obtained, and the delay coefficient of the AGV scheduling system sending the path adjustment scheduling command to the AGV in the grid to be adjusted is determined based on the network quality assessment result. If the delay coefficient of the path adjustment scheduling command is greater than the preset delay coefficient, the movement speed of the AGV in the grid to be adjusted will be reduced to a safe threshold, and the AGV forklifts with path conflicts will be switched to manual control to execute manual control path planning. If the delay coefficient of the path adjustment scheduling instruction is not greater than the preset delay coefficient, the Dijkstra algorithm performs secondary path planning on the AGV in the grid to be adjusted and outputs the secondary scheduling planning path. During the path execution process, when the coordinate deviation between the actual navigation trajectory of the AGV and the planned path exceeds the preset tolerance range, the heading angle is corrected in real time based on the inertial navigation data. If the deviation continues to increase after correction and reaches the maximum allowable deviation, the AGV will drive out of the AGV path planning area of ​​the target warehouse and switch to manual control to obtain an emergency avoidance planning path. The path planning scheme for AGVs is constructed by combining the current path automatically planned by the AGV scheduling system, the path planned by manual control, the path planned by secondary scheduling, and the path planned for emergency avoidance.

[0012] A second aspect of the present invention also provides an AGV intelligent warehousing path planning system based on inertial navigation. The system includes a memory and a processor. The memory includes a program for an AGV intelligent warehousing path planning method based on inertial navigation. When the processor executes the program, the program performs the following steps: Based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a spatiotemporal path distribution matrix for AGV scheduling system to schedule AGVs is constructed. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a preset time period is predicted, and a position occupancy grid map is constructed. The conflict probability density of each grid cell in the target warehouse is determined based on the location-occupied grid map. Evaluate the network transmission quality for each grid in the target warehouse; The path planning scheme for the AGV is determined based on the collision probability density of each grid and the network transmission quality.

[0013] This invention discloses an AGV intelligent warehousing path planning method and system based on inertial navigation. The method first constructs a spatiotemporal path distribution matrix for AGV scheduling based on historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse. Next, a Markov chain is used to analyze this matrix in depth to predict the position distribution of each AGV within a preset time period, thereby constructing a position occupancy grid map. This map determines the conflict probability density of each grid in the target warehouse, and network transmission quality is evaluated for each grid. Finally, by combining the conflict probability density and network transmission quality of each grid, a scientifically reasonable AGV path planning scheme is determined. This invention can effectively improve the efficiency and accuracy of AGV path planning in intelligent warehousing, reduce conflict risks, optimize network transmission, and provide strong support for the efficient operation of intelligent warehousing. Attached Figure Description

[0014] Figure 1 A flowchart of an AGV intelligent warehousing path planning method based on inertial navigation according to the present invention is shown; Figure 2 The flowchart illustrating the present invention for determining the collision probability density of each grid cell in a target warehouse is shown. Figure 3 A flowchart illustrating the network transmission quality assessment method of this invention is shown. Figure 4 A block diagram of an AGV intelligent warehousing path planning system based on inertial navigation according to the present invention is shown. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of an AGV intelligent warehousing path planning method based on inertial navigation according to the present invention is shown.

[0018] like Figure 1 As shown, the first aspect of the present invention provides an AGV intelligent warehousing path planning method based on inertial navigation, comprising: S102, Construct a spatiotemporal path distribution matrix for the AGV scheduling system to schedule AGVs based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse; S104, Analyze the spatiotemporal path distribution matrix based on the Markov chain, predict the position distribution of each AGV within a future preset time period, and construct a position occupancy grid map. S106, Determine the conflict probability density of each grid in the target warehouse based on the location-occupied grid map; S108, Evaluate the network transmission quality of each grid in the target warehouse; S110 determines the AGV's path planning scheme based on the collision probability density of each grid and the network transmission quality.

[0019] It should be noted that in the target intelligent warehouse, AGVs will travel according to the driving paths automatically planned by the AGV scheduling system. However, due to changes in scheduling tasks and other reasons, path conflicts may occur in the driving paths automatically planned by the AGV scheduling system. Therefore, by constructing a spatiotemporal path distribution matrix for AGV scheduling by the AGV scheduling system based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a comprehensive reflection of the spatiotemporal scheduling patterns of AGVs in historical operation can be achieved, providing a data foundation for subsequent path planning. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a future preset time period can be predicted, and a position occupancy grid map can be constructed, which can effectively predict the position of each AGV within the future time period. The system identifies potential path conflict risks in advance by analyzing the distribution of data; it determines the conflict probability density of each grid in the target warehouse based on the location-occupied grid map, quantifying the conflict likelihood of each grid and providing a clear priority basis for path optimization; it assesses the network transmission quality of each grid in the target warehouse, evaluating the impact of the network communication environment on the transmission of scheduling instructions, ensuring the timeliness and reliability of scheduling instructions; and it determines the AGV's path planning scheme based on the conflict probability density of each grid and the network transmission quality, comprehensively considering path conflict risks and network communication quality to generate a robust and adaptable dynamic path planning scheme, thereby effectively avoiding path conflict problems caused by changes in scheduling tasks and improving the operating efficiency and safety of AGVs in intelligent warehousing.

[0020] According to an embodiment of the present invention, the step of constructing a spatiotemporal path distribution matrix for AGV scheduling by the AGV scheduling system based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse specifically includes: The target warehouse is divided into an M×N gridded map. Based on the timestamp, the historical inertial navigation data of each AGV for a preset data acquisition period is divided into time windows. The real-time coordinate sequence, motion speed and heading angle data of the AGV in each time window are extracted to obtain spatiotemporal dynamic trajectory data. The task scheduling instruction data within the preset data acquisition time period is parsed to extract the current task type, target shelf coordinates, and task priority parameters for each AGV, and a task attribute vector is constructed. The spatiotemporal dynamic trajectory data and the task attribute vector are spatiotemporally aligned. Based on the AGV identifier, the trajectory data and task data of the same AGV are fused into a three-dimensional tensor structure. The first dimension of the tensor is the time slice index, the second dimension is the path segment number, and the third dimension is the fused multidimensional feature vector. The multidimensional feature vector includes coordinate offset, instantaneous acceleration, task priority weight, and shelf coordinate deviation. Tensor decomposition algorithm is introduced to perform low-rank approximation on the three-dimensional tensor. The product of the core tensor and the factor matrix is ​​calculated by alternating least squares method to extract the scheduling correlation features of each path segment of AGV driving in the spatiotemporal dimension. Based on the scheduling association features, a spatiotemporal path distribution matrix is ​​constructed, where the row vectors of the matrix represent the dynamic weights of each path segment within a time slice, the column vectors represent the probability distribution of the same path segment in different time slices, and the matrix element values ​​are determined by the product of the Euclidean distance of the path segment association features and the task priority weight.

[0021] It should be noted that in intelligent warehousing, the AGV scheduling system and inertial navigation construct multiple driving paths on the ground of the target warehouse for AGVs to choose from. When there are other obstacles on the path, the AGV will switch to another path. The task types include picking, replenishing, and returning tasks. The scheduling association features of each path segment in the spatiotemporal dimension are implicit features extracted from the three-dimensional tensor generated by fusing the spatiotemporal dynamic trajectory data of the AGV (including real-time coordinate sequence, speed, and heading angle) and task attribute vectors (such as task type, target shelf coordinates, and priority) using the tensor decomposition algorithm. Specifically, it manifests as the continuity of path segment usage in the time dimension (such as the periodic fluctuation of path congestion during peak periods) and the cooperative or competitive relationship between adjacent path segments in the spatial dimension (such as the probability of path intersection in densely packed shelf areas). The scheduling association features explain how the AGV scheduling system of the target warehouse automatically plans, avoids obstacles, and changes paths for AGVs. The spatiotemporal path distribution matrix is ​​a two-dimensional probability matrix dynamically generated based on the fusion of AGV inertial navigation data and task scheduling instructions. Its row dimension represents discretized time slices (e.g., 5-minute intervals), and the column dimension represents path segments (grids or nodes) divided in the warehouse map. The matrix element values ​​are calculated from the spatiotemporal correlation features extracted from the three-dimensional fusion tensor (time slice × path segment × multi-dimensional features) using a tensor decomposition algorithm. Specifically, it quantifies the dynamic weight or probability of each path segment being selected by the AGV scheduling system within different time slices. The weight calculation incorporates the temporal continuity of the path segments (e.g., periodic congestion); the more congested the path segment, the lower the dynamic weight of the path segment being selected by the AGV scheduling system, as well as spatial proximity and cooperative relationships (e.g., the competition probability of intersecting path segments). This matrix can predict the probability of path conflicts in real time.

[0022] According to an embodiment of the present invention, the step of analyzing the spatiotemporal path distribution matrix based on Markov chains to predict the position distribution of each AGV within a preset time period and constructing a position occupancy grid map specifically involves: The real-time task scheduling instruction data of each AGV in the target warehouse is obtained. The Markov chain method is introduced to map each path segment number in the spatiotemporal path distribution matrix to the state node of the Markov chain, and the real-time task scheduling instruction data is attached as the state attribute. Based on the column vector probability distribution of the spatiotemporal path distribution matrix, the basic transition probability of the path segment in different time slices is extracted. Then, according to the target shelf coordinate deviation and task priority parameters of the real-time scheduling task, the adjustment factor is calculated. The basic transition probability and the adjustment factor are weighted and fused to generate the state transition probability matrix. The current path segment of the AGV is taken as the initial state. Multiple candidate path sequences for future time slices are generated by Monte Carlo tree search. The selection weight of each candidate path is determined according to the state transition probability matrix. The AGV scheduling system predicts the scheduling path of the AGV in the future preset time period based on the selection weight. The location distribution of each AGV in the target warehouse is determined based on the scheduling prediction path within a future preset time period, and a grid map of the AGV's location occupancy at each time node within the future preset time period is constructed based on the location distribution.

[0023] It should be noted that by mapping each path segment number in the spatiotemporal path distribution matrix to a state node of a Markov chain and attaching real-time task scheduling instruction data as a state attribute, historical path data and real-time task information can be fully combined, improving the accuracy and adaptability of the prediction model. Secondly, based on the column vector probability distribution of the spatiotemporal path distribution matrix, the basic transition probability of the path segment is extracted, and an adjustment factor is calculated by combining the target shelf coordinate deviation and task priority parameters of the real-time scheduling task to generate a state transition probability matrix. This effectively reflects the impact of task priority on path selection, ensuring that the prediction results are consistent with actual scheduling needs. Thirdly, by generating multiple candidate path sequences for future time slices through Monte Carlo tree search and determining the selection weight of each candidate path in conjunction with the state transition probability matrix, multiple possible scheduling paths can be simulated, improving the comprehensiveness and robustness of the prediction. Finally, the location distribution of each AGV within a preset time period is determined based on the predicted scheduling path, and a location occupancy grid map is constructed, enabling the early identification of potential path conflict areas. Analyzing the real-time task scheduling instruction data and the spatiotemporal path distribution matrix using Markov chains can predict the AGV scheduling system's path planning for each AGV, thereby determining the AGV's driving status within each grid at each time point. The basic transition probability refers to the probability of a path segment transitioning between different time slices, extracted based on the column vector probability distribution of the spatiotemporal path distribution matrix. It reflects the probability that an AGV will transition from the current path segment to the next without additional task constraints. The calculation of the basic transition probability mainly relies on historical operational data, capturing the AGV's regular driving patterns in the warehousing environment. The adjustment factor is a dynamically corrected value calculated based on the target shelf coordinate deviation and task priority parameters of the real-time scheduling task. It is used to adjust the basic transition probability to reflect the impact of the current task on the AGV's path selection. For example, when the AGV needs to prioritize high-priority tasks, the adjustment factor will increase the transition probability of the target path segment and decrease the transition probability of other path segments. The introduction of the adjustment factor allows the state transition probability matrix to more accurately reflect actual scheduling needs. Markov chains stipulate that state transitions are only allowed between nodes in adjacent time slices and with spatially contiguous path segments, prohibiting illegal transitions across time slices or non-connected path segments.

[0024] Figure 2 A flowchart illustrating the present invention for determining the collision probability density of each grid cell in a target warehouse is shown.

[0025] According to an embodiment of the present invention, determining the conflict probability density of each grid cell in the target warehouse based on the location-occupied grid map specifically involves: S202, extract the AGV occupancy data of each time node in the future preset time period in the location occupancy grid map, and count the number of times each grid is occupied by multiple AGVs at different time nodes. S204, based on the time difference between the time node and the current time, assign a time decay factor to each time node, and weight the number of occupations with the corresponding time decay factor to generate an initial conflict probability value; S206, detect the intersection area of ​​each AGV predicted path within a future preset time period, extract the number of path intersections as a dynamic adjustment parameter, and perform weighted correction on the initial conflict probability value; S208 uses a kernel density estimation algorithm to spatially smooth the corrected collision probability values, eliminate local abnormal peaks, and output the collision probability density distribution of each grid cell.

[0026] It should be noted that the process involves extracting AGV occupancy data for each time node within a preset future time period, counting the number of times each grid is simultaneously occupied by multiple AGVs, and then weighting and summing these occupancy counts using a time decay factor to generate an initial conflict probability value. This quantifies the potential conflict frequency of a grid under multi-AGV collaborative operations. Secondly, by detecting the intersection areas of predicted paths for each AGV within the preset future time period, the number of path intersections is extracted as a dynamic adjustment parameter to weight and correct the initial conflict probability value, further improving the dynamism and accuracy of the conflict probability assessment. Finally, a kernel density estimation algorithm is used to spatially smooth the corrected conflict probability value, eliminating local abnormal peaks and making the conflict probability density distribution more continuous and reliable. The final output conflict probability density distribution provides the AGV scheduling system with accurate conflict risk information, helping the system to identify high-conflict areas in advance and optimize path planning, thereby significantly reducing the path conflict rate in multi-AGV collaborative operations and improving warehouse operation efficiency and safety.

[0027] Figure 3 A flowchart illustrating the network transmission quality assessment method of this invention is shown.

[0028] According to an embodiment of the present invention, the network transmission quality assessment of each grid in the target warehouse specifically includes: S302, based on the conflict probability density, predict the number of AGVs traveling in each grid at each time node within a future preset time period, and obtain the signal receiving location information of the AGVs from the AGV scheduling system; S304, obtain the wireless network transmission signal obstruction status between the signal receiving location and each grid, and determine the interaction signal strength loss value between the AGV and the AGV scheduling system in each grid based on the wireless network transmission signal obstruction status; S306, determine the network channel occupancy rate of each grid based on the AGV driving quantity information, and evaluate the network transmission quality of each grid based on the interaction signal strength loss value and the network channel occupancy rate to obtain the network quality evaluation result of each grid in the future preset time period.

[0029] It should be noted that in intelligent warehousing systems, AGV path planning and scheduling rely on real-time communication between the AGV scheduling system and the AGVs. However, in actual operation, the instability of network transmission quality directly affects the timeliness and accuracy of scheduling instructions, leading to reduced AGV operating efficiency and even path conflicts. Firstly, in complex warehousing environments, shelves, walls, and other obstacles can obstruct wireless network signals, causing signal attenuation, especially in areas with dense AGV traffic, where signal attenuation is more significant and may cause communication delays or interruptions. Secondly, when multiple AGVs operate in the same area, network channel resources are heavily occupied, leading to channel congestion, reduced data transmission rates, and even preventing some AGVs from receiving scheduling instructions in a timely manner. Thirdly, the number and location distribution of AGVs in the warehousing environment are dynamically changing, causing significant fluctuations in network transmission quality at different times and in different spaces. Evaluating the network transmission quality of each grid in the target warehouse, by predicting the number of AGVs in each grid within a preset future time period, analyzing signal obstruction and channel occupancy, and dynamically assessing network transmission quality provides more reliable data support for path planning, thereby effectively addressing the impact of network transmission quality instability on the AGV scheduling system.

[0030] According to an embodiment of the present invention, determining the AGV path planning scheme based on the collision probability density of each grid and the network transmission quality specifically involves: According to the conflict probability density, if the conflict probability density of each grid in the target warehouse is less than the preset conflict probability density, the current path automatically planned by the AGV scheduling system of the target warehouse will be used as the AGV scheduling planning path. It should be noted that when the conflict probability density of each grid in the target warehouse is less than the preset conflict probability density, it means that the path automatically planned by the AGV scheduling system will not have path conflicts, and the current path automatically planned by the AGV scheduling system can be used as the AGV scheduling planning path.

[0031] If the target warehouse is stored in a grid with a conflict probability density not less than a preset conflict probability density, the grid is marked as a grid to be adjusted, the network quality assessment result of the grid to be adjusted is obtained, and the delay coefficient of the AGV scheduling system sending the path adjustment scheduling command to the AGV in the grid to be adjusted is determined based on the network quality assessment result. If the delay coefficient of the path adjustment scheduling command is greater than the preset delay coefficient, the movement speed of the AGV in the grid to be adjusted will be reduced to a safe threshold, and the AGV forklifts with path conflicts will be switched to manual control to execute manual control path planning. It should be noted that when the target warehouse is stored in a grid with a conflict probability density not less than the preset conflict probability density, it indicates that there is a high risk of path conflict in that area within a preset time period. In this case, the AGV scheduling system needs to avoid conflict by sending path adjustment scheduling commands. However, if the delay coefficient of the path adjustment scheduling command is greater than the preset delay coefficient, it indicates that the network transmission quality in that area is poor, and the scheduling command cannot be transmitted to the AGV in time. This may lead to problems such as command delay exacerbating path conflict, network quality affecting scheduling reliability, and insufficient emergency response in dynamic environments. Therefore, by reducing the movement speed of the AGV in the grid to be adjusted to a safe threshold, the travel speed of the AGV can be effectively slowed down, giving the system more time to adjust the path and reducing the possibility of conflict. At the same time, switching the AGV with path conflict to manual control and executing manual control path planning can ensure the safe operation of the AGV and avoid collision accidents through manual intervention when the network transmission quality is poor.

[0032] If the delay coefficient of the path adjustment scheduling instruction is not greater than the preset delay coefficient, the Dijkstra algorithm performs secondary path planning on the AGV in the grid to be adjusted and outputs the secondary scheduling planning path. During the path execution process, when the coordinate deviation between the actual navigation trajectory of the AGV and the planned path exceeds the preset tolerance range, the heading angle is corrected in real time based on the inertial navigation data. If the deviation continues to increase after correction and reaches the maximum allowable deviation, the AGV will drive out of the AGV path planning area of ​​the target warehouse and switch to manual control to obtain an emergency avoidance planning path. The path planning scheme for AGVs is constructed by combining the current path automatically planned by the AGV scheduling system, the path planned by manual control, the path planned by secondary scheduling, and the path planned for emergency avoidance.

[0033] It should be noted that when the delay coefficient of the path adjustment scheduling command is not greater than the preset delay coefficient, although the network transmission quality meets the real-time scheduling requirements, secondary path planning is still required for high-conflict areas to ensure that the AGV can complete the task safely and efficiently. Therefore, when the delay coefficient of the path adjustment scheduling command is not greater than the preset delay coefficient, the Dijkstra algorithm is used to perform secondary path planning for the AGV in the grid to be adjusted, which can generate the optimal obstacle avoidance path and ensure the safe operation of the AGV in high-conflict areas. Secondly, during the path execution process, the heading angle is corrected in real time based on inertial navigation data, which can promptly correct the AGV's driving deviation and improve the accuracy of path tracking. Finally, when the deviation continues to increase and reaches the maximum allowable deviation, the AGV is driven out of the path planning area and switched to manual control, which can realize emergency avoidance and prevent accidents. By constructing the automatically planned current path, the manually controlled planned path, the secondary scheduling planned path, and the emergency avoidance planned path as the AGV path planning scheme, this invention can comprehensively cope with various challenges in dynamic environments and significantly improve the operational safety, path planning robustness, and warehousing operation efficiency of the AGV system.

[0034] According to an embodiment of the present invention, it further includes: A three-dimensional fluid dynamics simulation model is constructed by collecting real-time information on the stagnant location of faulty AGVs and the topology of surrounding paths in the target warehouse. The motion of the AGV cluster is abstracted as the motion of viscous fluid, and flow conservation equations and momentum conservation equations are set. The congestion propagation rate and direction of the path around the fault point are calculated through dynamic iteration of the density field. Based on real-time updated AGV position data, the boundary condition parameters of the fluid dynamics model are dynamically corrected, and an adaptive grid partitioning algorithm is used to predict the congestion index threshold of each path segment within the future time window. A dynamic diversion control coefficient matrix is ​​generated by combining the remaining capacity of the path network and task priority. The speed attenuation adjustment and path weight redistribution are performed on the upstream AGV through a distributed path planner. At the same time, virtual traffic lights are deployed at intersection nodes to realize time-slotted traffic control in conflict areas. A model predictive control closed loop is established to continuously optimize the diversion strategy until the path network returns to steady state.

[0035] It should be noted that in intelligent warehousing systems, when an AGV stops on the main road due to mechanical failure, traditional local avoidance strategies only adjust the detours of nearby AGVs, but cannot predict or prevent the path network-level cascading congestion caused by the failure. As subsequent AGVs continue to enter the failure area, the congestion will spread rapidly along the path topology, causing a sharp drop in the traffic capacity of multiple areas, large-scale task delays, and even system-wide paralysis. By constructing a three-dimensional fluid dynamics simulation model, the motion of the AGV cluster is mapped to viscous fluid motion. The flow conservation equation and momentum conservation equation are used to quantify the dynamic characteristics of congestion propagation, breaking through the accuracy limitations of traditional discrete event simulation. Dynamically correcting boundary condition parameters and combining adaptive mesh partitioning, real-time high-precision prediction of congestion propagation paths is achieved. A dynamic diversion control coefficient matrix is ​​generated and integrates the remaining capacity of the path network with task priority to avoid secondary congestion caused by random path adjustments. Speed ​​attenuation adjustment and path weight redistribution are coordinated with virtual traffic light control to achieve peak flow reduction in congested areas while ensuring the timeliness of critical tasks. The model predictive control closed loop continuously tracks system state changes through rolling optimization to ensure the dynamic optimality of the diversion strategy, ultimately achieving rapid steady-state recovery from local faults to the global path network. The virtual fluid dynamics simulation model is a mathematical simulation model that analogizes the motion of the AGV cluster in the path network to viscous fluid flow. It describes the macroscopic motion of the AGV group through fluid dynamics equations to predict congestion propagation trends. The density field dynamic iterative calculation divides the path network into grid cells and calculates the spatial gradient change of the AGV number density distribution within each cell in real time, predicting the direction of congestion spread through iterative updates. The dynamic diversion control coefficient matrix contains a mathematical matrix of diversion weights for each path segment. Its element values ​​are determined by the remaining capacity of the path, the urgency of the task, and the congestion propagation speed, and are used to generate the optimal diversion ratio. The distributed path planner, deployed locally on the AGV's intelligent decision-making module, receives the diversion coefficients from the central system and autonomously generates avoidance paths that conform to vehicle dynamics constraints, achieving coordination between global planning and local execution.

[0036] Figure 4 A block diagram of an AGV intelligent warehousing path planning system based on inertial navigation according to the present invention is shown.

[0037] A second aspect of the present invention also provides an AGV intelligent warehousing path planning system 4 based on inertial navigation. The system includes a memory 41 and a processor 42. The memory includes an AGV intelligent warehousing path planning method program based on inertial navigation. When the processor executes the AGV intelligent warehousing path planning method program based on inertial navigation, it performs the following steps: Based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a spatiotemporal path distribution matrix for AGV scheduling system to schedule AGVs is constructed. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a preset time period is predicted, and a position occupancy grid map is constructed. The conflict probability density of each grid cell in the target warehouse is determined based on the location-occupied grid map. Evaluate the network transmission quality for each grid in the target warehouse; The path planning scheme for the AGV is determined based on the collision probability density of each grid and the network transmission quality.

[0038] This invention discloses an AGV intelligent warehousing path planning method and system based on inertial navigation. The method first constructs a spatiotemporal path distribution matrix for AGV scheduling based on historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse. Next, a Markov chain is used to analyze this matrix in depth to predict the position distribution of each AGV within a preset time period, thereby constructing a position occupancy grid map. This map determines the conflict probability density of each grid in the target warehouse, and network transmission quality is evaluated for each grid. Finally, by combining the conflict probability density and network transmission quality of each grid, a scientifically reasonable AGV path planning scheme is determined. This invention can effectively improve the efficiency and accuracy of AGV path planning in intelligent warehousing, reduce conflict risks, optimize network transmission, and provide strong support for the efficient operation of intelligent warehousing.

[0039] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0040] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0041] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0042] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0043] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AGV intelligent warehouse path planning method based on inertial navigation, characterized in that, Includes the following steps: Based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a spatiotemporal path distribution matrix for AGV scheduling system to schedule AGVs is constructed. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a preset time period is predicted, and a position occupancy grid map is constructed. The conflict probability density of each grid cell in the target warehouse is determined based on the location-occupied grid map. Evaluate the network transmission quality for each grid in the target warehouse; The path planning scheme for the AGV is determined based on the collision probability density of each grid and the network transmission quality.

2. The AGV intelligent warehousing path planning method based on inertial navigation according to claim 1, characterized in that, The process of constructing a spatiotemporal path distribution matrix for AGV scheduling by the AGV scheduling system based on historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse is as follows: The target warehouse is divided into an M×N gridded map. Based on the timestamp, the historical inertial navigation data of each AGV for a preset data acquisition period is divided into time windows. The real-time coordinate sequence, motion speed and heading angle data of the AGV in each time window are extracted to obtain spatiotemporal dynamic trajectory data. The task scheduling instruction data within the preset data acquisition time period is parsed to extract the current task type, target shelf coordinates, and task priority parameters for each AGV, and a task attribute vector is constructed. The spatiotemporal dynamic trajectory data and the task attribute vector are spatiotemporally aligned. Based on the AGV identifier, the trajectory data and task data of the same AGV are fused into a three-dimensional tensor structure. The first dimension of the tensor is the time slice index, the second dimension is the path segment number, and the third dimension is the fused multidimensional feature vector. The multidimensional feature vector includes coordinate offset, instantaneous acceleration, task priority weight, and shelf coordinate deviation. Tensor decomposition algorithm is introduced to perform low-rank approximation on the three-dimensional tensor. The product of the core tensor and the factor matrix is ​​calculated by alternating least squares method to extract the scheduling correlation features of each path segment of AGV driving in the spatiotemporal dimension. Based on the scheduling association features, a spatiotemporal path distribution matrix is ​​constructed, where the row vectors of the matrix represent the dynamic weights of each path segment within a time slice, the column vectors represent the probability distribution of the same path segment in different time slices, and the matrix element values ​​are determined by the product of the Euclidean distance of the path segment association features and the task priority weight.

3. The AGV intelligent warehousing path planning method based on inertial navigation according to claim 1, characterized in that, The analysis of the spatiotemporal path distribution matrix based on Markov chains to predict the position distribution of each AGV within a preset time period and construct a position occupancy grid map is specifically as follows: The real-time task scheduling instruction data of each AGV in the target warehouse is obtained. The Markov chain method is introduced to map each path segment number in the spatiotemporal path distribution matrix to the state node of the Markov chain, and the real-time task scheduling instruction data is attached as the state attribute. Based on the column vector probability distribution of the spatiotemporal path distribution matrix, the basic transition probability of the path segment in different time slices is extracted. Then, according to the target shelf coordinate deviation and task priority parameters of the real-time scheduling task, the adjustment factor is calculated. The basic transition probability and the adjustment factor are weighted and fused to generate the state transition probability matrix. The current path segment of the AGV is taken as the initial state. Multiple candidate path sequences for future time slices are generated by Monte Carlo tree search. The selection weight of each candidate path is determined according to the state transition probability matrix. The AGV scheduling system predicts the scheduling path of the AGV in the future preset time period based on the selection weight. The location distribution of each AGV in the target warehouse is determined based on the scheduling prediction path within a future preset time period, and a grid map of the AGV's location occupancy at each time node within the future preset time period is constructed based on the location distribution.

4. The AGV intelligent warehousing path planning method based on inertial navigation according to claim 1, characterized in that, The step of determining the conflict probability density of each grid cell in the target warehouse based on the location-occupied grid map is specifically as follows: Extract AGV occupancy data for each time node within a future preset time period from the location-occupied grid map, and count the number of times each grid is simultaneously occupied by multiple AGVs at different time nodes; Based on the time difference between the time node and the current time, a time decay factor is assigned to each time node, and the number of occupations is weighted and summed with the corresponding time decay factor to generate an initial conflict probability value. The intersection area of ​​each AGV's predicted path within a preset time period is detected, and the number of path intersections is extracted as a dynamic adjustment parameter to perform weighted correction on the initial conflict probability value. The corrected collision probability values ​​are spatially smoothed using a kernel density estimation algorithm to eliminate local abnormal peaks and output the collision probability density distribution for each grid cell.

5. The AGV intelligent warehousing path planning method based on inertial navigation according to claim 1, characterized in that, The network transmission quality assessment for each grid cell in the target warehouse specifically involves: Based on the conflict probability density, predict the number of AGVs traveling in each grid at each time node within a future preset time period, and obtain the signal reception location information of the AGVs by the AGV scheduling system. The wireless network transmission signal obstruction between the signal receiving location and each grid is obtained, and the interaction signal strength loss value between the AGV and the AGV scheduling system in each grid is determined based on the wireless network transmission signal obstruction. The network channel occupancy rate of each grid is determined based on the AGV driving quantity information. The network transmission quality of each grid is evaluated based on the interaction signal strength loss value and the network channel occupancy rate, and the network quality evaluation result of each grid is obtained within a preset time period.

6. The AGV intelligent warehousing path planning method based on inertial navigation according to claim 1, characterized in that, The method for determining the AGV's path planning scheme based on the collision probability density of each grid and the network transmission quality is as follows: According to the conflict probability density, if the conflict probability density of each grid in the target warehouse is less than the preset conflict probability density, the current path automatically planned by the AGV scheduling system of the target warehouse will be used as the AGV scheduling planning path. If the target warehouse is stored in a grid with a conflict probability density not less than a preset conflict probability density, the grid is marked as a grid to be adjusted, the network quality assessment result of the grid to be adjusted is obtained, and the delay coefficient of the AGV scheduling system sending the path adjustment scheduling command to the AGV in the grid to be adjusted is determined based on the network quality assessment result. If the delay coefficient of the path adjustment scheduling command is greater than the preset delay coefficient, the movement speed of the AGV in the grid to be adjusted will be reduced to a safe threshold, and the AGV forklifts with path conflicts will be switched to manual control to execute manual control path planning. If the delay coefficient of the path adjustment scheduling instruction is not greater than the preset delay coefficient, the Dijkstra algorithm performs secondary path planning on the AGV in the grid to be adjusted and outputs the secondary scheduling planning path. During the path execution process, when the coordinate deviation between the actual navigation trajectory of the AGV and the planned path exceeds the preset tolerance range, the heading angle is corrected in real time based on the inertial navigation data. If the deviation continues to increase after correction and reaches the maximum allowable deviation, the AGV will drive out of the AGV path planning area of ​​the target warehouse and switch to manual control to obtain an emergency avoidance planning path. The path planning scheme for AGVs is constructed by combining the current path automatically planned by the AGV scheduling system, the path planned by manual control, the path planned by secondary scheduling, and the path planned for emergency avoidance.

7. An AGV intelligent warehousing path planning system based on inertial navigation, characterized in that, The AGV intelligent warehousing path planning system based on inertial navigation includes a storage unit and a processor. The storage unit includes an AGV intelligent warehousing path planning method program based on inertial navigation. When the processor executes the AGV intelligent warehousing path planning method program based on inertial navigation, it performs the following steps: Based on the historical inertial navigation data and historical task scheduling instruction data of each AGV in the target warehouse, a spatiotemporal path distribution matrix for AGV scheduling system to schedule AGVs is constructed. Based on the analysis of the spatiotemporal path distribution matrix using Markov chains, the position distribution of each AGV within a preset time period is predicted, and a position occupancy grid map is constructed. The conflict probability density of each grid cell in the target warehouse is determined based on the location-occupied grid map. Evaluate the network transmission quality for each grid in the target warehouse; The path planning scheme for the AGV is determined based on the collision probability density of each grid and the network transmission quality.

Images