A path optimization method for intelligent inventory robots based on environment modeling and collaborative optimization

Abstract

The present invention provides an intelligent inventory robot path optimization method based on environmental modeling and collaborative optimization, which relates to the technical field of intelligent warehouse management. The method comprises the following steps: scanning a warehouse area through an unmanned aerial vehicle platform and generating three-dimensional warehouse point cloud data; generating a two-dimensional warehouse plan after point cloud denoising, plane fitting, shelf boundary extraction, and channel detection processing, and transmitting the result to a robot data base; using an A* algorithm to generate an initial inventory planning path, and optimizing the initial path through a simulated annealing algorithm to generate a target inventory planning path; correcting the target path through a hybrid algorithm of multi-objective optimization and PSO dynamic weights based on dynamic obstacle influencing factors, combined with RFID signal strength and the remaining battery power of the robot; and viewing the inventory progress through an AR device at the operator's end and manually adjusting the target path. This allows the intelligent inventory robot's inventory path to be optimized in real time in a dynamic environment, thereby improving the robot's inventory efficiency, coverage, and endurance.

Description

Technical Field

[0001] The present invention relates to the field of intelligent warehouse management technology, and in particular to an intelligent inventory robot path optimization method based on environment modeling and collaborative optimization. Background Art

[0002] With the rapid development of modern warehouse management, intelligent inventory robots, with their high efficiency and precision, are increasingly being used, becoming a vital tool for improving warehouse efficiency and reducing labor costs. These robots significantly enhance the intelligence of warehouse management through automated inspections, data collection, and processing.

[0003] However, as warehouse environments become increasingly complex and dynamic, traditional path planning methods for inventory robots rely on pre-built static environmental models. These models often overlook the dynamic changes that can occur in the warehouse environment. In practice, warehouse environments are dynamic and can include obstacles such as temporarily stacked goods, moving forklifts, and personnel, all of which can significantly impact the robot's path planning. Furthermore, variations in RFID signal strength can affect the robot's ability to accurately identify the location of goods, and the robot's real-time power consumption is directly related to its operational continuity and efficiency.

[0004] Faced with these dynamic factors, traditional path planning methods often fail to quickly adjust to real-time environmental information. This can lead to robots being unable to avoid unexpected situations in a timely manner, or even getting stuck in an endless loop or failing to complete their tasks. This not only reduces inventory efficiency but also poses a potential threat to warehouse safety.

[0005] Therefore, it is necessary to provide an intelligent inventory robot path optimization method based on environmental modeling and collaborative optimization to solve the above technical problems. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides an intelligent inventory robot path optimization method based on environmental modeling and collaborative optimization, which is used to solve the problem that traditional inventory robot path planning methods are often based on static environmental models, which are difficult to adapt to dynamic changing factors in the warehouse environment and cannot achieve efficient path planning and collaborative optimization of intelligent inventory robots.

[0007] The present invention provides a path optimization method for an intelligent inventory robot based on environment modeling and collaborative optimization, the path optimization method comprising:

[0008] Scan the target warehouse area through the drone platform and generate 3D warehouse point cloud data, perform point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the 3D warehouse point cloud data to generate a 2D warehouse plan and transmit it to the robot data base;

[0009] An A* algorithm is used to generate an initial inventory planning path based on the two-dimensional warehouse plan, and the initial inventory planning path is optimized by a simulated annealing algorithm to generate a target inventory planning path;

[0010] Based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights;

[0011] The operator uses the AR device to view the robot's inventory progress and manually adjusts the target inventory planning path.

[0012] Preferably, the method of scanning the target storage area by the drone platform and generating three-dimensional storage point cloud data, performing point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the three-dimensional storage point cloud data to generate a two-dimensional storage plan and transmitting the two-dimensional storage plan to the robot data base specifically includes:

[0013] The drone platform is equipped with a 16-line laser radar and an IMU inertial navigation system, which performs overlapping scans along a preset grid route of the target storage area at a preset cruising speed to generate raw storage point cloud data;

[0014] The voxel grid filtering technology is used to divide the three-dimensional space corresponding to the original warehouse point cloud data into 0.08m 3 The voxels of the cube, and by the formula Performing voxel indexing on each of the original stored point cloud data and retaining the centroid point within each of the voxels to generate the three-dimensional stored point cloud data;

[0015] Among them, Voxel(p) represents the calculation point p(p x ,p y ,p z ) of voxels; v x 、v y 、v z Represents the side length of the voxel in the x, y, and z directions respectively; They represent the voxel indices to which the calculation point p belongs in the x, y, and z directions, respectively; W represents the total width of the voxel grid in the x direction, that is, the number of voxels in the x direction; H represents the total width of the voxel grid in the y direction, that is, the number of voxels in the y direction.

[0016] Preferably, the process of shelf boundary extraction and channel detection is as follows:

[0017] In the three-dimensional coordinate system corresponding to the three-dimensional warehouse point cloud data, for each plane, a local coordinate system orthogonal to the plane normal vector is constructed with the plane centroid as the origin, and the three-dimensional warehouse point cloud data is projected into the local coordinate system;

[0018] Perform Delaunay triangulation on the projected point cloud, generate a closed shelf boundary polygon by screening the triangle edges with the minimum circumscribed circle radius, and perform boundary error control on the closed shelf boundary polygon using the Douglas-Peucker algorithm to generate the shelf boundary contour;

[0019] The ground point cloud is converted into a bird's-eye view grid, and the pixel gaps in the three-dimensional warehouse point cloud data are filled by morphological closing operation, and the pixel occupancy probability P is calculated. occ ;

[0020] Get the shelf channel distinction probability and the preset width limit, the pixel occupancy probability P occ All pixels with a probability greater than the shelf channel distinction probability constitute shelf elements, and the pixel occupancy probability P occ All pixel points with a probability less than the shelf channel distinction probability constitute channel elements, and entrance and exit elements are marked at the same time, and the width between the pixel points corresponding to the two ends of the entrance and exit elements is greater than the preset width limit;

[0021] The shelf ID, the aisle width and the entrance and exit direction are respectively added to the shelf element, the aisle element and the entrance and exit element.

[0022] Preferably, the method of using the A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan, and optimizing the initial inventory planning path by a simulated annealing algorithm to generate a target inventory planning path, specifically includes:

[0023] Determine the inventory shelf location based on the two-dimensional warehouse plan, obtain the cargo asset type and inventory location priority corresponding to the inventory shelf location from the warehouse management library corresponding to the target storage area, and set the robot inventory motion parameters, namely the maximum walking speed, minimum safety distance, and single-shelf single-layer scanning time;

[0024] The A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated by combining the type of the goods asset and the priority of the inventory location;

[0025] The initial inventory planning path is locally optimized, that is, the initial inventory planning path is optimized by the simulated annealing algorithm to generate the target inventory planning path.

[0026] Preferably, the A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated in combination with the type of the goods asset and the priority of the inventory location, specifically including:

[0027] Set the node cost function as follows:

[0028] f(n)=g(n)+h(n)

[0029]

[0030] h(n)=|n x -t x |-|n y -t y |

[0031] Where f(n) represents the total estimated cost of inventory shelf location node n; g(n) represents the actual movement cost from the starting inventory location node to the inventory shelf location node n; Indicates the node from the starting inventory location to the previous inventory shelf location node The actual moving cost; move_cost(n) represents the cost of moving from the previous inventory shelf location node to node The basic moving cost to the inventory shelf location node n is If the node n moves horizontally or vertically to the inventory shelf location node n, then move_cost(n) = 1. If the node parent(n) moves diagonally to the inventory shelf location node n, then turn_penalty(n) represents the turning penalty of inventory rack location node n. If a turn is required from the previous inventory rack location node parent(n) to the inventory rack location node n, then turn_penalty(n) = γ, where γ represents the turning penalty coefficient. Otherwise, turn_penalty(n) = 0. width_penalty(n) represents the aisle width penalty of inventory rack location node n. β represents the width penalty coefficient, w0 represents the standard channel width, w(n) represents the channel width of the inventory shelf location node n; h(n) represents the heuristic estimated cost from the inventory shelf location node n to the end inventory location node; (n x ,n y ) represents the coordinates of the inventory shelf location node n; (t x ,t y ) represents the coordinates of the node where the inventory is terminated;

[0032] The inventory shelf location node with the smallest total estimated cost is expanded and selected through a priority queue until all the inventory shelf location nodes are traversed, thereby generating a continuous walking sequence including turning points and straight path segments, that is, the shortest path from the starting inventory location to the ending shelf location, and combining the cargo asset type and the inventory location priority to generate the initial inventory planning path.

[0033] Preferably, the locally optimizing the initial inventory planning path, i.e., optimizing the initial inventory planning path by the simulated annealing algorithm to generate the target inventory planning path, specifically includes:

[0034] Set the key parameters of the simulated annealing algorithm, namely the initial temperature T, cooling rate α, and termination temperature T ' and the number of disturbances Q at each temperature;

[0035] Defining a path perturbation operation includes randomly selecting two path nodes in the initial inventory planning path P, deleting the intermediate node between the two path nodes and reconnecting the initial inventory planning path P, randomly swapping the order of two non-adjacent path nodes in the initial inventory planning path P, and randomly selecting a subpath in the initial inventory planning path P and reversing the direction of the subpath;

[0036] The path perturbation operation is performed on the initial inventory planning path P to generate a new inventory planning path P ' , calculate the path length change ΔL = LL ' , where L represents the initial path length of the initial inventory planning path P, L ' Represents the new inventory planning path P ' The new path length;

[0037] If ΔL<0, then P1=P ' ; If ΔL ≥ 0, the path acceptance probability is calculated based on the Metropolis acceptance criterion If the path acceptance probability q is greater than the random number r, 0≤r<1, then P1=P ' , otherwise P1=P;

[0038] The initial temperature T gradually converges according to the cooling rate α, and the path perturbation operation is performed Q times at each temperature until the temperature is less than the termination temperature T ' , terminate the simulated annealing algorithm and output the target inventory planning path.

[0039] Preferably, the target inventory planning path is corrected by a hybrid algorithm of multi-objective optimization and PSO dynamic weight based on the collected dynamic obstacle impact factors, combined with the RFID signal strength and the remaining battery power of the robot, specifically including:

[0040] The RFID signal strength and the remaining battery power of the robot are normalized, and based on the target path length of the target inventory planning path, the normalized RFID signal strength and the remaining battery power of the robot, and the dynamic obstacle impact factor, a multi-objective optimization function is constructed as follows:

[0041]

[0042] Where F represents the comprehensive cost function; w1, w2, w3, w4 represent the target weight coefficients; L goal represents the target path length of the target inventory planning path; S norm represents the normalized RFID signal strength; B norm represents the remaining power of the robot after normalization; O represents the dynamic obstacle influence factor, d obs represents the distance from the intelligent inventory robot to the nearest obstacle, d safe represents the safety distance threshold of the intelligent inventory counting robot, and λ represents the dynamic obstacle influence coefficient;

[0043] Each particle represents a set of target weight vectors (w1, w2, w3, w4), and the particle speed and position are updated through PSO iteration to determine the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w ' 4);

[0044] Based on the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w ' 4) The node cost function is adjusted to obtain an improved node cost function, and the target inventory planning path is corrected based on the improved node cost function. The expression of the improved node cost function is as follows:

[0045]

[0046] Where h L (n) represents the Manhattan distance from the inventory shelf location node n to the end inventory location node; h S (n) indicates signal heuristic; h B (n) represents the power heuristic; h O (n) represents the obstacle heuristic.

[0047] An intelligent inventory robot path optimization system based on environmental modeling and collaborative optimization, the path optimization system comprising:

[0048] The drone scanning and data transmission module is used to scan the target storage area through the drone platform and generate three-dimensional warehouse point cloud data. The three-dimensional warehouse point cloud data is processed by point cloud denoising, plane fitting, shelf boundary extraction and channel detection to generate a two-dimensional warehouse plan map and transmit it to the robot data base;

[0049] An inventory path generation and optimization module is used to generate an initial inventory planning path based on the two-dimensional warehouse plan using an A* algorithm, and optimize the initial inventory planning path using a simulated annealing algorithm to generate a target inventory planning path;

[0050] The multi-objective optimization and PSO dynamic weight correction module is used to correct the target inventory planning path based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, through a hybrid algorithm of multi-objective optimization and PSO dynamic weight;

[0051] The inventory progress viewing and manual adjustment module is used for the operator to view the robot inventory progress through the AR device and manually adjust the target inventory planning path.

[0052] An electronic device includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the steps of an intelligent inventory robot path optimization method for environmental modeling and collaborative optimization as described in any one of the above.

[0053] A readable storage medium having a computer program stored therein, wherein the computer program, when executed by a processor, is used to implement the steps of an intelligent inventory robot path optimization method for environment modeling and collaborative optimization as described in any one of the above.

[0054] Compared with related technologies, the intelligent inventory robot path optimization method based on environment modeling and collaborative optimization provided by the present invention has the following beneficial effects:

[0055] The present invention uses a drone platform to scan the target storage area and generate three-dimensional storage point cloud data, performs point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the three-dimensional warehouse point cloud data to generate a two-dimensional warehouse plan map and transmit it to the robot data base; uses the A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan map, and optimizes the initial inventory planning path through a simulated annealing algorithm to generate a target inventory planning path; based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights; the operator checks the robot's inventory progress through an AR device and manually adjusts the target inventory planning path, so that the inventory path of the intelligent inventory robot can be optimized in real time in a dynamic environment, thereby improving the robot's inventory efficiency, coverage and endurance.

[0056] The present invention can deeply integrate RFID signal strength, robot remaining power and dynamic obstacle distribution data, dynamically adjust path priority through real-time data collection and intelligent analysis technology, and automatically plan priority detours for areas with weak signal coverage and low power threshold ranges, ensuring the continuity and accuracy of inventory tasks and improving task execution capabilities in complex environments. The method of the present invention can, with the help of the PSO algorithm, flexibly allocate the weight ratios of path length, signal coverage density, energy consumption and obstacle avoidance safety according to the real-time needs of warehousing operations. While ensuring path efficiency, it dynamically coordinates signal integrity and energy management requirements to achieve intelligent strategy adaptation in different operating scenarios, significantly improving inventory efficiency and equipment endurance performance. The present invention constructs a multi-level obstacle avoidance protection mechanism through dynamic obstacle trajectory prediction and segmented path planning technology, which can capture dynamic obstacle information in the environment in real time, predict its motion trajectory and generate a safe avoidance path, effectively reducing the risk of collision and ensuring the stable operation of the robot in a warehousing environment with high-frequency personnel flow and equipment movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flow chart of a method for intelligent inventory robot path optimization based on environment modeling and collaborative optimization provided by an embodiment of the present invention;

[0058] Figure 2 A system block diagram of an intelligent inventory robot path optimization system for environment modeling and collaborative optimization provided by an embodiment of the present invention;

[0059] Figure 3 A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0061] like Figure 1 FIG. 1 is a flow chart of a method for intelligent inventory robot path optimization based on environment modeling and collaborative optimization provided by an embodiment of the present invention. Figure 1 The execution subject of the method shown may be a software and / or hardware device. The execution subject of the present application may include but is not limited to at least one of the following: user equipment, network equipment, etc. Among them, the user equipment may include but is not limited to computers, smart phones, personal digital assistants (Personal Digital Assistant, referred to as: PDA) and the electronic devices mentioned above. Network equipment may include but is not limited to a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers, wherein cloud computing is a type of distributed computing, a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. It includes steps S1 to S4, as follows:

[0062] S1, using the drone platform to scan the target storage area and generate three-dimensional storage point cloud data, performing point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the three-dimensional storage point cloud data to generate a two-dimensional storage plan and transmit it to the robot data base;

[0063] Among them, drone platforms refer to unmanned aerial vehicles equipped with sensors such as lidar and cameras. They can conduct all-round scans of target storage areas according to preset routes, collecting three-dimensional spatial data on a large scale and with high efficiency. They are particularly suitable for storage scenarios with high-rise shelves or complex layouts. Three-dimensional warehouse point cloud data refers to a collection of massive discrete points formed by the reflection of the laser beam emitted by the lidar in space. Each point contains x, y, z coordinates and reflection intensity information, which can fully characterize the geometric form of the storage environment and accurately restore the spatial position of shelves, aisles, and obstacles. A two-dimensional warehouse plan refers to a grid map formed by projecting three-dimensional point cloud data onto a horizontal plane, which is marked with information such as shelf location, aisle width, and entrances and exits. The data base of the intelligent inventory robot is usually integrated with a wireless communication module and charging facilities, which can receive, store, and distribute two-dimensional map data.

[0064] In practical applications, a multi-sensor drone platform can be used to perform a full-scale scan of the target storage area. The point cloud acquisition capability of the lidar and the posture calculation function of the IMU inertial navigation system can then be used to generate three-dimensional storage point cloud data containing spatial coordinates and reflection intensity information.

[0065] Specifically, voxel grid filtering can first be used to divide the continuous three-dimensional space into uniform cubic units. By retaining the voxel centroid, point cloud downsampling and noise suppression can be achieved. Subsequently, the Random Sample Consensus Algorithm (RANSAC) algorithm can be applied to fit planar structures, such as the ground surface and shelf shelves, from the point cloud. Delaunay triangulation and Alpha Shapes algorithms are used to extract shelf boundary polygons, and the Douglas-Peucker algorithm is used for outline simplification, thereby reducing data complexity while preserving geometric features. Finally, bird's-eye view rasterization and morphological closing operations can be used to distinguish between shelf-occupied areas and navigable aisles. A two-dimensional warehouse plan with shelf layout, aisle widths, and entrance and exit locations can be generated. This planar plan can then be transmitted to the robot data base via a wireless communication link, providing a static environment model for subsequent path planning.

[0066] S2, using an A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan, and optimizing the initial inventory planning path using a simulated annealing algorithm to generate a target inventory planning path;

[0067] It can be understood that the A* algorithm (A-Star Algorithm) is a heuristic path search algorithm that can quickly and accurately determine the shortest path in a two-dimensional grid map. The initial inventory planning path refers to the initial path for inventorying all shelves planned based on a two-dimensional warehouse plan. The simulated annealing algorithm (SA) is a global optimization algorithm that simulates the physical annealing process and allows differential solutions to be accepted in the search to jump out of the local optimum, thereby optimizing the initial path, reducing repeated scanning and invalid walking, and improving the path smoothness. The target inventory planning path refers to the path for the intelligent inventory robot to perform shelf inventory after being optimized by the simulated annealing algorithm.

[0068] Specifically, based on a two-dimensional warehouse floor plan, the A* algorithm can be used to evaluate the actual cost and target heuristic estimate of the function's balanced path. The actual cost includes travel distance, number of turns, and aisle width constraints, while the heuristic function uses Manhattan distance to quickly estimate the shortest path from a node to the destination. Through a priority queue expansion mechanism, the A* algorithm ensures that the node with the lowest overall cost is selected for expansion each time. This generates a sequence of shortest paths covering all inventory shelves in a grid map, consisting of continuous turning points and straight path segments.

[0069] To further improve path quality, a simulated annealing algorithm can be used to locally optimize the initial path. This algorithm generates new paths through random perturbations and dynamically accepts inferior solutions during the cooling process according to the Metropolis acceptance criterion to escape the local optimum. The corresponding operations include performing multiple path perturbations within each temperature iteration cycle, calculating the change in path length, and directly accepting the new path if it is superior; otherwise, accepting the inferior solution with a probability related to the temperature. As the temperature parameter decays exponentially, the algorithm gradually converges to a globally optimal solution, generating a target inventory planning path with no repeated scans and low redundancy.

[0070] S3, based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights;

[0071] It should be noted that the dynamic obstacle impact factor is a factor used to quantify the impact of dynamic obstacles on the path. RFID signal strength refers to the signal strength value of the ultra-high frequency radio frequency identification tag, which can reflect the communication quality between the tag and the robot. The robot's remaining battery life refers to the remaining battery power of the intelligent inventory robot at a specific moment in the shelf inventory process. AR devices are augmented reality terminals based on SLAM technology. They use spatial mapping to overlay the virtual path with the physical environment and support gesture interaction and voice commands.

[0072] During robot operations, they must respond to changes in the warehouse environment in real time. LiDAR and cameras collect dynamic obstacle location data, combined with RFID signal strength and the robot's remaining battery life to establish multi-dimensional constraints. RFID signal strength can also reflect the shielding effect of metal shelves. A multi-objective optimization function integrates conflicting objectives such as path length, signal coverage quality, energy consumption, and obstacle impact. Using a particle swarm optimization (PSO) algorithm, the weights of these objectives are dynamically adjusted to achieve adaptive strategies for different operational scenarios.

[0073] The PSO algorithm simulates the collaborative mechanism of a flock of birds, treating each set of weight coefficients as a particle. It guides the particles to search for the optimal solution in the weight space through individual and global optimization. The optimized weights are then injected into the improved A* algorithm to adjust the heuristic function and balance multiple objectives. For example, this can increase scanning priority in areas with weak signals, prioritize charging routes when the battery is low, or generate obstacle avoidance trajectories around dynamic obstacles.

[0074] This hybrid algorithm can realize closed-loop control of static planning and dynamic correction, significantly improving the robot's path adaptability in dynamic scenarios such as personnel movement and shelf adjustment, and reducing path redundancy.

[0075] S4, the operator checks the robot's inventory progress through the AR device and manually adjusts the target inventory planning path.

[0076] The operator is responsible for tasks such as scheduling, path monitoring, and exception handling during the robotic inventory process. Robot inventory progress is a core indicator used to quantify and monitor the robot's operational status. It reflects the robot's progress in scanning, identifying, and verifying stored goods within a specified timeframe. This serves as an important basis for human-robot collaborative intervention, providing operators with visual feedback on task execution.

[0077] AR devices can create a virtual-reality monitoring interface, visualizing the robot's inventory progress. Using spatial registration technology, AR devices align virtual path information with the physical warehouse environment, allowing operators to intuitively view key indicators such as robot location, areas counted, and RFID reading status.

[0078] In addition, when signal reading anomalies, equipment failures, or emergency tasks are detected, operators can use the AR interface to mark temporary restricted areas, adjust shelf inventory priorities, or issue immediate obstacle avoidance instructions. After receiving the instructions, the robot triggers the path replanning mechanism and updates the operation strategy in real time.

[0079] This human-machine collaboration mechanism compensates for the lack of flexibility of purely automated systems in complex and unusual scenarios, forming a two-way feedback loop between manual monitoring and intelligent execution. For example, when encountering a temporary stacking obstacle, the operator can manually draw a detour using an AR device. The robot then quickly generates and executes a new path based on an improved A* algorithm, ensuring task continuity and improving the reliability of inventory operations.

[0080] In the specific implementation process, the drone platform is used to scan the target storage area and generate three-dimensional storage point cloud data, and the three-dimensional storage point cloud data is processed by point cloud denoising, plane fitting, shelf boundary extraction and channel detection to generate a two-dimensional storage plan and transmit it to the robot data base. Specifically, it includes:

[0081] The drone platform is equipped with a 16-line laser radar and an IMU inertial navigation system, which performs overlapping scans along a preset grid route of the target storage area at a preset cruising speed to generate raw storage point cloud data;

[0082] The voxel grid filtering technology is used to divide the three-dimensional space corresponding to the original warehouse point cloud data into 0.08m 3 The voxels of the cube, and by the formula Performing voxel indexing on each of the original stored point cloud data and retaining the centroid point within each of the voxels to generate the three-dimensional stored point cloud data;

[0083] Among them, Voxel(p) represents the calculation point p(p x ,p y ,p z ) of voxels; v x 、v y 、v z Represents the side length of the voxel in the x, y, and z directions respectively; They represent the voxel indices to which the calculation point p belongs in the x, y, and z directions, respectively; W represents the total width of the voxel grid in the x direction, that is, the number of voxels in the x direction; H represents the total width of the voxel grid in the y direction, that is, the number of voxels in the y direction.

[0084] The drone platform, equipped with a 16-line laser radar and an inertial navigation system (IMU), performs overlapping scans of target storage areas at preset cruising speeds and grid routes. This multi-sensor collaboration enables three-dimensional coverage of the storage space. The 16-line laser radar emits laser beams and receives reflected signals to construct raw point cloud data containing three-dimensional coordinates and reflection intensity information. The IMU inertial navigation system calculates the drone's pose parameters, namely the translation and rotation matrices, in real time, providing a reference alignment of the spatial coordinate system for the point cloud data, ensuring accurate splicing of data from different scanned areas.

[0085] Furthermore, the voxel grid filtering technology can be used to discretize the three-dimensional space and map the original point cloud data to the 0.08m 3 Then, based on the 3D coordinates of each original point cloud data point, it can be assigned to the corresponding voxel unit. By calculating the centroid coordinates of all points in each voxel, the representative points that can characterize the spatial characteristics of the voxel are retained to generate 3D warehouse point cloud data. This can remove redundant data while retaining the geometric characteristics of the environment. This voxelization process not only reduces the size of the point cloud data, but also suppresses the influence of random noise through the statistical characteristics of the centroid points, thereby improving the reliability of the data.

[0086] The core mechanism of voxel indexing is to achieve group management of point clouds through quantitative mapping of spatial coordinates. For any calculation point, its position in three-dimensional space can be decomposed into three axial coordinate components. By rounding the ratio of each component to the voxel side length, the index position of the point in the voxel grid can be determined. Specifically, the voxel index of the calculation point in the x, y, and z directions is obtained by rounding the ratio of its coordinate value to the corresponding axial voxel side length, so that the three-dimensional space point cloud can be converted into a discrete voxel unit index sequence. This indexing method not only facilitates the computer's storage and retrieval of spatial data, but also provides a structured data foundation for subsequent operations such as plane fitting and boundary extraction.

[0087] The process of shelf boundary extraction and channel detection is as follows:

[0088] In the three-dimensional coordinate system corresponding to the three-dimensional warehouse point cloud data, for each plane, a local coordinate system orthogonal to the plane normal vector is constructed with the plane centroid as the origin, and the three-dimensional warehouse point cloud data is projected into the local coordinate system;

[0089] Perform Delaunay triangulation on the projected point cloud, generate a closed shelf boundary polygon by screening the triangle edges with the minimum circumscribed circle radius, and perform boundary error control on the closed shelf boundary polygon using the Douglas-Peucker algorithm to generate the shelf boundary contour;

[0090] The ground point cloud is converted into a bird's-eye view grid, and the pixel gaps in the three-dimensional warehouse point cloud data are filled by morphological closing operation, and the pixel occupancy probability P is calculated. occ ;

[0091] Get the shelf channel distinction probability and the preset width limit, the pixel occupancy probability P occ All pixels with a probability greater than the shelf channel distinction probability constitute shelf elements, and the pixel occupancy probability P occ All pixel points with a probability less than the shelf channel distinction probability constitute channel elements, and entrance and exit elements are marked at the same time, and the width between the pixel points corresponding to the two ends of the entrance and exit elements is greater than the preset width limit;

[0092] The shelf ID, the aisle width and the entrance and exit direction are respectively added to the shelf element, the aisle element and the entrance and exit element.

[0093] It is understandable that in the processing framework of three-dimensional warehouse point cloud data, a local coordinate system is constructed for each fitted plane, such as shelf shelves, walls, etc., with the plane's centroid as the origin. The coordinate axes of this local coordinate system are orthogonal to the plane normal vector, that is, the normal vector serves as a principal axis of the local coordinate system, usually the z-axis, and the other two axes extend along the plane tangent direction. Through this transformation, the three-dimensional point cloud data is projected onto a two-dimensional plane of the local coordinate system, such as the xy plane, so that the dimensionality reduction processing of the point cloud data can be achieved through the coordinate system transformation, and the geometric prior knowledge of the plane can be used to improve the efficiency and accuracy of feature extraction.

[0094] Furthermore, Delaunay triangulation can be performed on the point cloud projected into the local coordinate system to construct a triangular mesh covering all points. This mesh has the characteristic of an empty circle, meaning that the circumcircle of any triangle contains no other points, ensuring the geometric rationality of the mesh. By screening the triangle edges with a minimum circumcircle radius less than a threshold, the boundary edges that represent the shelf outline are extracted, generating a closed shelf boundary polygon. This eliminates redundant edges caused by noise or discrete points, retaining the contour line that closely matches the shelf point cloud distribution.

[0095] To further improve the practicality of the boundary, the Douglas-Peucker algorithm can be used to optimize the closed polygon. This algorithm recursively evaluates the contribution of vertices to the boundary shape and removes vertices whose distance from the fitted line is less than a preset error threshold. This reduces the number of vertices while maintaining the overall shape of the boundary. This process effectively reduces the complexity of the boundary data, avoiding computational redundancy caused by excessive details, while ensuring that the geometric accuracy of the shelf outline in the warehouse plan meets the requirements of path planning.

[0096] The ground point cloud data can then be converted into a bird's-eye view raster, where each grid cell corresponds to a fixed dimension in the physical space. A morphological closing operation, dilation followed by erosion, is used to process the raster data, filling pixel gaps caused by point cloud sparsity or noise and smoothing the boundaries between shelves and aisles. The application of closing operations effectively connects adjacent shelf point cloud regions, avoiding misidentification of aisles due to missing points, while maintaining the geometric connectivity of the shelf regions.

[0097] Based on the rasterization, the occupancy probability of each pixel can be calculated. This probability reflects the likelihood that a shelf is occupied at that location. It is usually generated by counting the point cloud density or reflection intensity within the grid. The occupancy probability ranges from [0, 1], with higher values indicating a greater likelihood of a shelf being present.

[0098] Based on the preset probability threshold for distinguishing shelves and aisles, the raster pixels can be binarized and segmented. Pixels with an occupation probability greater than the threshold are classified as shelf elements, and pixels with an occupation probability less than the threshold are classified as channel elements. The setting of this threshold needs to be combined with prior knowledge of the warehousing scenario. For example, in areas with dense metal shelves, the threshold can be set to 0.7 to ensure the integrity of the shelf area. At the same time, by detecting connected areas in channel elements with a width greater than the preset limit and marking the entrance and exit elements, it can be ensured that the entrances and exits meet the robot's two-way passage or turning operations. The limit is usually set according to the robot size and passage requirements.

[0099] To achieve semantic management of warehouse elements, attribute information can be added to shelf elements, aisle elements, and entrance and exit elements: shelf ID, used to identify the shelf's physical location and storage attributes, such as location number and cargo type; aisle width, used to record aisle width parameters and provide capacity constraints for path planning; entrance and exit directions, used to mark the directional attributes of entrances and exits, such as north-south and east-west, to assist robot navigation in path selection. This attribute information is embedded in the two-dimensional warehouse floor plan as metadata, forming a structured semantic map, providing multidimensional data support for subsequent path planning, task scheduling, and conflict avoidance.

[0100] The method of using the A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan, and optimizing the initial inventory planning path through a simulated annealing algorithm to generate a target inventory planning path specifically includes:

[0101] Determine the inventory shelf location based on the two-dimensional warehouse plan, obtain the cargo asset type and inventory location priority corresponding to the inventory shelf location from the warehouse management library corresponding to the target storage area, and set the robot inventory motion parameters, namely the maximum walking speed, minimum safety distance, and single-shelf single-layer scanning time;

[0102] The A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated by combining the type of the goods asset and the priority of the inventory location;

[0103] The initial inventory planning path is locally optimized, that is, the initial inventory planning path is optimized by the simulated annealing algorithm to generate the target inventory planning path.

[0104] In practical applications, all shelves requiring inventory can be identified based on a two-dimensional warehouse floor plan. The corresponding asset type and inventory priority data for each shelf can then be retrieved from the warehouse management database. The asset type determines the scanning strategy, such as requiring repeated verification for high-value assets, while the inventory priority is used to optimize the routing. Furthermore, the motion parameters of the intelligent inventory robot can be configured, including maximum travel speed, minimum safe distance, and scanning time per shelf and per layer. These parameters act as constraints to ensure the physical feasibility of the generated path.

[0105] The A* algorithm can then be used to generate an initial inventory path. This algorithm uses a two-dimensional grid map as its search space and calculates node costs using a comprehensive evaluation function. The actual movement cost accounts for distance, number of turns, and aisle width, while the heuristic cost estimate is based on the spatial distance to the target location. The algorithm iteratively expands the minimum-cost nodes using a priority queue. It then sorts the path nodes based on the type of asset and inventory priority, generating a sequence of shortest paths from the starting location to all target shelves. This sequence consists of continuous straight line segments and turning points, ensuring coverage of all inventory tasks and satisfying access constraints.

[0106] To further improve the path quality, a simulated annealing algorithm can be used to optimize the initial path. This algorithm generates a new path through random perturbation operations and dynamically accepts inferior solutions during the cooling process according to the Metropolis acceptance criterion to escape the local optimum. The specific operations include: performing multiple path transformations in each temperature iteration cycle, calculating the changes in path length and constraint satisfaction, and directly adopting the new path if it is better. Otherwise, the inferior solution is accepted with a probability related to the current temperature. With the exponential decay of the temperature parameter, the algorithm gradually converges to the global optimal solution and generates a target inventory planning path with no redundant detours and the least number of turns. Theoretically, this path is shorter than the initial path by a certain proportion and has better motion smoothness and task execution efficiency.

[0107] The A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated in combination with the type of the goods asset and the priority of the inventory location, specifically including:

[0108] Set the node cost function as follows:

[0109] f(n)=g(n)+h(n)

[0110]

[0111] h(n)=|n x -t x |-|n y -t y |

[0112] Where f(n) represents the total estimated cost of inventory shelf location node n; g(n) represents the actual movement cost from the starting inventory location node to the inventory shelf location node n; Indicates the node from the starting inventory location to the previous inventory shelf location node The actual moving cost; move_cost(n) represents the cost of moving from the previous inventory shelf location node to node The basic moving cost to the inventory shelf location node n is If the node n moves horizontally or vertically to the inventory shelf location node n, then move_cost(n) = 1. If the node parent(n) moves diagonally to the inventory shelf location node n, then turn_penalty(n) represents the turning penalty of inventory rack location node n. If a turn is required from the previous inventory rack location node parent(n) to the inventory rack location node n, then turn_penalty(n) = γ, where γ represents the turning penalty coefficient. Otherwise, turn_penalty(n) = 0. width_penalty(n) represents the aisle width penalty of inventory rack location node n. β represents the width penalty coefficient, w0 represents the standard channel width, w(n) represents the channel width of the inventory shelf location node n; h(n) represents the heuristic estimated cost from the inventory shelf location node n to the end inventory location node; (n x ,n y ) represents the coordinates of the inventory shelf location node n; (t x ,t y ) represents the coordinates of the node where the inventory is terminated;

[0113] The inventory shelf location node with the smallest total estimated cost is expanded and selected through a priority queue until all the inventory shelf location nodes are traversed, thereby generating a continuous walking sequence including turning points and straight path segments, that is, the shortest path from the starting inventory location to the ending shelf location, and combining the cargo asset type and the inventory location priority to generate the initial inventory planning path.

[0114] Among them, the node cost function adopts a two-component model, and the total estimated cost is composed of the actual movement cost and the heuristic estimated cost. The calculation of the actual movement cost includes four parts: the accumulated cost of the previous node, the basic movement cost, the turning penalty, and the channel width penalty. Among them, the basic movement cost sets different weights according to the difference in movement direction, that is, horizontal or vertical movement corresponds to a unit cost, and diagonal movement is set with a higher weight due to the longer distance. The turning penalty mechanism is used to suppress frequent turns. If the movement direction of the current node and the previous node changes, the penalty coefficient is superimposed; the channel width penalty is calculated by the ratio of the standard width to the width of the channel where the current node is located, so as to guide the path to prioritize wider channels to reduce the risk of collision.

[0115] The heuristic cost estimation uses the Manhattan distance model to quickly estimate the remaining path length, ensuring the algorithm converges to the optimal solution in polynomial time. The priority control mechanism obtains data on the type of goods and inventory priority from the warehouse management system. Nodes corresponding to high-priority shelves receive a higher expansion weight in the priority queue, ensuring that intelligent inventory robots prioritize access to key areas during the inventory process.

[0116] The node with the lowest total estimated cost can then be dynamically expanded through a priority queue, maintaining a closed list of visited nodes and an open list of nodes to be expanded. In each iteration, the node with the lowest cost is selected from the open list for neighborhood expansion, the total estimated cost of the adjacent nodes is calculated, and the priority queue is updated until all inventory rack nodes have been traversed. The resulting path consists of a series of turning points and straight path segments, satisfying both geometric minimization and business-demand-driven path sorting through priority weights, forming an initial inventory planning path that complies with warehouse operation specifications. This path balances mobility efficiency, operational smoothness, and channel safety through the multi-factor weighting of the cost function while ensuring collision-free operation.

[0117] The locally optimizing the initial inventory planning path, i.e., optimizing the initial inventory planning path by the simulated annealing algorithm to generate the target inventory planning path, specifically includes:

[0118] Set the key parameters of the simulated annealing algorithm, namely the initial temperature T, cooling rate α, and termination temperature T ' and the number of disturbances Q at each temperature;

[0119] Defining a path perturbation operation includes randomly selecting two path nodes in the initial inventory planning path P, deleting the intermediate node between the two path nodes and reconnecting the initial inventory planning path P, randomly swapping the order of two non-adjacent path nodes in the initial inventory planning path P, and randomly selecting a subpath in the initial inventory planning path P and reversing the direction of the subpath;

[0120] The path perturbation operation is performed on the initial inventory planning path P to generate a new inventory planning path P ' , calculate the path length change ΔL = LL ' , where L represents the initial path length of the initial inventory planning path P, L ' Represents the new inventory planning path P ' The new path length;

[0121] If ΔL<0, then P1=P ' ; If ΔL ≥ 0, the path acceptance probability is calculated based on the Metropolis acceptance criterion If the path acceptance probability q is greater than the random number r, 0≤r<1, then P1=P ' , otherwise P1=P;

[0122] The initial temperature T gradually converges according to the cooling rate α, and the path perturbation operation is performed Q times at each temperature until the temperature is less than the termination temperature T ' , terminate the simulated annealing algorithm and output the target inventory planning path.

[0123] It's important to note that the algorithm parameters must first be initialized, setting the initial temperature, cooling rate, termination temperature, and number of single-temperature perturbations. The initial temperature determines the breadth of the search space; a higher initial temperature allows the algorithm to explore more path variants early on. The cooling rate controls the temperature decay rate and is typically a constant close to 1 to balance exploration and convergence. The termination temperature, serving as the algorithm's stopping condition, ensures solution stability. The number of single-temperature perturbations limits the local search depth at each temperature level.

[0124] The path perturbation operation achieves path structural variation through three basic transformations: randomly selecting two nodes in the path, removing the intermediate node between them, and directly connecting them to shorten the path; randomly swapping the order of non-adjacent nodes to adjust access priorities; and randomly reversing the direction of subpaths to explore local optimal arrangements. These operations are combined with a certain probability to generate diverse new path variants, breaking through the local structural limitations of the initial path.

[0125] After each perturbation, the algorithm calculates the change in length between the old and new paths. If the new path is shorter, it is accepted directly. If the new path is longer, the acceptance probability is calculated based on the Metropolis acceptance criterion. This probability is negatively exponentially related to the temperature and the path length difference, allowing the algorithm to accept inferior solutions with a higher probability during high-temperature phases, thus escaping the local optimum. By generating a random number between 0 and 1 and comparing it with the acceptance probability, the algorithm can determine whether to retain the new path, forming a probabilistic optimization mechanism.

[0126] The temperature decay mechanism is central to the convergence of the simulated annealing algorithm. The initial temperature decays exponentially at the same rate as the cooling rate. A preset number of perturbations are performed at each temperature level. As the temperature decreases, the algorithm's probability of accepting inferior solutions gradually decreases, ultimately converging to a local optimal solution at low temperatures. When the temperature falls below the termination temperature, the algorithm terminates and outputs an optimized target inventory planning path. This path significantly reduces the total trip length and improves path smoothness by reducing redundant turns and merging the order of visits to adjacent shelves.

[0127] Based on the collected dynamic obstacle impact factors, combined with the RFID signal strength and the remaining battery power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights, specifically including:

[0128] The RFID signal strength and the remaining battery power of the robot are normalized, and based on the target path length of the target inventory planning path, the normalized RFID signal strength and the remaining battery power of the robot, and the dynamic obstacle impact factor, a multi-objective optimization function is constructed as follows:

[0129]

[0130] Where F represents the comprehensive cost function; w1, w2, w3, w4 represent the target weight coefficients; L goal represents the target path length of the target inventory planning path; S norm represents the normalized RFID signal strength; B norm represents the remaining power of the robot after normalization; O represents the dynamic obstacle influence factor, d obs represents the distance from the intelligent inventory robot to the nearest obstacle, d safe represents the safety distance threshold of the intelligent inventory counting robot, and λ represents the dynamic obstacle influence coefficient;

[0131] Each particle represents a set of target weight vectors (w1, w2, w3, w4), and the particle speed and position are updated through PSO iteration to determine the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w ' 4);

[0132] Based on the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w '4) The node cost function is adjusted to obtain an improved node cost function, and the target inventory planning path is corrected based on the improved node cost function. The expression of the improved node cost function is as follows:

[0133]

[0134] Where h L (n) represents the Manhattan distance from the inventory shelf location node n to the end inventory location node; h S (n) indicates signal heuristic; h B (n) represents the power heuristic; h O (n) represents the obstacle heuristic.

[0135] In practical applications, the real-time collected RFID signal strength and the remaining battery power of the robot can first be normalized to convert the physical quantity into a dimensionless indicator in the [0,1] interval, so as to facilitate the unified measurement and fusion calculation of multi-source data.

[0136] Then, by combining the inherent length properties of the target inventory planning path and the spatial impact factors of dynamic obstacles, we can construct a multi-objective optimization function that includes path length, signal quality, device endurance, and obstacle avoidance safety. This function integrates the optimization objectives of each dimension into a comprehensive cost evaluation indicator through weighted summation.

[0137] The particle swarm optimization algorithm, the core mechanism for dynamic weight adjustment, maps the weight coefficients of each set of objectives to individuals in the particle swarm. By simulating the iterative search of a flock of birds, it dynamically optimizes the priority allocation of each objective. The particle speed and position update rules combine the bidirectional guidance of individual historical optimal solutions with the swarm's global optimal solution, ensuring that the weight vector efficiently searches for the optimal combination in multidimensional space. After several iterations, the optimal weight coefficient combination is obtained to balance the current environmental constraints. For example, in areas with weak signal, the weight of signal quality objectives is automatically increased, or the priority of battery life objectives is increased when the battery is low.

[0138] Based on the optimized weight coefficients, the original node cost function can be reconstructed, introducing signal heuristics, battery heuristics, and obstacle heuristics. The signal heuristic is negatively correlated with the normalized signal strength, which can guide the path to prioritize areas with reliable signals and improve inventory accuracy. The battery heuristic is negatively correlated with the remaining battery power to ensure that low-battery risks are avoided in advance during path planning. The obstacle heuristic is positively correlated with the dynamic obstacle impact factor, forcing the path to stay away from dangerous areas. Through the improved node cost function, the priority of path nodes can be recalculated and local path replanning can be implemented to generate a corrected path that adapts to real-time environmental changes. This path can dynamically balance multiple objectives such as efficiency, accuracy, endurance, and safety while ensuring the completion of the inventory task, significantly improving the robustness of the robot's operations in dynamic warehousing environments.

[0139] like Figure 2 FIG. 1 is a system block diagram of an intelligent inventory robot path optimization system for environment modeling and collaborative optimization provided by an embodiment of the present invention. The path optimization system includes:

[0140] The drone scanning and data transmission module is used to scan the target storage area through the drone platform and generate three-dimensional warehouse point cloud data. The three-dimensional warehouse point cloud data is processed by point cloud denoising, plane fitting, shelf boundary extraction and channel detection to generate a two-dimensional warehouse plan map and transmit it to the robot data base;

[0141] An inventory path generation and optimization module is used to generate an initial inventory planning path based on the two-dimensional warehouse plan using an A* algorithm, and optimize the initial inventory planning path using a simulated annealing algorithm to generate a target inventory planning path;

[0142] The multi-objective optimization and PSO dynamic weight correction module is used to correct the target inventory planning path based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, through a hybrid algorithm of multi-objective optimization and PSO dynamic weight;

[0143] The inventory progress viewing and manual adjustment module is used for the operator to view the robot inventory progress through the AR device and manually adjust the target inventory planning path.

[0144] Figure 2 The apparatus of the embodiment shown can be used to perform Figure 1 The implementation principles and technical effects of the steps in the method embodiment shown are similar and will not be repeated here.

[0145] An electronic device includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the steps of an intelligent inventory robot path optimization method for environmental modeling and collaborative optimization as described in any one of the above.

[0146] like Figure 3 FIG. 1 is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32 and a computer program;

[0147] The memory 32 is used to store the computer program, which may also be a flash memory. The computer program is, for example, an application program or a functional module for implementing the above method.

[0148] The processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant description in the above method embodiment.

[0149] Optionally, the memory 32 may be independent or integrated with the processor 31 .

[0150] When the memory 32 is a device independent of the processor 31, the device may further include:

[0151] The bus 33 is used to connect the memory 32 and the processor 31 .

[0152] A readable storage medium having a computer program stored therein, wherein the computer program, when executed by a processor, is used to implement the steps of an intelligent inventory robot path optimization method for environment modeling and collaborative optimization as described in any one of the above.

[0153] Among them, the readable storage medium can be a computer storage medium or a communication medium. Communication media include any medium that facilitates the transmission of computer programs from one place to another. Computer storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the readable storage medium can also exist in a communication device as discrete components. The readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0154] The present invention also provides a program product, which includes execution instructions stored in a readable storage medium. At least one processor of a device can read the execution instructions from the readable storage medium, and at least one processor executes the execution instructions so that the device implements the methods provided in the various embodiments described above.

[0155] In the embodiments of the above-mentioned devices, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0156] Through the introduction of the above embodiments, the present invention adopts an intelligent inventory robot path optimization method based on environmental modeling and collaborative optimization, scans the target storage area through a drone platform and generates three-dimensional warehouse point cloud data, performs point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the three-dimensional warehouse point cloud data to generate a two-dimensional warehouse plan and transmit it to the robot data base; adopts the A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan, and optimizes the initial inventory planning path through the simulated annealing algorithm to generate a target inventory planning path; based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weight; the operator views the robot inventory progress through the AR device and manually adjusts the target inventory planning path, so that the inventory path of the intelligent inventory robot can be optimized in real time in a dynamic environment, thereby improving the robot's inventory efficiency, coverage and endurance.

[0157] The present invention can deeply integrate RFID signal strength, robot remaining power and dynamic obstacle distribution data, dynamically adjust path priority through real-time data collection and intelligent analysis technology, and automatically plan priority detours for areas with weak signal coverage and low power threshold ranges, ensuring the continuity and accuracy of inventory tasks and improving task execution capabilities in complex environments. The method of the present invention can, with the help of the PSO algorithm, flexibly allocate the weight ratios of path length, signal coverage density, energy consumption and obstacle avoidance safety according to the real-time needs of warehousing operations. While ensuring path efficiency, it dynamically coordinates signal integrity and energy management requirements to achieve intelligent strategy adaptation in different operating scenarios, significantly improving inventory efficiency and equipment endurance performance. The present invention constructs a multi-level obstacle avoidance protection mechanism through dynamic obstacle trajectory prediction and segmented path planning technology, which can capture dynamic obstacle information in the environment in real time, predict its motion trajectory and generate a safe avoidance path, effectively reducing the risk of collision and ensuring the stable operation of the robot in a warehousing environment with high-frequency personnel flow and equipment movement.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A path optimization method for an intelligent inventory robot based on environment modeling and collaborative optimization, characterized in that: The path optimization method comprises: Scan the target warehouse area through the drone platform and generate 3D warehouse point cloud data, perform point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the 3D warehouse point cloud data to generate a 2D warehouse plan and transmit it to the robot data base; An A* algorithm is used to generate an initial inventory planning path based on the two-dimensional warehouse plan, and the initial inventory planning path is optimized by a simulated annealing algorithm to generate a target inventory planning path; Based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights; The operator checks the robot's inventory progress through the AR device and manually adjusts the target inventory planning path; The method of using the A* algorithm to generate an initial inventory planning path based on the two-dimensional warehouse plan, and optimizing the initial inventory planning path through a simulated annealing algorithm to generate a target inventory planning path specifically includes: Determine the inventory shelf location based on the two-dimensional warehouse plan, obtain the cargo asset type and inventory location priority corresponding to the inventory shelf location from the warehouse management library corresponding to the target storage area, and set the robot inventory motion parameters, namely the maximum walking speed, minimum safety distance, and single-shelf single-layer scanning time; The A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated by combining the type of the goods asset and the priority of the inventory location; Locally optimizing the initial inventory planning path, that is, optimizing the initial inventory planning path by using the simulated annealing algorithm to generate the target inventory planning path; The A* algorithm is used to search for the shortest path from the starting inventory location to the ending shelf location, and the initial inventory planning path is generated in combination with the type of the goods asset and the priority of the inventory location, specifically including: Set the node cost function as follows: f(n)=g(n)+h(n) h(n)=|n x -t x |-|n y -t y | Where f(n) represents the total estimated cost of inventory shelf location node n; g(n) represents the actual movement cost from the starting inventory location node to the inventory shelf location node n; Indicates the node from the starting inventory location to the previous inventory shelf location node The actual moving cost; move_cost(n) represents the cost of moving from the previous inventory shelf location node to node The basic moving cost to the inventory shelf location node n is If the node moves horizontally or vertically to the inventory shelf location node n, then move_cost(n) = 1. If the node moves diagonally from the previous inventory shelf location node parent(n) to the inventory shelf location node n, then turn_penalty(n) represents the turning penalty of inventory rack location node n. If a turn is required from the previous inventory rack location node parent(n) to the inventory rack location node n, then turn_penalty(n) = γ, where γ represents the turning penalty coefficient. Otherwise, turn_penalty(n) = 0. width_penalty(n) represents the aisle width penalty of inventory rack location node n. β represents the width penalty coefficient, w0 represents the standard channel width, w(n) represents the channel width of the inventory shelf location node n; h(n) represents the heuristic estimated cost from the inventory shelf location node n to the end inventory location node; (n x ,n y ) represents the coordinates of the inventory shelf location node n; (t x ,t y ) represents the coordinates of the node where the inventory is terminated; Expanding and selecting the inventory rack location node with the smallest total estimated cost through a priority queue until all the inventory rack location nodes are traversed, generating a continuous walking sequence including turning points and straight path segments, that is, the shortest path from the starting inventory location to the ending inventory location, and combining the cargo asset type and the inventory location priority to generate the initial inventory planning path; Based on the collected dynamic obstacle impact factors, combined with the RFID signal strength and the remaining battery power of the robot, the target inventory planning path is corrected through a hybrid algorithm of multi-objective optimization and PSO dynamic weights, specifically including: The RFID signal strength and the remaining battery power of the robot are normalized, and based on the target path length of the target inventory planning path, the normalized RFID signal strength and the remaining battery power of the robot, and the dynamic obstacle impact factor, a multi-objective optimization function is constructed as follows: Where F represents the comprehensive cost function; w1, w2, w3, w4 represent the target weight coefficients; L goal represents the target path length of the target inventory planning path; S norm represents the normalized RFID signal strength; B norm represents the remaining power of the robot after normalization; O represents the dynamic obstacle influence factor, d obs represents the distance from the intelligent inventory robot to the nearest obstacle, d safe represents the safety distance threshold of the intelligent inventory counting robot, and λ represents the dynamic obstacle influence coefficient; Each particle represents a set of target weight vectors (w1, w2, w3, w4), and the particle speed and position are updated through PSO iteration to determine the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w ' 4); Based on the optimal combination of the target weight coefficients (w ' 1,w ' 2,w ' 3,w ' 4) The node cost function is adjusted to obtain an improved node cost function, and the target inventory planning path is corrected based on the improved node cost function. The expression of the improved node cost function is as follows: Where h L (n) represents the Manhattan distance from the inventory shelf location node n to the end inventory location node; h S (n) indicates signal heuristic; h B (n) represents the power heuristic; h O (n) represents the obstacle heuristic.

2. The intelligent inventory robot path optimization method based on environment modeling and collaborative optimization according to claim 1 is characterized in that: The method comprises: scanning the target storage area through the drone platform and generating three-dimensional storage point cloud data; performing point cloud denoising, plane fitting, shelf boundary extraction and channel detection on the three-dimensional storage point cloud data to generate a two-dimensional storage plan and transmitting the plan to the robot data base; The drone platform is equipped with a 16-line laser radar and an IMU inertial navigation system, which performs overlapping scans along a preset grid route of the target storage area at a preset cruising speed to generate raw storage point cloud data; The voxel grid filtering technology is used to divide the three-dimensional space corresponding to the original warehouse point cloud data into 0.08m 3 The voxels of the cube, and by the formula Performing voxel indexing on each of the original stored point cloud data and retaining the centroid point within each of the voxels to generate the three-dimensional stored point cloud data; Among them, Voxel(p) represents the calculation point p(p x ,p y ,p z ) of voxels; v x 、v y 、v z Represents the side length of the voxel in the x, y, and z directions respectively; They represent the voxel indices to which the calculation point p belongs in the x, y, and z directions, respectively; W represents the total width of the voxel grid in the x direction, that is, the number of voxels in the x direction; H represents the total width of the voxel grid in the y direction, that is, the number of voxels in the y direction.

3. The intelligent inventory robot path optimization method based on environment modeling and collaborative optimization according to claim 1 is characterized in that: The process of shelf boundary extraction and channel detection is as follows: In the three-dimensional coordinate system corresponding to the three-dimensional warehouse point cloud data, for each plane, a local coordinate system orthogonal to the plane normal vector is constructed with the plane centroid as the origin, and the three-dimensional warehouse point cloud data is projected into the local coordinate system; Perform Delaunay triangulation on the projected point cloud, generate a closed shelf boundary polygon by screening the triangle edges with the minimum circumscribed circle radius, and perform boundary error control on the closed shelf boundary polygon using the Douglas-Peucker algorithm to generate the shelf boundary contour; The ground point cloud is converted into a bird's-eye view grid, and the pixel gaps in the three-dimensional warehouse point cloud data are filled by morphological closing operation, and the pixel occupancy probability P is calculated. occ ; Get the shelf channel distinction probability and the preset width limit, the pixel occupancy probability P occ All pixels with a probability greater than the shelf channel distinction probability constitute shelf elements, and the pixel occupancy probability P occ All pixel points with a probability less than the shelf channel distinction probability constitute channel elements, and entrance and exit elements are marked at the same time, and the width between the pixel points corresponding to the two ends of the entrance and exit elements is greater than the preset width limit; The shelf ID, the aisle width and the entrance and exit direction are respectively added to the shelf element, the aisle element and the entrance and exit element.

4. The method for intelligent inventory robot path optimization based on environment modeling and collaborative optimization according to claim 1 is characterized in that: The locally optimizing the initial inventory planning path, i.e., optimizing the initial inventory planning path by the simulated annealing algorithm to generate the target inventory planning path, specifically includes: Set the key parameters of the simulated annealing algorithm, namely the initial temperature T, cooling rate α, and termination temperature T ' and the number of disturbances Q at each temperature; Defining a path perturbation operation includes randomly selecting two path nodes in the initial inventory planning path P, deleting the intermediate node between the two path nodes and reconnecting the initial inventory planning path P, randomly swapping the order of two non-adjacent path nodes in the initial inventory planning path P, and randomly selecting a subpath in the initial inventory planning path P and reversing the direction of the subpath; The path perturbation operation is performed on the initial inventory planning path P to generate a new inventory planning path P ' , calculate the path length change ΔL = LL ' , where L represents the initial path length of the initial inventory planning path P, L ' Represents the new inventory planning path P ' The new path length; If ΔL<0, then P1=P ' ; If ΔL ≥ 0, the path acceptance probability is calculated based on the Metropolis acceptance criterion If the path acceptance probability q is greater than the random number r, 0≤r<1, then P1=P ' , otherwise P1=P; The initial temperature T gradually converges according to the cooling rate α, and the path perturbation operation is performed Q times at each temperature until the temperature is less than the termination temperature T ' , terminate the simulated annealing algorithm and output the target inventory planning path.

5. An intelligent inventory robot path optimization system based on environmental modeling and collaborative optimization, applied to an intelligent inventory robot path optimization method based on environmental modeling and collaborative optimization as claimed in any one of claims 1 to 4, characterized in that: The path optimization system includes: The drone scanning and data transmission module is used to scan the target storage area through the drone platform and generate three-dimensional warehouse point cloud data. The three-dimensional warehouse point cloud data is processed by point cloud denoising, plane fitting, shelf boundary extraction and channel detection to generate a two-dimensional warehouse plan map and transmit it to the robot data base; An inventory path generation and optimization module is used to generate an initial inventory planning path based on the two-dimensional warehouse plan using an A* algorithm, and optimize the initial inventory planning path using a simulated annealing algorithm to generate a target inventory planning path; The multi-objective optimization and PSO dynamic weight correction module is used to correct the target inventory planning path based on the collected dynamic obstacle influencing factors, combined with the RFID signal strength and the remaining battery power of the robot, through a hybrid algorithm of multi-objective optimization and PSO dynamic weight; The inventory progress viewing and manual adjustment module is used for the operator to view the robot inventory progress through the AR device and manually adjust the target inventory planning path.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor runs the computer program stored in the memory, the processor executes the steps of the intelligent inventory robot path optimization method for environment modeling and collaborative optimization as described in any one of claims 1 to 4.

7. A readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, it is used to implement the steps of an intelligent inventory robot path optimization method for environment modeling and collaborative optimization as described in any one of claims 1 to 4.

Images