A stereoscopic shelf unmanned aerial vehicle inventory method and system

Abstract

The application discloses a kind of stereoscopic shelf unmanned aerial vehicle inventory methods and systems, comprising: according to the warehouse map data and inventory task point of collection constructs warehouse topology graph, and based on warehouse topology graph planning global inventory path;Unmanned aerial vehicle according to global inventory path and collected warehouse environment information, and in combination with the initial obstacle avoidance trajectory of unmanned aerial vehicle position information generation, the warehouse environment information indicates the perception data of the spatial distribution of obstacle in the current local airspace of the unmanned aerial vehicle;Trajectory parameter optimization is carried out to the initial obstacle avoidance trajectory, and execution obstacle avoidance trajectory is generated;The unmanned aerial vehicle based on the execution obstacle avoidance trajectory gathers goods identification information, and triggers indicating device to carry out physical indication to goods, and outputs goods information.Compared with prior art, it can significantly improve the traversal efficiency of stereoscopic shelf inventory, obstacle avoidance safety and goods positioning accuracy, overcome the technical shortcomings of path redundancy, obstacle avoidance harsh and weak interactive feedback in traditional scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing and logistics automation technology, and more specifically, to a method and system for inventory counting using drones on automated storage and retrieval systems. Background Technology

[0002] With the large-scale development of intelligent warehousing and modern logistics, automated storage and retrieval systems (AS / RS), as a core facility for improving space utilization, have become crucial in influencing logistics turnover efficiency due to the increasing sophistication and automation of their inventory management. Modern warehousing scenarios require inventory systems to possess comprehensive capabilities such as high-frequency, full-coverage scanning, rapid location of specific goods, and safe operation in complex environments. Traditional manual, gun-wielding inventory checks or ground-based AGV-based planar inspections, limited by operating height, blind spots, and mobility, can no longer meet the efficient management needs of large-scale, high-density AS / RS. Drones, with their three-dimensional spatial mobility, are gradually becoming an important carrier for AS / RS inventory checks. Their path planning efficiency, obstacle avoidance capabilities in confined spaces, and interactive experience combining data and reality directly determine the level of intelligence and operational safety in inventory management.

[0003] While some existing technologies attempt to achieve drone inventory through preset waypoints or remote control, they still suffer from a lack of coordination between global scheduling efficiency and local flight safety. This results in existing drone inventory systems exhibiting problems such as redundant traversal paths, insufficient dynamic obstacle avoidance capabilities, and weak interactive feedback in complex application scenarios involving automated storage and retrieval systems.

[0004] Therefore, how to provide a method and system for inventory counting of automated storage and retrieval systems using drones, which can significantly improve the traversal efficiency, obstacle avoidance safety, and cargo positioning accuracy of automated storage and retrieval systems, while overcoming the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance, and weak interactive feedback, has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for inventorying automated storage and retrieval systems using drones. This method significantly improves the traversal efficiency, obstacle avoidance safety, and cargo positioning accuracy of automated storage and retrieval systems, overcoming the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance, and weak interactive feedback.

[0006] The first technical solution provided by this invention is as follows:

[0007] This invention provides a method for inventorying automated storage and retrieval systems using drones. S1: A ground station collects warehouse map data and inventory task points, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map. S2: The drone receives the global inventory path and collects warehouse environment information. Based on the global inventory path and the warehouse environment information, and combined with the drone's position information, an initial obstacle avoidance trajectory is generated. The warehouse environment information represents the perceived spatial distribution of obstacles within the drone's current local airspace. S3: The trajectory parameters of the initial obstacle avoidance trajectory are optimized to generate an executed obstacle avoidance trajectory. S4: The drone collects cargo identification information based on the executed obstacle avoidance trajectory, triggers an indicator device to physically indicate the cargo, and outputs cargo information.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of planning a global inventory route based on the warehouse topology map includes:

[0009] Construct a warehouse topology map, which includes: spatial nodes representing path intersections and inventory task points, and guide path segments representing the passability between adjacent spatial nodes;

[0010] Extract all inventory task points, calculate the shortest travel distance between any two inventory task points based on the warehouse topology map, and construct a distance matrix based on several of the shortest travel distances. The distance matrix is ​​used to quantify the travel cost of the travel distance.

[0011] The LKH-3 algorithm is used to iteratively search the distance matrix to generate a task access sequence. The corresponding guiding path segments are then connected according to the task access sequence to generate a global inventory path.

[0012] Furthermore, in a preferred embodiment of the present invention, the step of constructing a distance matrix based on a plurality of the shortest travel distances includes:

[0013] Construct an N×N matrix, where N represents the total number of the inventory task points;

[0014] Traverse the inventory task points. For any two different inventory task points i and j, search for the shortest path between inventory task points i and j based on Dijkstra's algorithm, and calculate the physical travel distance between inventory task points i and j.

[0015] Assign the physical travel distance to the element in the i-th row and j-th column of the matrix to construct the distance matrix.

[0016] Further, in a preferred embodiment of the present invention, the step of iteratively searching the distance matrix using the LKH-3 algorithm to generate a task access sequence includes:

[0017] Construct an initial traversal path, which includes all the inventory task points;

[0018] Side-swapping operation: Remove the preset old path segments of the initial traversal path, introduce new path segments with the same number of preset old path segments, reconnect the path breakpoints, and construct traversal candidate paths;

[0019] Distance calculation operation: Identify the adjacent connection relationships of the traversal candidate paths, and calculate the physical travel distance of the traversal candidate paths based on the distance matrix;

[0020] If the physical travel distance of the candidate path is less than that of the initial traversal path, the initial traversal path is replaced, and the edge-swapping operation and distance calculation operation are repeated until the physical travel distance of any candidate path generated by the edge-swapping operation is not less than that of the previous candidate path. The current candidate path is then output as the task access sequence.

[0021] Furthermore, in a preferred embodiment of the present invention, the follow-up inventory mode specifically includes:

[0022] Receive a follow-up command, which contains the main feature information of the moving subject to be followed, wherein the moving subject is a warehouse worker or a mobile transportation device;

[0023] The drone searches for and locks onto the moving subject, calculates the motion vector of the moving subject in real time, and generates a dynamic following trajectory that maintains a preset relative spatial position with the moving subject;

[0024] Based on the dynamic following trajectory, the drone scans the cargo identification information on the shelves on both sides of the flight path in real time, and maps the cargo identification information to the warehouse map in combination with the drone's current real-time position coordinates to complete the follow-up inventory.

[0025] Further, in a preferred embodiment of the present invention, the step of generating an initial obstacle avoidance trajectory based on the global inventory path and the warehouse environment information, combined with the UAV location information, includes:

[0026] The global inventory path is parsed, and subsequent target waypoints in the flight sequence are extracted from the global inventory path based on the UAV's location information. The subsequent target waypoints are then set as the local planning endpoints.

[0027] Based on the warehouse environment information, a voxel grid map is constructed, and the obstacle occupancy status of each voxel unit in the voxel grid map is identified.

[0028] Based on the obstacle occupancy status, a jump point search algorithm is used to generate an initial obstacle avoidance trajectory, starting from the UAV's location information and ending at the local planning endpoint.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of optimizing the trajectory parameters of the initial obstacle avoidance trajectory includes:

[0030] The initial obstacle avoidance trajectory is converted into a B-spline curve, which contains several control points for determining the shape of the B-spline curve.

[0031] A multidimensional objective function is constructed based on the B-spline curve. The multidimensional objective function is as follows:

[0032] ;

[0033] in, , , as well as Indicates the weight of each item;

[0034] The smoothness cost is represented by the sum of the squares of the acceleration and jerk of the control point;

[0035] The obstacle distance cost is represented as: ;

[0036] This indicates the distance between the control point and the nearest obstacle. This represents the preset penalty function;

[0037] This represents the dynamic cost, used to indicate the portion exceeding the maximum speed limit of the drone. This indicates the total flight time of the drone.

[0038] Furthermore, in a preferred embodiment of the present invention, a targeted inventory mode is also included, specifically:

[0039] Receive and respond to the goods selection instruction initiated by the user, and identify the specified goods according to the goods selection instruction;

[0040] Based on the warehouse map data parsing, the target location coordinates corresponding to the specified goods are obtained, and the target location coordinates are sent to the drone;

[0041] The drone receives the target location coordinates, generates a directional trajectory, and flies to the designated cargo according to the directional trajectory, obtains cargo identification information, and triggers the indication device to execute physical instructions.

[0042] Furthermore, in a preferred embodiment of the present invention, a scheduling algorithm for switching the targeted inventory mode is further included, specifically:

[0043] The system monitors input signals on the human-computer interaction interface in real time. When it detects that a user has executed a selection command for a specific item in the warehouse map data, it generates a task preemption command containing the three-dimensional coordinates of the specified item.

[0044] The UAV receives the task preemption command via a wireless communication link and parses the command words in the command to identify the current task type as a targeted inventory task.

[0045] In response to the directional inventory task, the drone immediately suspends the currently executing global traversal process or exits the standby hovering state and clears the current flight waypoint cache queue.

[0046] The drone loads the three-dimensional coordinates of the designated cargo, identifies the target navigation point, calculates the navigation path based on the target navigation point, and initiates the directional inventory mode.

[0047] The present invention provides a second technical solution as follows:

[0048] The present invention also provides a three-dimensional shelf drone inventory system, comprising:

[0049] The global path planning module collects warehouse map data and inventory task points from the ground station, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map.

[0050] The flight path planning module allows the UAV to receive the global inventory path and collect warehouse environment information. Based on the global inventory path and the warehouse environment information, and combined with the UAV's position information, it generates an initial obstacle avoidance trajectory. The warehouse environment information represents the perception data of the spatial distribution of obstacles in the current local airspace of the UAV.

[0051] The route optimization module optimizes the trajectory parameters of the initial obstacle avoidance trajectory to generate an execution obstacle avoidance trajectory;

[0052] The dual-verification module identifies cargo identification information based on the obstacle avoidance trajectory, triggers an indicator device to physically indicate the cargo, and outputs cargo information.

[0053] This invention provides a drone-based inventory method for automated storage and retrieval systems (AS / RS), which significantly improves the traversal efficiency, obstacle avoidance safety, and cargo positioning accuracy of AS / RS inventory operations, overcoming the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance, and weak interactive feedback. The drone-based AS / RS inventory method includes: S1. A ground station collects warehouse map data and inventory task points, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map; S2. The drone receives the global inventory path and collects warehouse environment information, generates an initial obstacle avoidance trajectory based on the global inventory path and the warehouse environment information, combined with the drone's position information, where the warehouse environment information represents the perceived data of the spatial distribution of obstacles in the drone's current local airspace; S3. The initial obstacle avoidance trajectory is optimized to generate an executed obstacle avoidance trajectory; S4. The drone collects cargo identification information based on the executed obstacle avoidance trajectory, triggers an indicator device to physically indicate the cargo, and outputs cargo information. The ground station utilizes collected warehouse map data and inventory task points to map the unstructured physical warehouse space into a warehouse topology map composed of spatial nodes and guiding path segments. Based on the topological connection relationship, the access order of all task points is forcibly locked, which can significantly improve the traversal efficiency of automated racking inventory. Then, the UAV combines warehouse environment information to generate an initial obstacle avoidance trajectory under global path guidance, and performs trajectory parameter optimization on the initial trajectory to generate an execution obstacle avoidance trajectory. A feasible channel that meets the geometric collision-free conditions is constructed through the initial trajectory, and the spatial position avoidance logic is established. Subsequently, through parameter optimization, the polyline path in the channel is processed to limit the UAV's acceleration abrupt changes and attitude oscillations during avoidance actions, thereby improving the UAV's obstacle avoidance safety. Based on the execution obstacle avoidance trajectory, the indicator device is triggered to physically indicate the goods while collecting information, effectively overcoming the problems of weak interactive feedback and difficulty in confirming the location of goods caused by traditional operations. Compared with existing technologies, this solution can significantly improve the traversal efficiency, obstacle avoidance safety and goods positioning accuracy of automated racking inventory, and overcome the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance and weak interactive feedback. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart illustrating the steps of the automated shelving drone inventory method provided in this embodiment of the invention;

[0056] Figure 2 This is a schematic diagram of the inventory site provided in an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of an inventory rack provided in an embodiment of the present invention;

[0058] Figure 4 The drone inventory route map provided for embodiments of the present invention;

[0059] Figure 5 This is a schematic diagram of the process for generating a global inventory path according to an embodiment of the present invention;

[0060] Figure 6 A schematic diagram illustrating the process of constructing a distance matrix according to an embodiment of the present invention;

[0061] Figure 7 This is a structural block diagram of the automated shelving drone inventory system provided in an embodiment of the present invention. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0063] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly set on the other component; when a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to the other component.

[0064] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "first", "second", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0065] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" or "several" means two or more, unless otherwise explicitly specified.

[0066] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0067] like Figures 1 to 7 As shown in the figure, the UAV inventory method for three-dimensional shelving provided by the present invention can significantly improve the traversal efficiency, obstacle avoidance safety and cargo positioning accuracy of three-dimensional shelving inventory, and overcome the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance and weak interactive feedback.

[0068] This invention provides a method for inventorying automated storage and retrieval systems using drones, specifically including: S1, a ground station collects warehouse map data and inventory task points, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map; S2, the drone receives the global inventory path and collects warehouse environment information, generates an initial obstacle avoidance trajectory based on the global inventory path and the warehouse environment information, combined with the drone's position information, wherein the warehouse environment information represents the perceived data of the spatial distribution of obstacles in the drone's current local airspace; S3, the trajectory parameters of the initial obstacle avoidance trajectory are optimized to generate an execution obstacle avoidance trajectory; S4, the drone collects cargo identification information based on the execution obstacle avoidance trajectory, triggers an indicator device to physically indicate the cargo, and outputs cargo information. The ground station utilizes collected warehouse map data and inventory task points to map the unstructured physical warehouse space into a warehouse topology map composed of spatial nodes and guiding path segments. Based on the topological connection relationship, the access order of all task points is forcibly locked, which can significantly improve the traversal efficiency of automated racking inventory. Then, the UAV combines warehouse environment information to generate an initial obstacle avoidance trajectory under global path guidance, and performs trajectory parameter optimization on the initial trajectory to generate an execution obstacle avoidance trajectory. A feasible channel that meets the geometric collision-free conditions is constructed through the initial trajectory, and the spatial position avoidance logic is established. Subsequently, through parameter optimization, the polyline path in the channel is processed to limit the UAV's acceleration abrupt changes and attitude oscillations during avoidance actions, thereby improving the UAV's obstacle avoidance safety. Based on the execution obstacle avoidance trajectory, the indicator device is triggered to physically indicate the goods while collecting information, effectively overcoming the problems of weak interactive feedback and difficulty in confirming the location of goods caused by traditional operations. Compared with existing technologies, this solution can significantly improve the traversal efficiency, obstacle avoidance safety and goods positioning accuracy of automated racking inventory, and overcome the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance and weak interactive feedback.

[0069] Tests have verified that the total inventory time of this invention is less than 55 minutes, while the total time for manual inventory is more than 80 minutes, representing an improvement of 31%. The inventory accuracy of this invention is greater than 99.9%, with a missed scan rate of less than 0.2%, while the traditional manual inventory is greater than 99%, with a missed scan rate of more than 1.0%. Therefore, the technical solution involved in this invention, compared with the prior art, can significantly improve the traversal efficiency, obstacle avoidance safety, and cargo positioning accuracy of automated shelving inventory, overcoming the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance, and weak interactive feedback.

[0070] More specifically, in this embodiment, the UAV adopts a heterogeneous dual-controller architecture consisting of Jetson Orin Nano, CUAV V5, and others. Jetson Orin Nano, as the upper-level companion computer, is responsible for processing all visual SLAM mapping, target recognition algorithms, and trajectory parameter optimization calculations based on B-splines, achieving high-performance local edge computing. CUAV V5+, as the lower-level flight controller, is only responsible for receiving the final control commands output by the upper-level computer and driving the motors to execute them.

[0071] The following detailed description of the steps and procedures of the automated shelving drone inventory method and system, using specific embodiments, is provided.

[0072] Specifically, such as Figure 5 As shown, in a specific embodiment of the present invention, the step of planning a global inventory path based on the warehouse topology map includes: S21 constructing a warehouse topology map, which includes: spatial nodes representing path intersections and inventory task points, and guiding path segments representing the traversable relationship between adjacent spatial nodes; S22 extracting all inventory task points, calculating the shortest traversal distance between any two inventory task points based on the warehouse topology map, and constructing a distance matrix based on several shortest traversal distances, the distance matrix being used to quantify the traversal cost of the traversal distance; S23 using the LKH-3 algorithm to iteratively search the distance matrix to generate a task access sequence, connecting the corresponding guiding path segments according to the task access sequence to generate a global inventory path.

[0073] In a specific embodiment of the present invention, the ground station first loads pre-stored warehouse map data and a list of inventory task points to be executed through a data interface. The warehouse map data records in detail the physical boundaries of the warehouse, the layout of the shelves, the location of obstacles, and the information of walkable passages. In order to transform the complex physical environment into a mathematical model that can be processed by a computer, the system constructs a warehouse topology map based on the above data. This construction process specifically adopts the idea of ​​graph theory modeling, which abstracts the key geographical locations in the warehouse map data. For example, the intersections and turning points of warehouse aisles and the hovering positions of each inventory task point are defined as spatial nodes in the topology. At the same time, the physical corridors or aisles that connect these spatial nodes and meet the conditions for drone passage are defined as guiding path segments, thereby forming a digital network structure that can represent the spatial connectivity of the warehouse. Based on this, the system plans a global inventory path using the warehouse topology map, solving for an efficient access sequence that covers all inventory task points. The system analyzes the connection relationships between spatial nodes in the topology map and the physical attributes of the guiding path segments, such as path length, assesses the passage cost of the UAV moving between different nodes, and uses combinatorial optimization strategies or path search logic to search for a connected route in the topology network that satisfies traversal constraints. This route establishes the order of accessing each inventory task point, and the corresponding guiding path segments are connected end to end in this order, ultimately generating a global inventory path that starts from the starting point and includes the coordinate sequence of all target task points, providing macro-level navigation guidance for the subsequent autonomous flight of the UAV.

[0074] Specifically, such as Figure 6 As shown, in a specific embodiment of the present invention, a distance matrix is ​​constructed based on several shortest travel distances, including: S221 establishing an N×N matrix, where N represents the total number of inventory task points; S222 traversing the inventory task points, for any two different inventory task points i and j, searching for the shortest path between inventory task points i and j based on Dijkstra's algorithm, and calculating the physical travel distance between inventory task points i and j; S223 assigning the physical travel distance to the element in the i-th row and j-th column of the matrix to construct the distance matrix.

[0075] In an embodiment of the present invention, in order to transform the complex physical access network in the warehousing environment into a digital cost model that can be efficiently processed by the algorithm, the system performs a distance matrix construction step based on the shortest access distance. Specifically, the system first reads the task point list of the inventory task to be performed, counts the total number of task points, and constructs a two-dimensional square matrix as a distance matrix. The row and column dimensions of this matrix strictly correspond to the total number of inventory task points, ensuring that the matrix can accommodate the connection weight information between any two nodes in the network. Subsequently, the system enters the path calculation stage of full permutation, which calculates the path weight information between any two nodes in the matrix. For elements in different combinations of inventory task points, the system uses Dijkstra's shortest path algorithm to traverse and search the pre-constructed warehouse topology map. Since shelves obstruct the view in the automated warehouse, points typically cannot travel in a straight line. The system starts with one task point and expands the search outwards along accessible guide path segments in the topology network until a target node leading to another task point is found. During this process, the algorithm automatically accumulates the actual physical length of each guide path segment on the shortest path, thus calculating the actual shortest physical distance between the two points that needs to bypass obstacles. After calculation, the system directly assigns this specific distance value to the storage unit at the intersection of the corresponding row and column in the distance matrix. For example, if there are task points A and B in the system, the algorithm calculates that the shortest path from point A to point B along the alley is a specific number of meters. This number of meters is then written into the element in the A-th row and B-th column of the matrix. By performing the above calculation and assignment operations on all task point pairs, a complete numerical matrix that quantifies the overall passage cost is finally formed, providing accurate data support for subsequent global planning.

[0076] Specifically, in a specific embodiment of the present invention, the steps employ the LKH-3 algorithm to iteratively search the distance matrix and generate a task access sequence, including: constructing an initial traversal path, which includes all inventory task points; a side-swapping operation: removing preset old path segments from the initial traversal path and introducing new path segments with the same number as the preset old path segments, reconnecting path breakpoints, and constructing a traversal candidate path; a distance calculation operation: identifying the adjacent connection relationships of the traversal candidate paths and calculating the physical travel distance of the traversal candidate paths based on the distance matrix; if the physical travel distance of a candidate path is less than that of the initial traversal path, the initial traversal path is replaced, and the side-swapping operation and distance calculation operation are repeated until the physical travel distance of any candidate path generated by the side-swapping operation is not less than that of the previous candidate path, and the current candidate path is output as the task access sequence.

[0077] In an embodiment of the present invention, after the system constructs a distance matrix that quantifies the passage cost, it iteratively searches the matrix using the LKH-3 heuristic algorithm to find the globally optimal task access order. Specifically, the system first constructs an initial traversal path that connects all task points to be inventoried in any reasonable order, ensuring that each task point is visited exactly once, forming a closed or open initial link. Subsequently, the system enters the core iterative optimization phase, performing a topology adjustment through "side-swapping." In each iteration, based on preset swapping rules (e.g., 3-opt swapping rules), a certain number of old path segments are selected and removed from the current traversal path. For example, the connection between three different locations is broken, temporarily disconnecting the path into multiple isolated segments. Then, an equal number of new path segments are introduced to reconnect these broken points in different combinations, thereby constructing a traversal candidate path with a changed topology. Following this, the system executes the distance... In the computational operation, by identifying the connection relationships of all adjacent task point pairs in the candidate traversal path, the system retrieves the corresponding physical travel distance values ​​from the pre-constructed distance matrix one by one, and sums these values ​​to obtain the total physical travel cost of the candidate path. The system compares the total cost of the candidate path with the total cost of the initial traversal path (or the current optimal path) before optimization. If the total cost of the candidate path is lower, it indicates that the new connection order is more efficient, and the candidate path is used to replace the original traversal path as the new benchmark. The system repeats the above steps of disconnection, reconnection, calculation, and comparison, continuously trying various possible connection combinations in the solution space until the physical travel distance of any new candidate path generated by the edge-swapping operation is not lower than the previously retained path, that is, reaching a local optimum or satisfying the convergence condition. At this time, the system outputs the arrangement order of each task point in the last retained path as the final task access sequence to guide the UAV flight.

[0078] Specifically, in this embodiment of the invention, the follow-up inventory mode is as follows: receiving a follow-up command, which includes the main feature information of the moving entity to be followed, the moving entity being a warehouse worker or a mobile transportation device; the drone searches for and locks onto the moving entity, calculates the motion vector of the moving entity in real time, and generates a dynamic following trajectory that maintains a preset relative spatial position with the moving entity; based on the dynamic following trajectory, the drone scans the cargo identification information on the shelves on both sides of the flight path in real time, and maps the cargo identification information to the warehouse map in combination with the drone's current real-time position coordinates to complete the follow-up inventory.

[0079] In this embodiment of the invention, the drone first receives a follow-up command, which specifies the main feature information of the moving subject to be followed. For example, when the moving subject is a warehouse worker, the main feature information may be the specific color of the reflective vest worn by the worker or a specific visual label affixed to the back; when the moving subject is a forklift or AGV, the main feature information may be an ArUco code, QR code, or specific outline appearance data of the vehicle affixed to the roof or rear of the vehicle. The drone activates its onboard visual sensor to perform an omnidirectional search. When a target matching the main feature information is identified within the field of view, the drone immediately locks onto the target and enters a visual servo tracking state.In this state, the drone does not fly via pre-set GPS waypoints, but instead calculates in real time the displacement and dimensional change rate of the moving subject in the image coordinate system to determine the motion vector of the moving subject, including its speed and direction. Based on the calculated motion vector, the drone generates a dynamic following trajectory. The core constraint of this trajectory is to always maintain a preset relative spatial position with the moving subject. For example, to ensure operational safety and obtain the best observation angle, this relative spatial position can be set to two meters above and behind the moving subject, meaning the drone always hovers above the operator or forklift. It maintains a certain advantage in height and position; furthermore, this preset relative spatial position is not a single fixed value, but is configured differently within a preset range according to the type of mobile subject and its operational characteristics: when the mobile subject is a warehouse worker, considering the high mobility but small size of the personnel, and the need to avoid interference from the rotor airflow, the relative spatial position is preferably set within the range of 1.5 to 2.5 meters behind and 1.8 to 2.2 meters above. This distance ensures that the visual sensor can clearly capture the visual tag on the back of the personnel, while also ensuring that the drone is outside the personnel's safe psychological distance; when When the mobile transport vehicle is a forklift or AGV, considering its large size, high speed, and potential obstruction of view by stacked goods, the preferred relative spatial position is set within a range of 2.5 to 4.0 meters behind and 2.5 to 3.5 meters above. Increasing the rear distance provides a longer braking buffer space to handle sudden stops, while increasing the height provides a bird's-eye view over the forklift mast or stacked goods, preventing obstruction and ensuring unrestricted scanning of the side racks. If the mobile vehicle accelerates, turns, or stops, the UAV adjusts its position using a PID control algorithm. The drone monitors the drone's flight attitude, including pitch and yaw angles, and corrects its flight speed and heading in real time to eliminate positional deviations and ensure a constant relative distance between them. While flying along this dynamic tracking trajectory, the drone scans the goods on the shelves on both sides of the flight path in real time. Then, it directly maps the goods identification information and real-time position coordinates to the warehouse map data. This method allows the drone to complete the inventory count and position calibration of goods along the path by taking advantage of the fragmented time while following the movement of the staff, without the need for a specially planned inventory route, which greatly improves the overall efficiency of warehouse operations.

[0080] Specifically, in this embodiment of the invention, the step of generating an initial obstacle avoidance trajectory based on the global inventory path and warehouse environment information, combined with the UAV location information, includes: parsing the global inventory path, extracting subsequent target waypoints in the flight sequence from the global inventory path based on the UAV location information, and setting the subsequent target waypoints as the local planning endpoint; constructing a voxel grid map based on the warehouse environment information, identifying the obstacle occupancy status of each voxel unit in the voxel grid map; and generating an initial obstacle avoidance trajectory based on the obstacle occupancy status using a jump point search algorithm, starting from the UAV location information and ending at the local planning endpoint.

[0081] In this embodiment of the invention, after the UAV obtains the global inventory path at the macro level, it needs to be converted into a local flight command that can be executed at the micro level; first, the initial obstacle avoidance trajectory generation step based on real-time perception data is executed. Specifically, the system first parses the received global inventory path and discretizes it into a series of ordered spatial coordinate points, i.e., several UAV target waypoints. Then, the system reads the UAV's current positioning information in real time, such as local 3D coordinates. By calculating the distance relationship between the current position and each waypoint in the global path, the system selects the next waypoint that is closest to the current position along the flight direction and has not yet been reached from the sequence, and determines it as the current adjacent target waypoint or local planning endpoint. At the same time, using airborne sensors, such as LiDAR or depth cameras, the system collects warehouse environment information in real time to construct a local 3D voxel grid map. In this map, the actual physical space is cut into tiny cubic units (voxels). Preferably, the voxel size of the voxel grid map can be set to 15cm×15cm×15cm. The selection of this size parameter is mainly based on a comprehensive consideration of the accuracy of environmental modeling and the real-time computing load of the airborne system. On the one hand, from the perspective of environmental feature adaptation, the physical cross-sectional width of key structural components such as beams and columns of shelves in the warehouse environment is usually between 5cm and 10cm. Setting the voxel size to 15cm ensures that each voxel unit, while enveloping these small obstacles, naturally forms a safe expansion buffer with a physical size of approximately 5cm to 10cm. This granularity avoids misjudging narrow, passable gaps as blockages due to excessively large grids, and also avoids misidentifying sensor noise as physical obstacles due to excessively small grids. On the other hand, from the perspective of navigation planning efficiency, this size allows standard rack aisles to be discretized into a sufficient number of grids (approximately 8 to 13) in the horizontal dimension of the map. This provides sufficient resolution for the jump point search algorithm to plan a safe path to the center of the aisle, while keeping the amount of map data within the real-time processing capacity of the onboard edge computing main controller, thus preventing navigation delays. The system determines whether there are obstacles in each voxel unit based on sensor echo data. If there are obstacles, the voxel is marked as occupied; otherwise, it is marked as idle, thus forming a digital environment model. Based on this, the system uses the voxel where the UAV is currently located as the starting node and the determined adjacent target waypoints as the ending nodes. It calls a jump point search algorithm to search for a path in the voxel grid map. This algorithm moves quickly along a straight line and only generates a critical jump point when it encounters the edge of an obstacle. This allows it to quickly calculate a collision-free polyline path connecting the starting and ending points while avoiding all voxel units marked as occupied. Although this path may be dynamically rigid, it strictly guarantees geometric connectivity in complex obstacle environments and is defined as the initial obstacle avoidance trajectory.

[0082] More specifically, in this embodiment, the step length of a single jump point is set to 0.5 meters / step. The step length of the jump point is determined based on the width of the three-dimensional rack aisle (1.2-2 meters), which is adapted to the obstacle avoidance requirements of the narrow aisle of the rack. This can avoid collisions caused by excessive step length and trajectory redundancy caused by excessive step length.

[0083] Specifically, in a specific embodiment of the present invention, the step of optimizing the trajectory parameters of the initial obstacle avoidance trajectory includes: converting the initial obstacle avoidance trajectory into a B-spline curve, wherein the B-spline curve contains several control points for determining the shape of the B-spline curve; and constructing a multidimensional objective function based on the B-spline curve, wherein the multidimensional objective function is:

[0084] ;

[0085] in, , , as well as Indicates the weight of each item; The smoothness cost is represented by the sum of the squares of the acceleration and jerk at the control point; The obstacle distance cost is represented as: ; This indicates the distance between the control point and the nearest obstacle. This represents the preset penalty function; This represents the dynamic cost, used to indicate the portion exceeding the maximum speed limit of the drone. This indicates the total flight time of the drone.

[0086] In a specific embodiment of the present invention, after the system generates an initial obstacle avoidance trajectory with only geometric connectivity, in order to ensure the stability and safety of the UAV's flight in narrow rack aisles, a trajectory parameter optimization step based on dynamic constraints needs to be further performed. Specifically, the system first introduces parametric modeling technology to transform the initial obstacle avoidance trajectory composed of discrete waypoints into a mathematically continuous and differentiable uniform B-spline curve. This curve does not directly pass through each initial waypoint, but its geometry is defined by a series of control points distributed around the curve. Utilizing the convex hull property unique to B-spline curves, it ensures that the generated flight trajectory is always confined within the polygonal area enclosed by the control points, thus providing a geometric basis for subsequent safety constraints. Based on this, the system constructs a multidimensional objective function for evaluating trajectory quality. This function is not a single index, but is composed of a weighted sum of costs from four key dimensions: the first being the smoothness cost. The smoothness of the trajectory is measured by calculating the acceleration at the second derivative level and the jerk (i.e., the rate of change of acceleration) at the third derivative level for all control points, and then summing the squares of these physical quantities; a smaller value indicates a smoother flight. Secondly, obstacle distance cost is considered. The system uses the Euclidean symbolic distance field to query the physical distance from each point on the trajectory to the nearest obstacle in real time. To construct effective penalty rules, the system presets a safety threshold, typically set as the sum of the drone's fuselage radius and a reserved safety margin. Simultaneously, a penalty function is defined, designed as a piecewise non-linear growth function: when the distance value of the evaluation point is greater than the safety threshold, it indicates that the drone is in an absolutely safe zone; at this time, the obstacle distance cost is zero, and the algorithm does not impose any penalty, allowing the trajectory to flow freely. When the distance value of the evaluation point is less than or equal to the safety threshold, it indicates that the drone has intruded into the safety buffer zone, at which point the penalty mechanism is activated. Finally, the system sums the penalty values ​​calculated from all sampling points on the entire trajectory to obtain the final total obstacle distance cost. This total cost serves as one of the loss function terms in the optimizer iteration, forcing the trajectory to actively move away from obstacles. Finally, there is the dynamic cost. The system detects the velocity vector at any point on the trajectory. If the calculated velocity exceeds the physical limit of the drone's motors or the preset maximum cruising speed, a high penalty value is generated; finally, there is the time cost. The system calculates the total time required to fly along the trajectory to encourage the algorithm to find more efficient paths while ensuring safety and stability.

[0087] It should be noted that, considering the narrow and long straight operating environment within the automated storage and retrieval system (AS / RS) aisles, the weighting of various parameters follows a principle of prioritizing safety while also considering smooth dynamic configuration. The specific weighting strategy is as follows: Due to the limited space in the storage aisles and the extremely small physical distance between the drone and the racks, the weight corresponding to obstacle distance cost is given the highest priority to ensure absolute flight safety. Secondly, to ensure that the onboard camera can clearly capture cargo markings during flight and avoid image blurring due to excessive abrupt changes, the weight corresponding to smoothness cost is given the second highest priority. Dynamics cost and time cost, as constraint and efficiency terms, have relatively low weights. As a set of experimentally verified optimal empirical values, the weights can be set as follows: smoothness weight is set to 0.4 to primarily suppress trajectory abruptness; obstacle distance weight is set to 1.0, and can be designed as a dynamic variable that increases exponentially with decreasing distance, thereby generating strong repulsive potential energy; dynamics weight is set to 0.2 as a soft constraint to limit speed; and time weight is set to 0.1 to appropriately optimize operational efficiency while ensuring safety.

[0088] In the solution phase, the system preferably employs a nonlinear numerical optimization algorithm, such as the L-BFGS algorithm, to iteratively solve the aforementioned multidimensional objective function. By continuously fine-tuning the spatial coordinates of each control point in the B-spline curve, it seeks the combination of control points that minimizes the total agency value, ultimately generating an obstacle avoidance trajectory that is collision-free, meets dynamic smoothness requirements, and has optimal flight efficiency. Specifically, in a specific embodiment of the present invention, the directional inventory mode is as follows:

[0089] Receive and respond to the goods selection instruction initiated by the user, and identify the specified goods according to the goods selection instruction;

[0090] Based on the analysis of warehouse map data, the target location coordinates corresponding to the specified goods are obtained, and the target location coordinates are sent to the drone;

[0091] The drone receives the target location coordinates, generates a directional trajectory, and flies to the designated cargo according to the directional trajectory, obtains cargo identification information, and triggers the indication device to execute physical instructions.

[0092] In this embodiment of the invention, in addition to a global inventory check, a targeted inventory mode is also provided to meet the needs of rapid verification or temporary spot checks of specific goods. During the execution of this mode, the ground station first enters a command listening state. When the operator selects a specific shelf location on the electronic map of the human-machine interface, or directly enters the unique number of a specific item and initiates a query request, the ground station immediately responds to the item selection command and retrieves the warehouse map data stored internally. By matching the item number or shelf ID, it parses the precise target location coordinates of the specified item in the three-dimensional space of the warehouse, including longitude, latitude, and altitude information, or the X, Y, and Z axis coordinates relative to the local coordinate system. Subsequently, the ground station sends the target location coordinates as a single-point task command to the drone via a wireless communication link. After receiving the target location coordinates, the drone no longer relies on a pre-planned global long path, but... Starting from the current hovering position and ending at the received target position coordinates, the onboard path search algorithm (such as the A* search algorithm) is invoked to calculate a point-to-point geometric path directly to the target in the current obstacle environment, generating a directional trajectory. The drone then rapidly flies along this directional trajectory to the front of the designated cargo and hovers. The onboard camera collects the marking information on the cargo surface for identification and verification. Once the target cargo is confirmed to be correct, the drone immediately triggers the onboard indicating device to perform a physical indicating action, such as controlling a relay to close and illuminating a high-brightness laser pointer, projecting a visible beam of light onto the surface of the designated cargo for a preset duration (1 second in this embodiment). This visually marks the target cargo in physical space, assisting ground personnel in manual verification or confirmation. Testing verified that the directional inventory mode was triggered several times during global inventory checks, and the switching response time and interruption rate were recorded. Test results show that the mode switching time of this invention is less than 0.5 seconds and the interruption rate is less than 0.1%, while the traditional method has a switching time greater than 1 second and an interruption rate greater than 0.5%. Therefore, this invention can effectively reduce the interruption rate and switching time.

[0093] Specifically, in embodiments of the present invention, a scheduling algorithm for switching the targeted inventory mode is further included, specifically:

[0094] The system monitors input signals on the human-computer interaction interface in real time. When it detects that the user has executed a selection command for a specific item in the warehouse map data, it generates a task preemption command containing the three-dimensional coordinates of the specified item.

[0095] The drone receives the task preemption instruction via a wireless communication link and parses the command words in the instruction to identify the current task type as a directional inventory task.

[0096] In response to the directional inventory task, the drone immediately suspends the currently executing global traversal process or exits the standby hovering state and clears the current flight waypoint cache queue.

[0097] The drone loads the three-dimensional coordinates of the specified cargo, identifies the target navigation point, calculates the navigation path based on the target navigation point, and initiates the directional inventory mode.

[0098] To achieve efficient emergency inventory checks of specific goods, the system employs a priority-based task scheduling algorithm to logically switch between fully automated traversal operations and manually designated directional operations. Specifically, the ground station software continuously runs a real-time monitoring program in the background, constantly monitoring input events on the human-machine interface. When it detects that an operator has executed a selection command on a specific shelf or goods icon on the visualized warehouse map data interface (e.g., double-clicking the shelf icon or entering a goods code for confirmation), the system immediately responds and extracts the absolute three-dimensional coordinates of the specified goods from the map database. It then generates a task preemption command with the highest execution level. The data frame of this command contains specific command words (e.g., defined as a fixed-point inspection command word) and target coordinate data. Furthermore, to ensure the reliability of command transmission and the accuracy of parsing, the data frame of this task preemption command adopts a strictly defined communication protocol format. Specifically, the data frame consists of a frame header, priority identifier, function command word, data payload, checksum, and frame trailer. The frame header identifies the start of the data packet, for example, 0xEB90, enabling the UAV communication module to synchronize the frame structure from a continuous bit stream. The priority field is set to the highest preset level, for example, 0xFF, to instruct the receiver's underlying driver to trigger a high-priority hardware interrupt or task preemption logic. The function command word is encoded as a dedicated code for directional inventory, for example, 0xA5, to uniquely identify the current task type. The data payload occupies 12 bytes and stores the X, Y, and Z axis three-dimensional coordinate data of the target cargo in single-precision floating-point format. The checksum uses CRC16 cyclic redundancy check to detect bit errors during wireless transmission and ensure the integrity of the coordinate data. After receiving the data frame via the wireless communication link, the UAV immediately unpacks and parses the command. Once it identifies the command as a directional inventory task, the system will trigger a hardware interrupt or a high-priority software callback to execute the task preemption logic: if the UAV is currently in a standby hovering state, it will immediately wake up and enter the operation preparation state; if the UAV is currently executing a global inventory path, i.e., in traversal inventory mode, it will immediately suspend the current traversal process, forcibly interrupt the ongoing route task, and perform a critical cache clearing operation, clearing all pending global flight waypoint cache queues in the flight controller to prevent old waypoint data from interfering with the new task, thereby releasing flight control; subsequently, the UAV loads the parsed three-dimensional coordinates of the specified cargo into the navigation system's register, locks it as the current unique navigation target point, and uses this as the endpoint and the current hovering position as the starting point to call the path search algorithm to recalculate a collision-free navigation path, officially starting the control loop of the directional inventory mode, ensuring that the UAV can immediately respond to human requests and fly directly to the target.

[0099] This invention also provides a three-dimensional shelf drone inventory system, such as Figure 7 As shown, the system includes:

[0100] The global path planning module collects warehouse map data and inventory task points from the ground station, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans the global inventory path based on the warehouse topology map.

[0101] The flight path planning module receives the global inventory path and collects warehouse environment information. Based on the global inventory path and warehouse environment information, and combined with the drone's position information, it generates an initial obstacle avoidance trajectory. The warehouse environment information represents the perception data of the spatial distribution of obstacles in the drone's current local airspace.

[0102] The route optimization module optimizes the trajectory parameters of the initial obstacle avoidance trajectory and generates the obstacle avoidance trajectory to be executed.

[0103] The dual-verification module identifies cargo identification information based on the obstacle avoidance trajectory executed by the drone, triggers the indicator device to physically indicate the cargo, and outputs cargo information.

[0104] More specifically, in this embodiment, since the drone relies primarily on optical flow or SLAM algorithms for hovering in indoor environments without GPS, and is constrained by the complex airflow within the warehouse—such as the ventilation system or the near-wall effect of its own propellers—the fuselage inevitably experiences slight high-frequency shaking or low-frequency drift. Because the laser designator is typically rigidly fixed to the fuselage, any slight yaw (e.g., a 0.5-degree deflection) projected onto a shelf several meters away will cause a visible displacement or jitter in the laser spot. This makes it difficult for ground personnel to identify which item the laser is referring to, severely impacting the accuracy of physical indication and user experience.

[0105] To address this issue, during the activation of the indicator device, an airborne vision sensor is used to acquire feedback images containing the laser spot and the target cargo markings in real time. Within the image pixel coordinate system of the feedback image, image processing algorithms are used to extract the centroid coordinates of the laser spot and the geometric center coordinates of the target cargo markings, and to calculate the pixel deviation vector pointing from the centroid coordinates to the geometric center coordinates. This pixel deviation vector is then input as an error feedback signal to the attitude control loop of the airborne mission computer. A PID control algorithm is used to calculate the fuselage yaw and pitch angle compensation amounts used to compensate for the pixel deviation vector. These compensation amounts are then converted into flight attitude fine-tuning commands and sent to the flight controller, driving the UAV to adjust its hovering attitude in real time, thus forming a closed-loop negative feedback control that minimizes the pixel deviation vector.

[0106] Thanks to the introduction of a visual feedback-based closed-loop compensation mechanism, the system can resist pointing deviations caused by airflow disturbances or hovering errors in real time. Even if the drone body experiences unavoidable physical vibrations in the air, the control algorithm can make millisecond-level attitude fine adjustments so that the projected laser spot appears to be perfectly stationary in the center of the target cargo. This method overcomes the problem of ambiguous indication caused by spot jitter under traditional open-loop control from the perspective of control methods, significantly improving the accuracy and readability of high-level inventory counting on automated racking systems.

[0107] More specifically, in this embodiment, although the drone's flight along the obstacle avoidance trajectory optimized by the B-spline ensures smoothness of dynamics and flight safety, this prioritization of flight continuity and obstacle avoidance safety leads to a physical deviation between the drone's actual flight trajectory and the ideal optimal shooting point. Specifically: on the one hand, to meet safety obstacle avoidance constraints, obstacle avoidance algorithms (such as jump point search algorithms) typically plan paths located at the geometric center of the alleyway space or in safe corridors away from obstacles. This results in the physical shooting distance between the drone's onboard camera and the shelf cargo label often being greater than the ideal clear imaging distance, leading to insufficient effective pixel proportions of the cargo label in the image and difficulty in conveying sufficient texture details. On the other hand, because the B-spline curve smooths the broken-line path, the drone is in a continuous dynamic cruise state when performing inventory tasks. This means the drone sweeps across the shelves in continuous motion, rather than flying to a point and stopping to take a picture. This causes the cargo label to frequently slide quickly past the edge of the camera's field of view. Due to the inherent optical distortion characteristics of wide-angle lenses, the image clarity and geometric reproduction at the edge of the field of view are significantly lower than in the central area.

[0108] To address this issue, an adaptive acquisition control strategy based on imaging geometric validity is adopted in the cargo identification information acquisition step. Specifically, a central effective recognition area and a minimum pixel percentage threshold are preset in the image acquisition logic of the airborne vision sensor. In this embodiment, the central effective recognition area can be set to 60% of the image center, and the minimum pixel percentage threshold can be set to 10%. When the vision sensor detects a cargo identification target in the field of view, it does not immediately perform decoding and reading operations, but instead calculates the geometric center coordinates and pixel coverage area of ​​the target in the image coordinate system in real time. The system performs dual verification based on the above geometric data: determining whether the geometric center coordinates fall within the central effective recognition area and whether the pixel coverage area is greater than the minimum pixel percentage threshold. If the position conditions are not met, a translation control command is generated to drive the UAV to adjust its position to bring the target back to the center of the image. If the size conditions are not met, a proximity control command is generated to drive the UAV to move forward along the line of sight to increase the imaging area. The system only locks the current frame to perform decoding and data storage operations when the cargo identification target simultaneously meets the conditions of being located in the central area and meeting the size requirements.

[0109] By introducing this geometrically based active filtering mechanism, the system is essentially equipped with an automatic image composition decision logic for the drone, which can forcibly eliminate interference from lens edge distortion and long-distance blur from the image acquisition source. This method can overcome the problems of low decoding rate and wasted computing power caused by inconsistent imaging quality in complex environments from the data acquisition decision level, and significantly improve the success rate of a single inventory and the standardization of data with extremely low computing cost.

[0110] More specifically, in this embodiment, when the drone is in the follow-up inventory mode, although the drone can maintain its relative position with the moving subject through visual servoing, in actual complex warehousing operation scenarios, the moving subject itself is very likely to become a source of visual obstruction. For example, when a forklift is loading large-volume goods or lifting its forks, its large physical outline often blocks the drone's side-facing camera's field of view, causing the drone to be unable to see the shelf area on the other side or behind the moving subject. If the system rigidly maintains the preset fixed following angle at this time, the drone's perception line of sight will be blocked by the moving subject for a long time, forming an inventory blind spot. This will cause a large number of goods markings along the path to be missed, resulting in a serious lack of follow-up inventory data, which greatly reduces the actual efficiency of human-machine collaboration.

[0111] To address this issue, an adaptive adjustment strategy for occlusion by a moving subject is adopted during the follow-up inventory process. Specifically, the UAV acquires a lateral field-of-view image, identifies the shelf area and non-shelf area in the image, and calculates the effective visible proportion of the shelf area in the lateral field-of-view image. The effective visible proportion is compared with a preset occlusion threshold. When the effective visible proportion is lower than the occlusion threshold, it is determined that there is a moving subject occlusion. The UAV maintains a constant longitudinal following speed along the direction of movement of the moving subject and generates a position compensation vector. Based on the position compensation vector, the UAV is driven to perform lateral movement or altitude increase relative to the moving subject until the effective visible proportion is detected to recover to above the occlusion threshold.

[0112] By introducing this dynamic position compensation mechanism based on visual proportion perception, the system empowers the UAV with the intelligent decision-making ability to actively seek the best observation window during the following process, effectively solving the problem of blind spots caused by equipment or personnel obstruction in human-machine mixed operations. After testing and verification, this method can control the missed scan rate of the follow-up inventory to below 0.2%, which is about 80% lower than the missed scan rate of the conventional fixed position following scheme, and significantly improves the robustness of the system and the integrity of data acquisition in dynamic and complex environments.

[0113] As described above, the UAV inventory method for automated storage and retrieval systems provided by this invention can significantly improve the traversal efficiency, obstacle avoidance safety, and cargo positioning accuracy of automated storage and retrieval systems, overcoming the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance, and weak interactive feedback. The UAV inventory method for automated storage and retrieval systems includes: S1. A ground station collects warehouse map data and inventory task points, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map; S2. The UAV receives the global inventory path and collects warehouse environment information, generates an initial obstacle avoidance trajectory based on the global inventory path and the warehouse environment information, combined with the UAV's position information, where the warehouse environment information represents the perceived data of the spatial distribution of obstacles in the current local airspace of the UAV; S3. The trajectory parameters of the initial obstacle avoidance trajectory are optimized to generate an execution obstacle avoidance trajectory; S4. The UAV collects cargo identification information based on the execution obstacle avoidance trajectory, triggers an indicator device to physically indicate the cargo, and outputs cargo information. The ground station utilizes collected warehouse map data and inventory task points to map the unstructured physical warehouse space into a warehouse topology map composed of spatial nodes and guiding path segments. Based on the topological connection relationship, the access order of all task points is forcibly locked, which can significantly improve the traversal efficiency of automated racking inventory. Then, the UAV combines warehouse environment information to generate an initial obstacle avoidance trajectory under global path guidance, and performs trajectory parameter optimization on the initial trajectory to generate an execution obstacle avoidance trajectory. A feasible channel that meets the geometric collision-free conditions is constructed through the initial trajectory, and the spatial position avoidance logic is established. Subsequently, through parameter optimization, the polyline path in the channel is processed to limit the UAV's acceleration abrupt changes and attitude oscillations during avoidance actions, thereby improving the UAV's obstacle avoidance safety. Based on the execution obstacle avoidance trajectory, the indicator device is triggered to physically indicate the goods while collecting information, effectively overcoming the problems of weak interactive feedback and difficulty in confirming the location of goods caused by traditional operations. Compared with existing technologies, this solution can significantly improve the traversal efficiency, obstacle avoidance safety and goods positioning accuracy of automated racking inventory, and overcome the technical shortcomings of traditional solutions such as path redundancy, rigid obstacle avoidance and weak interactive feedback.

[0114] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for inventorying automated shelving systems using drones, characterized in that, include: The S1 ground station collects warehouse map data and inventory task points, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map. The S2 drone receives the global inventory path and collects warehouse environment information. Based on the global inventory path and the warehouse environment information, and combined with the drone's position information, it generates an initial obstacle avoidance trajectory. The warehouse environment information represents the perception data of the spatial distribution of obstacles in the drone's current local airspace. S3 optimizes the trajectory parameters of the initial obstacle avoidance trajectory to generate an execution obstacle avoidance trajectory; The UAV described in S4 collects cargo identification information based on the obstacle avoidance trajectory and triggers the indicator device to physically indicate the cargo, outputting cargo information; The triggering indicator device provides physical indication of the goods, specifically including: During the activation of the indicator device, an airborne vision sensor is used to acquire feedback images in real time, including laser spot and target cargo markings; In the image pixel coordinate system of the feedback image, the centroid coordinates of the laser spot and the geometric center coordinates of the target cargo mark are extracted by the image processing algorithm, and the pixel deviation vector from the centroid coordinates of the laser spot to the geometric center coordinates is calculated. The pixel deviation vector is input as an error feedback signal to the attitude control loop of the airborne mission computer; The fuselage yaw angle compensation and pitch angle compensation are calculated based on the PID control algorithm to compensate for the pixel deviation vector. The fuselage yaw angle compensation and the pitch angle compensation are converted into flight attitude fine-tuning commands and sent to the flight controller to drive the UAV to adjust its hovering attitude in real time.

2. The method for inventorying automated shelving systems using drones according to claim 1, characterized in that, The steps involve planning a global inventory route based on the warehouse topology map, including: Construct a warehouse topology map, which includes: spatial nodes representing path intersections and inventory task points, and guide path segments representing the passability between adjacent spatial nodes; Extract all inventory task points, calculate the shortest travel distance between any two inventory task points based on the warehouse topology map, and construct a distance matrix based on several of the shortest travel distances. The distance matrix is ​​used to quantify the travel cost of the travel distance. The LKH-3 algorithm is used to iteratively search the distance matrix to generate a task access sequence. The corresponding guiding path segments are then connected according to the task access sequence to generate a global inventory path.

3. The method for inventorying automated storage and retrieval systems using drones according to claim 2, characterized in that, The steps include constructing a distance matrix based on several of the shortest travel distances, including: Construct an N×N matrix, where N represents the total number of the inventory task points; Iterate through the inventory task points, for any two distinct inventory task points... i and j， The inventory task points were searched based on Dijkstra's algorithm. i and j Find the shortest path and calculate the inventory task point. i and j The physical travel distance; Assign the physical travel distance to the first element in the matrix. i Line number j Use the elements of the columns to construct a distance matrix.

4. The method for inventorying automated storage and retrieval systems using drones according to claim 3, characterized in that, The steps involve iteratively searching the distance matrix using the LKH-3 algorithm to generate a task access sequence, including: Construct an initial traversal path, which includes all the inventory task points; Side-swapping operation: Remove the preset old path segments of the initial traversal path, introduce new path segments with the same number of preset old path segments, reconnect the path breakpoints, and construct traversal candidate paths; Distance calculation operation: Identify the adjacent connection relationships of the traversal candidate paths, and calculate the physical travel distance of the traversal candidate paths based on the distance matrix; If the physical travel distance of the candidate path is less than that of the initial traversal path, the initial traversal path is replaced, and the edge-swapping operation and distance calculation operation are repeated until the physical travel distance of any candidate path generated by the edge-swapping operation is not less than that of the previous candidate path. The current candidate path is then output as the task access sequence.

5. The method for inventorying automated shelving systems using drones according to claim 1, characterized in that, This also includes a dynamic inventory management mode, specifically: Receive a follow-up command, which contains the main feature information of the moving subject to be followed, wherein the moving subject is a warehouse worker or a mobile transportation device; The drone searches for and locks onto the moving subject, calculates the motion vector of the moving subject in real time, and generates a dynamic following trajectory that maintains a preset relative spatial position with the moving subject; Based on the dynamic following trajectory, the drone scans the cargo identification information on the shelves on both sides of the flight path in real time, and maps the cargo identification information to the warehouse map in combination with the drone's current real-time position coordinates to complete the follow-up inventory.

6. The method for inventorying automated storage and retrieval systems using drones according to claim 1, characterized in that, The steps include generating an initial obstacle avoidance trajectory based on the global inventory path, the warehouse environment information, and the drone's location information, including: The global inventory path is parsed, and subsequent target waypoints in the flight sequence are extracted from the global inventory path based on the UAV's location information. The subsequent target waypoints are then set as the local planning endpoints. Based on the warehouse environment information, a voxel grid map is constructed, and the obstacle occupancy status of each voxel unit in the voxel grid map is identified. Based on the obstacle occupancy status, a jump point search algorithm is used to generate an initial obstacle avoidance trajectory, starting from the UAV's location information and ending at the local planning endpoint.

7. The method for inventorying automated shelving systems using drones according to claim 6, characterized in that, The step of optimizing the trajectory parameters of the initial obstacle avoidance trajectory includes: The initial obstacle avoidance trajectory is converted into a B-spline curve, which contains several control points for determining the shape of the B-spline curve. A multidimensional objective function is constructed based on the B-spline curve. The multidimensional objective function is as follows: ； in, , , as well as Indicates the weight of each item; The smoothness cost is represented by the sum of the squares of the acceleration and jerk of the control point; The distance cost to the obstacle is represented as: ; This indicates the distance between the control point and the nearest obstacle. This represents the preset penalty function; This represents the dynamic cost, used to indicate the portion exceeding the maximum speed limit of the drone. This indicates the total flight time of the drone.

8. The method for inventorying automated shelving systems using drones according to claim 1, characterized in that, This also includes a targeted inventory mode, specifically: Receive and respond to the goods selection instruction initiated by the user, and identify the specified goods according to the goods selection instruction; Based on the warehouse map data parsing, the target location coordinates corresponding to the specified goods are obtained, and the target location coordinates are sent to the drone; The drone receives the target location coordinates, generates a directional trajectory, and flies to the designated cargo according to the directional trajectory, obtains cargo identification information, and triggers the indication device to execute physical instructions.

9. The method for inventorying automated shelving systems using drones according to claim 8, characterized in that, It also includes a scheduling algorithm for switching the targeted inventory mode, specifically: The system monitors input signals on the human-computer interaction interface in real time. When it detects that a user has executed a selection command for a specific item in the warehouse map data, it generates a task preemption command containing the three-dimensional coordinates of the specified item. The UAV receives the task preemption command via a wireless communication link and parses the command words in the command to identify the current task type as a targeted inventory task. In response to the directional inventory task, the drone immediately suspends the currently executing global traversal process or exits the standby hovering state and clears the current flight waypoint cache queue. The drone loads the three-dimensional coordinates of the designated cargo, identifies the target navigation point, calculates the navigation path based on the target navigation point, and initiates the directional inventory mode.

10. A three-dimensional shelving drone inventory system, characterized in that, The system includes the automated shelving drone inventory method according to any one of claims 1 to 9, wherein the system comprises: The global path planning module collects warehouse map data and inventory task points from the ground station, constructs a warehouse topology map based on the warehouse map data and inventory task points, and plans a global inventory path based on the warehouse topology map. The flight path planning module allows the UAV to receive the global inventory path and collect warehouse environment information. Based on the global inventory path and the warehouse environment information, and combined with the UAV's position information, it generates an initial obstacle avoidance trajectory. The warehouse environment information represents the perception data of the spatial distribution of obstacles in the current local airspace of the UAV. The route optimization module optimizes the trajectory parameters of the initial obstacle avoidance trajectory to generate an execution obstacle avoidance trajectory; The dual-verification module identifies the UAV by collecting cargo identification information based on the obstacle avoidance trajectory and triggering an indicator device to physically indicate the cargo and output cargo information. The triggering indicator device provides physical indication of the goods, specifically including: During the activation of the indicator device, an airborne vision sensor is used to acquire feedback images in real time, including laser spot and target cargo markings; In the image pixel coordinate system of the feedback image, the centroid coordinates of the laser spot and the geometric center coordinates of the target cargo mark are extracted by the image processing algorithm, and the pixel deviation vector from the centroid coordinates of the laser spot to the geometric center coordinates is calculated. The pixel deviation vector is input as an error feedback signal to the attitude control loop of the airborne mission computer; The fuselage yaw angle compensation and pitch angle compensation are calculated based on the PID control algorithm to compensate for the pixel deviation vector. The fuselage yaw angle compensation and the pitch angle compensation are converted into flight attitude fine-tuning commands and sent to the flight controller to drive the UAV to adjust its hovering attitude in real time.

Images