Warehouse path planning method based on virtual reality

Abstract

The application relates to the technical field of intelligent path planning in a virtual reality environment, and discloses a warehouse path planning method based on virtual reality. A unified coordinate system is established on the ground of the warehouse, the picking points are uniquely numbered and positioned, and the data consistency and traceability are ensured; the path generation adopts the nearest neighbor principle and combines node order and numbering rules to obtain a preliminary reasonable route; the iterative local node exchange and corner sum evaluation are used to optimize the path smoothness and continuity; the path overlap is avoided through the cross detection and backtracking mechanism, and the executability and safety are improved; the buffer point is introduced for high corners, and the projection correction is carried out when the boundary is exceeded, so that the artificial or robot following is facilitated, and the picking efficiency is improved; finally, the path visualization and interactivity in the virtual reality are realized through coordinate mapping, and the problems of space modeling confusion, unreasonable path, cross confusion and unfriendly high corner in the traditional path planning are solved.

Description

A Virtual Reality-Based Warehouse Path Planning Method Technical Field

[0001] This invention relates to the field of intelligent path planning technology in virtual reality environments, specifically a warehouse path planning method based on virtual reality. Background Technology

[0002] With the intelligent and information-based development of modern logistics and warehousing, emerging technologies such as Virtual Reality (VR) have been gradually introduced into warehouse management and picking processes. Traditional warehouse path planning typically relies on two-dimensional maps and classic shortest path algorithms to achieve efficient movement of pickers or automated guided vehicles (AGVs) between goods points. However, existing path planning technologies generally focus on physical efficiency indicators such as shortest distance and least time, and give less consideration to the continuous visual experience of pickers in actual operation or virtual environments.

[0003] In recent years, the application of virtual reality (VR) technology in smart warehousing has been expanding. For example, pickers can use VR headsets to simulate warehouse routes in advance, improving training efficiency and operational safety. Path planning in a VR environment not only needs to ensure full coverage and accessibility of task points, but also needs to consider the visual smoothness and comfort of users in the virtual scene. However, in practical applications, existing shortest path-based paths often have many sharp turns, path intersections, or abrupt changes, causing pickers to frequently experience sudden changes in perspective, dizziness, or spatial disorientation in VR scenes, seriously affecting the immersive experience. In addition, most existing methods only focus on path optimization in physical space, lacking a systematic consideration of the characteristics of the VR visual field and human-computer interaction. There are still no mature and efficient solutions to key issues such as reducing corners in the virtual environment, buffering visual impact, and improving visual continuity while ensuring picking efficiency. Especially in scenarios with multiple target points, complex shelving layouts, and personnel training or auxiliary operations, how to achieve picking paths that both conform to warehousing business logic and are user-friendly for VR experiences has become a technical challenge that the industry urgently needs to overcome.

[0004] Therefore, this case aims to propose a warehouse path planning method based on virtual reality. While ensuring effective coverage of all picking task points, it effectively reduces sharp turns, abrupt changes and intersections in the path through systematic mathematical modeling and a self-developed path regularization algorithm, thereby achieving spatial smoothness and visual continuity of the path. Summary of the Invention

[0005] This invention provides a warehouse path planning method based on virtual reality, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a warehouse path planning method based on virtual reality, comprising:

[0007] S1. Establish a two-dimensional rectangular coordinate system on the floor of the rectangular warehouse, determine the warehouse boundary range, determine the plane coordinates and unique numbers of the picking start point and several picking target points, and form a set of path nodes including the start point and all target points.

[0008] S2. Starting from the starting point, select nodes step by step according to the Euclidean distance between each pair of nodes. Prioritize selecting the unvisited target point that is closest to the current end node. When there are nodes with the same distance, determine them in order of their planar coordinates and numbering to obtain the initial visiting order that covers all target points.

[0009] S3. Connect adjacent nodes in the initial access order into several path segments, and calculate the turning angle between each segment based on the directional relationship between adjacent path segments;

[0010] S4. The sum of the turning metrics of each adjacent path segment on the entire path is used as the main evaluation index, and can be combined with constraints such as path length and node arrival order to form the objective function.

[0011] S5. In the current access order, try to swap nodes at different positions one by one. If the angle cost after the swap is reduced, keep the swap and update the sequence. If the cost is not reduced, abandon the swap. Through multiple rounds of iteration, determine the termination condition based on the preset maximum number of iterations and the continuous unimproved threshold to obtain a converged optimization path.

[0012] S6. Perform line segment intersection detection on the path generated during the optimization process. When any two segments are detected to have a strict intersection, backtrack to the previous stable sequence and re-execute the local exchange optimization.

[0013] S7. For high corner positions that exceed the preset threshold, insert buffer points along the adjacent path segments before and after the corner. When the interpolation point exceeds the warehouse boundary, project it into the boundary range and update the node set and access order.

[0014] S8. Calculate the coordinate scaling relationship based on the width and height of the virtual scene, map the planar coordinates of each node in the final path to the virtual scene coordinates, and output the final path sequence including interpolation points.

[0015] Optionally, establishing a two-dimensional rectangular coordinate system on the rectangular warehouse floor to determine the warehouse boundary, the picking starting point, and the planar coordinates and unique numbers of several picking target points, forming a path node set containing the starting point and all target points, specifically includes:

[0016] Establish a two-dimensional rectangular coordinate system on the floor of the rectangular warehouse. Choose any vertex as the origin. The two sides connecting the rectangular warehouse floor to the origin are respectively taken as shaft and axis;

[0017] Get the length of the warehouse, denoted as Get the width of the warehouse, denoted as . ;

[0018] Construct warehouse coordinate domain ;

[0019] Let the starting number of the picker be... The coordinates in the coordinate system are ;in, Starting point The coordinates of the axis; Starting point The coordinates of the axis;

[0020] Obtain the total number of picking destinations, denoted as . Each target point is numbered as , Two-dimensional coordinates are ;in, The index is for enumeration and does not represent the order of access; it only indicates the order of access. The enumeration order when defining each storage location; For the first The number of each picking target point; For the first The coordinates of each picking target point;

[0021] Construct the complete set of path nodes as .

[0022] Optionally, the process begins with the starting point as the sequence and proceeds step-by-step based on the Euclidean distance between each pair of nodes, prioritizing the selection of unvisited target points that are closest to the current last node. When there are nodes with the same distance, the selection is made sequentially based on the planar coordinates and node numbers to obtain an initial visiting order that covers all target points. This process specifically includes:

[0023] Calculate any two points , Euclidean distance between ;in, and These are the numbers of the two path points;

[0024] Initialize path point number sequence ;

[0025] Initialize the set of unvisited points ;

[0026] right to Perform steps S201 to S205 sequentially:

[0027] in, For the path access sequence position;

[0028] S201. Let the number of the current path's end point be... ;

[0029] S202. Calculate the distance to all unvisited points: ;in, For accessing the location The set of identifiers of the remaining unvisited nodes after completion;

[0030] S203. Select the nearest point: ;in, To be selected as the access location The global node identifier;

[0031] If multiple minimum distances exist ,choose That is, according to coordinates First, then If they are still tied, then the one with the smallest node number is selected;

[0032] S204, Update Sequence ;

[0033] S205, Update Set ;

[0034] Final generation .

[0035] Optionally, the step of connecting adjacent nodes in the initial access order into several path segments and calculating the turning angle between each segment based on the directional relationship between adjacent path segments specifically includes:

[0036] Two consecutive points in the path , Construct segment vector Specifically:

[0037] ;in, For access location Arrive at the location The displacement vector;

[0038] like If the two points coincide, the segment is considered invalid, and the angle and interpolation calculations are skipped; where, It is a norm 2;

[0039] like and ,but:

[0040] ;in, For the first Duan Dao The corner of the segment;

[0041] Among them, dot product ; parameters Limited to the range Inside, specifically:

[0042] .

[0043] Optionally, the sum of the turning metrics of all adjacent path segments along the entire path is used as the main evaluation index, and can be combined with constraints such as path length and node arrival order to form an objective function, specifically including:

[0044] Calculate the total rotation angle ;

[0045] Minimize objective: Select path number sequence make Minimum, that is .

[0046] Optionally, the step of attempting pairwise swaps between nodes at different positions in the current access order, retaining the swap and updating the sequence when the angle cost after the swap decreases, and abandoning the swap when the cost does not decrease, through multiple rounds of iteration, determining the termination condition based on the preset maximum number of iterations and the threshold of continuous no improvement, and obtaining a converged optimization path, specifically includes:

[0047] Let the current path number sequence be denoted as The angle cost is ;in, For the first The path numbering sequence under round iteration; This represents the number of iterations. For the first Sum of path turning angles under round-by-round iteration;

[0048] For any , Path switching is performed sequentially using steps S501 to S502:

[0049] in, For the sequence of interchangeable pathpoint numbers;

[0050] S501, Execution Number Exchange: ; where, function Indicates will No. 1 in the sequence With the By swapping the position numbers, a new sequence is obtained. ;

[0051] S502, Calculate the cost of the new angle: ;

[0052] like Then receive the exchange ;like Then refuse and remain constant;

[0053] Let the maximum number of consecutive failures be the upper limit. The maximum number of iteration rounds is ;

[0054] If continuous Wheel not updated or maximum number of wheels reached Then it will terminate.

[0055] Optionally, the step of performing line segment intersection detection on the path generated during the optimization process, and when any two segments are detected to have a strict intersection, reverting to the previous stable sequence and re-performing the local exchange optimization, specifically includes:

[0056] Constructing the cross product function ;in, ; ;

[0057] Taking any two ends of the path, the endpoints of the first segment are... , The endpoint of the second segment is , ;

[0058] Path segments intersect if and only if: and ;

[0059] If the path segment intersection condition is met, then backtrack to the previous path. Then from the first Click to re-execute step S5.

[0060] Optionally, for high-angle locations exceeding a preset threshold, buffer points are inserted along adjacent path segments before and after the angle. When the interpolation point exceeds the warehouse boundary, it is projected onto the boundary range and the node set and access order are updated. Specifically, this includes:

[0061] Let the maximum allowable turning angle be ;

[0062] If a path segment exists If so, then insert two buffer points;

[0063] Let the first The length of the interpolation segment is ;

[0064] like or If , then no interpolation is performed;

[0065] Otherwise, set the first The interpolation point is , set the Post-interpolation point ;

[0066] like or Exceeding Then project it onto On the boundary:

[0067] Build ;in, For boundary projection functions;

[0068] right , Apply respectively And replace;

[0069] Build a new set of nodes ;

[0070] In the original sequence Insert between , forming a new sequence .

[0071] Optionally, the step of calculating the coordinate scaling relationship based on the width and height of the virtual scene, mapping the planar coordinates of each node in the final path to virtual scene coordinates, and outputting the final path sequence including interpolation points specifically includes:

[0072] Get Scene width, denoted as ; obtain Scene height, denoted as The mapping coefficients are: , ;in, for Direction mapping coefficient; for Direction mapping coefficient;

[0073] For any path point Obtain virtual coordinates: , ;

[0074] Output path sequence ;in, This represents the total number of nodes in the final path, including interpolation points.

[0075] The present invention has the following beneficial effects:

[0076] 1. A mathematical model of the actual warehouse space is constructed, establishing a unified two-dimensional coordinate system and effective boundary areas. All picking points and starting points are standardized into coordinate data. This design, with full warehouse space digitization and unique numbering throughout the entire process, differs from the approach of many traditional warehouses that only use rough map marking or local coding. By accurately collecting the coordinates of the starting point and each picking point, all subsequent path optimization and virtual mapping are based on the same benchmark, eliminating deviations and inconsistencies between actual operations and simulations. This not only improves the accuracy and reusability of the path algorithm but also lays the foundation for data integration between VR and the physical world. Compared to existing technologies, this greatly avoids navigation failures caused by waypoint confusion, numbering conflicts, or coordinate errors, supporting advanced applications such as warehouse automation and digital twins.

[0077] 2. Employing a spatial distance greedy algorithm, this method generates a fully covered, backtrack-free feasible initial path by progressively selecting the closest unvisited point to the current location, without relying on any traditional TSP (Traveling Salesman Problem) or classic path libraries. This method is simple, efficient, and adaptable to various warehouse layouts and target point distributions, significantly reducing the initial computation time for path generation and facilitating subsequent deep optimization. It not only ensures that picking paths do not involve redundant backtracking, improving operational efficiency, but also provides optimization space for further path smoothing and regularization. Compared to existing algorithms that only pursue the shortest path and ignore "prior rationality," this method offers both a good initial solution and flexibility and subsequent adaptability, making it suitable for dynamic picking and large-scale warehouse scenarios.

[0078] 3. Through rigorous vector operations, discrete path nodes are transformed into directed line segments, and the continuity of all path segments is further quantified by angle quantification. This approach differs from traditional picking path solutions that rely solely on node sequences without considering turning changes. By introducing numerically safe angle calculation and boundary correction mechanisms, the algorithm can automatically identify all sharp turns, abrupt changes, or discontinuities in the path, providing a precise data foundation for subsequent smoothing, interpolation, and other operations. This injects continuity constraints into VR path planning, effectively reducing visual "jumps," disorientation, and dizziness during virtual walking. Compared to existing technologies, this method no longer treats the path as a simple set of points but directly incorporates human visual experience and spatial curvature into the core of the algorithm, achieving dual optimization of technology and user experience.

[0079] 4. The "global path smoothness" is transformed into an optimizable total turning angle metric. Using the total turning angle as the cost function and the sole objective for all subsequent path adjustments, it overcomes the limitations of previous methods that relied solely on distance and steps. It proactively avoids paths that, while short in distance, involve sharp turns that degrade the user experience, prioritizing routes with spatial continuity and gentler turns. This cost function provides a theoretical basis and evaluation system for path optimization in VR, filling the technical gap in quantitatively evaluating experience continuity and significantly improving the usability and comfort of the final picking simulation.

[0080] 5. A path adjustment strategy combining local swapping and global search was developed. By systematically swapping and iteratively evaluating the order of any two picking points, it can escape the local extrema of a single greedy solution and find a solution with better corner smoothness. Its optimization process is strictly controlled by the number of failed rounds and the maximum number of steps, ensuring efficient convergence and preventing the path from getting bogged down in unnecessary adjustments. Compared to existing technologies, this method does not require global enumeration of all permutations (avoiding explosive computation) and can significantly improve path smoothness within a finite number of iterations, making it particularly suitable for dynamic, large-scale target point scenarios and saving significant time costs in practical deployment.

[0081] 6. The solution incorporates a geometric intersection detection and backtracking mechanism during path generation and optimization to ensure that the final output path is completely free of self-intersections. Compared to traditional methods that rely solely on distance and simple sorting, this step proactively detects and eliminates intersection problems that may arise from adjustments to point order. This effectively prevents abnormal situations such as path crossings and overlaps in VR virtual scenes, greatly improving the reliability of picking guidance and the feasibility of real-world operations. This measure avoids operational interference, visual confusion, and spatial chaos caused by intersections from the outset, filling the gaps in existing technologies regarding the continuity of three-dimensional space and the consistency of human-computer interaction.

[0082] 7. The solution innovatively incorporates a sharp turn detection and automatic buffer segment interpolation mechanism. Whenever a sharp turn occurs on the path exceeding a set threshold, the algorithm adaptively generates transition interpolation points based on the lengths of the path segments on both sides, thus significantly mitigating the angle change. This not only optimizes the visual continuity in VR scenes, preventing users from experiencing a sense of spatial discontinuity at sharp turns, but also ensures that the actual picking staff's walking movements are natural and smooth, reducing safety hazards. Compared to existing technologies that only correct node order and cannot dynamically buffer abrupt changes, this method achieves full-process spatial smoothing, making it particularly suitable for large-scale scenes, complex shelf layouts, and virtual-real collaborative training scenarios.

[0083] 8. A consistent mapping between the physical dimensions of the warehouse and the virtual reality space was adopted, achieving a lossless conversion of paths from reality to virtuality. All path points were strictly projected using scaling factors to ensure a one-to-one correspondence between the picking trajectory in the virtual environment and the actual warehouse. This ensured the bidirectional availability of virtual and physical data, providing a solid foundation for various scenarios such as warehouse digital twins, VR simulation, and autonomous driving. This mapping mechanism can adapt to warehouses of different scales and multiple VR platforms, greatly enhancing the system's universality and portability, eliminating the information gap between the real and virtual worlds, and providing theoretical and technical support for the deep integration of future intelligent logistics and human-computer interaction. Attached Figure Description

[0084] Figure 1 is a schematic diagram of the process of this invention. Detailed Implementation

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] Example: Referring to Figure 1, a warehouse path planning method based on virtual reality includes:

[0087] S1. Establish a two-dimensional rectangular coordinate system on the floor of the rectangular warehouse, determine the warehouse boundary range, determine the plane coordinates and unique numbers of the picking start point and several picking target points, and form a set of path nodes including the start point and all target points.

[0088] S2. Starting from the starting point, select nodes step by step according to the Euclidean distance between each pair of nodes. Prioritize selecting the unvisited target point that is closest to the current end node. When there are nodes with the same distance, determine them in order of their planar coordinates and numbering to obtain the initial visiting order that covers all target points.

[0089] S3. Connect adjacent nodes in the initial access order into several path segments, and calculate the turning angle between each segment based on the directional relationship between adjacent path segments;

[0090] S4. The sum of the turning metrics of each adjacent path segment on the entire path is used as the main evaluation index, and can be combined with constraints such as path length and node arrival order to form the objective function.

[0091] S5. In the current access order, try to swap nodes at different positions one by one. If the angle cost after the swap is reduced, keep the swap and update the sequence. If the cost is not reduced, abandon the swap. Through multiple rounds of iteration, determine the termination condition based on the preset maximum number of iterations and the continuous unimproved threshold to obtain a converged optimization path.

[0092] S6. Perform line segment intersection detection on the path generated during the optimization process. When any two segments are detected to have a strict intersection, backtrack to the previous stable sequence and re-execute the local exchange optimization.

[0093] S7. For high corner positions that exceed the preset threshold, insert buffer points along the adjacent path segments before and after the corner. When the interpolation point exceeds the warehouse boundary, project it into the boundary range and update the node set and access order.

[0094] S8. Calculate the coordinate scaling relationship based on the width and height of the virtual scene, map the planar coordinates of each node in the final path to the virtual scene coordinates, and output the final path sequence including interpolation points.

[0095] This paper proposes a complete virtual reality-based warehouse path planning method, covering the entire process from warehouse space modeling, waypoint determination, path generation, path optimization, intersection detection and correction, high-angle smoothing, to virtual reality output. By establishing a unified coordinate system on the warehouse floor, all picking points are uniquely numbered and located, ensuring data consistency and traceability. The path generation process employs the nearest neighbor principle, supplemented by node order and numbering rules, to obtain a preliminary reasonable picking route. Furthermore, through iterative local node exchange and angle sum evaluation, the smoothness and coherence of the path are continuously optimized. For intersections between path segments, intersection detection and automatic backtracking mechanisms effectively avoid path overlap during actual picking, improving the executability and safety of the path. For high-angle problems, the method introduces interpolation buffer points, making the path easier for humans or robots to follow and improving picking efficiency. Finally, the results are applied to the virtual reality environment through coordinate mapping, realizing path visualization and interactivity. In summary, this series of steps, when combined organically, solves the practical problems that are prone to occur in traditional warehouse path planning, such as chaotic spatial modeling, unreasonable paths, chaotic intersections, and unfriendly high-angle paths. It improves the smoothness and accuracy of path execution and the adaptability to virtual reality applications, and is an important technological foundation for the intelligent upgrading of warehousing and logistics.

[0096] The process involves establishing a two-dimensional Cartesian coordinate system on the rectangular warehouse floor to determine the warehouse boundary, the picking start point, and the planar coordinates and unique numbers of several picking target points, forming a path node set that includes the start point and all target points. Specifically, this includes:

[0097] Establish a two-dimensional rectangular coordinate system on the floor of the rectangular warehouse. Choose any vertex as the origin. The two sides connecting the rectangular warehouse floor to the origin are respectively taken as shaft and axis;

[0098] Get the length of the warehouse, denoted as Get the width of the warehouse, denoted as . ;

[0099] Construct warehouse coordinate domain ;

[0100] Unify the coordinate reference system for all path points and geometric operations, and define the boundary set of the walkable space. This provides a consistent physical reference (unit: meter) for subsequent distance, vector, projection and other calculations, and avoids path points falling into invalid areas;

[0101] Let the starting number of the picker be... The coordinates in the coordinate system are ;in, Starting point The coordinates of the axis; Starting point The coordinates of the axis; uniquely determine the starting node within the global coordinate system, serving as the 0th access position of the sequence, and providing the starting point for initial path construction and subsequent segment vector calculation;

[0102] Obtain the total number of picking destinations, denoted as . Each target point is numbered as , Two-dimensional coordinates are ;in, The index is for enumeration and does not represent the order of access; it only indicates the order of access. The enumeration order when defining each storage location; For the first The number of each picking target point; For the first The coordinates of each picking target point are used; all locations are structured in the form of coordinate points to form a set to be accessed, ensuring that each location can be uniquely indexed and calculated at both the geometric and sequence levels.

[0103] Construct the complete set of path nodes as Establish a unified data container to centrally manage the origin and cargo location, which will facilitate subsequent algorithms to filter, map and update at the set level.

[0104] The implementation methods for warehouse area modeling and path point initialization have been refined. By establishing a Cartesian coordinate system on the warehouse floor and selecting appropriate vertices as the origin, the physical space of the warehouse is abstracted into a unified data space. This step requires accurate acquisition of the actual length and width of the warehouse, and uses this to construct spatial boundaries, effectively preventing issues such as out-of-bounds errors and invalid paths in subsequent path planning. All path points—including picking start and destination points—are numbered and assigned coordinates, and managed uniformly in a single node set. This not only provides a clear data foundation for path calculation but also makes data retrieval and verification in each subsequent step more convenient and reliable. Through rigorous data initialization and physical boundary control, common problems in warehouse path algorithms, such as chaotic spatial data, inconsistent coordinate systems, and non-standard node management, are resolved. This ensures the accurate operation of each subsequent algorithm step, reduces the workload of algorithm development, system maintenance, and site adaptation, and improves the engineering feasibility and versatility of the solution, showing significant improvement compared to the drawbacks of existing technologies such as scattered data and non-standard spatial structures.

[0105] The process begins with the starting point as the sequence, and proceeds step-by-step by selecting nodes based on their Euclidean distances. Priority is given to selecting unvisited target points that are closest to the current last node. When distances are tied, the nodes are determined sequentially by their planar coordinates and node numbers to obtain an initial visiting order covering all target points. Specifically, this includes:

[0106] Calculate any two points , Euclidean distance between ;in, and These are the numbering of the two path points; they provide a unified geometric metric for calculating "nearest neighbor selection" and "path cost"; they ensure that the metric is differentiable and dimensionally consistent, facilitating integration with subsequent angle cost calculations.

[0107] Initialize path point number sequence ; Construct an expandable sequence using the starting point as the seed to ensure the consistency of the order of subsequent insertion operations;

[0108] Initialize the set of unvisited points Centralized identification of all nodes to be visited, supporting the selection of the best candidate from the remaining nodes in each round;

[0109] right to Perform steps S201 to S205 sequentially:

[0110] in, For the path access sequence position;

[0111] S201. Let the number of the current path's end point be... This provides a geometric reference point for the "next node selection" in this round, clarifying the starting point for measurement;

[0112] S202. Calculate the distance to all unvisited points: ;in, For accessing the location The set of remaining unvisited node identifiers is completed; the spatial cost from the current endpoint to all candidate nodes is quantified to provide a ranking basis for the "nearest point" decision;

[0113] S203. Select the nearest point: ;in, To be selected as the access location The global node identifier;

[0114] If multiple minimum distances exist ,choose That is, according to coordinates First, then If they are still tied, then the one with the smallest node number is selected;

[0115] Determine the location for this round of visits. The nodes form a coherent initial trajectory skeleton;

[0116] S204, Update Sequence ;

[0117] S205, Update Set ;

[0118] The decision results are written into the sequence and removed from the candidate set to ensure access uniqueness and progress.

[0119] Final generation Output a geometrically reasonable starting path that can be used as the starting path for subsequent "corner smoothing optimization".

[0120] This paper elaborates on the generation of the initial path sequence, clarifying the recursive generation method. Specifically, starting from the starting point, each time an unvisited point closest to the current end node is selected first. If equal distances are found, further filtering is performed based on coordinate size and node number, ensuring a scientific and orderly selection process. After each visit, the current path sequence and the set of unvisited points are updated to guarantee the uniqueness of subsequent nodes and the integrity of the path. Ultimately, a preliminary, coherent path skeleton covering all target points is obtained. This meticulous process effectively solves problems such as path discontinuity, node omissions or redundancy, and decision conflicts that may occur in traditional greedy algorithms. Simultaneously, this method provides a reasonable starting point for subsequent angle smoothing and local optimization, avoiding convergence difficulties or unsatisfactory final results caused by disordered initialization. This makes the initial path generation both efficient and structurally sound, balancing practical path feasibility with subsequent optimization potential. It is particularly suitable for warehouse picking scenarios with a large number of target points and complex distributions, and is more scientific and practical than existing methods that only consider distance or manually set the order.

[0121] The step of connecting adjacent nodes in the initial access order into several path segments and calculating the turning angle between each segment based on the directional relationship between adjacent path segments specifically includes:

[0122] Two consecutive points in the path , Construct segment vector Specifically:

[0123] ;in, For access location Arrive at the location The displacement vector; transforms the discrete access order into geometric kinematic quantities (displacement vectors), providing basic variables for angle, buffer length, and cross detection;

[0124] like If the two points coincide, the segment is considered invalid, and the angle and interpolation calculations are skipped; where, It is a 2-norm; it avoids division by zero and meaningless angles; at the same time, it provides an operation target point for "clearing zero segments" for subsequent exchanges, improving the controllability of path quality;

[0125] like and ,but:

[0126] ;in, For the first Duan Dao The corner of the segment;

[0127] Among them, dot product ; parameters Limited to the range Inside, specifically:

[0128] ; Quantify the steering amplitude of two consecutive segments, for use The core indicators of visual continuity; Numerical protection to suppress floating-point errors. Input from outside the domain ensures robust computing.

[0129] This method focuses on calculating path segment vectors and turning angles. By converting each pair of consecutive access points into displacement vectors and then calculating the angle between the two vectors segment by segment, the algorithm quantitatively manages the turning changes of the path. For zero-distance segments and overlapping nodes, the algorithm automatically skips them to avoid invalid data interfering with the overall calculation, while ensuring the accuracy of subsequent optimization and visualization. Furthermore, the algorithm employs angle interval restrictions and numerical protection measures to effectively prevent computational instability caused by floating-point errors. Through these meticulous processing steps, common problems such as data anomalies, angle anomalies, and numerical inconsistencies in the transformation of discrete path sequences into kinematic trajectories are resolved. This ensures that the foundational data for subsequent algorithms such as path smoothing optimization and cross-detection remains valid and reliable, while also giving the path itself better continuity and navigability in physical space. Compared to traditional approaches that only consider path length or lack rigorous angle processing, this method provides a solid data foundation for further intelligent path optimization while ensuring the robustness of the algorithm.

[0130] The objective function is formed by using the sum of the turning metrics of all adjacent path segments along the entire path as the main evaluation index, and can be combined with constraints such as path length and node arrival order. Specifically, it includes:

[0131] Calculate the total rotation angle Local corners are accumulated into a global metric, which serves as the optimization target for the "visual smoothness" of the path, and the merits of different access sequences are evaluated uniformly.

[0132] Minimize objective: Select path number sequence make Minimum, that is The objective is clearly defined: select the sequence with the smallest total turning angle from all feasible access sequences to provide an evaluation benchmark for the exchange mechanism in step S5.

[0133] This paper proposes using the sum of the turning angles of adjacent segments in a path as the primary evaluation index, which can be combined with constraints such as path length and node order to construct the objective function. This method not only pursues the shortest path but also emphasizes the smoothness of the path and the operability of the movement. By accumulating all turning angles, the algorithm can quantitatively reflect the turning complexity of the path in actual operation, thereby guiding the optimization algorithm to select the most easily implemented solution from multiple possible sequences. This avoids the efficiency losses and safety hazards caused by high-frequency, large-scale turning, and also ensures the smooth movement of pickers or automated equipment in the warehouse. Compared with traditional techniques that only optimize distance, ignore turning angles, or rely solely on experience for adjustment, this method, through a clearly defined smoothness objective, makes the final path both efficient and easy to execute. It improves the rationality of picking routes and the operational experience, and is particularly suitable for real-world warehousing scenarios with high requirements for continuity, safety, and ease of operation.

[0134] The process involves attempting pairwise swaps between nodes at different positions in the current access order. If the angle cost of the swap decreases, the swap is retained and the sequence is updated; otherwise, the swap is abandoned. Through multiple iterations, a termination condition is determined based on a preset maximum number of iterations and a threshold of no continuous improvement, thereby obtaining a convergent optimization path. Specifically, this includes:

[0135] Let the current path number sequence be denoted as The angle cost is ;in, For the first The path numbering sequence under round iteration; This represents the number of iterations. For the first The sum of path turning angles in each iteration; in the iteration rounds Solidify the "path-indicator" pair so that each exchange only makes improvement judgments based on the "current baseline";

[0136] For any , Path switching is performed sequentially using steps S501 to S502:

[0137] in, For the sequence of interchangeable pathpoint numbers;

[0138] S501, Execution Number Exchange: ; where, function Indicates will No. 1 in the sequence With the By swapping the position numbers, a new sequence is obtained. It provides "local topological transformations" with minimal operational granularity, exploring the sequence neighborhood with low complexity and progressively approximating the low-complexity sequence. The solution;

[0139] S502, Calculate the cost of the new angle: ;

[0140] like Then receive the exchange ;like Then refuse and remain The objective function remains unchanged, ensuring it is monotonically non-increasing, avoiding ineffective oscillations, and improving the interpretability of convergence.

[0141] Let the maximum number of consecutive failures be the upper limit. The maximum number of iteration rounds is ;

[0142] If continuous Wheel not updated or maximum number of wheels reached If the problem size is not met, the algorithm terminates; the search upper limit and stopping condition are defined endogenously by the problem size to ensure that the algorithm will terminate and cover the upper bound of all single swap pairs.

[0143] The implementation method of the local exchange-based path optimization mechanism is clarified. Its core approach is to swap nodes at different positions within the current path pairwise at each iteration, using the total number of path turns as the criterion for accepting a new solution; only solutions that improve upon the optimization result are retained. A maximum number of iterations and a threshold for consecutive unimproved results are introduced to effectively prevent the algorithm from getting stuck in invalid loops or local maxima. This mechanism has advantages such as fast convergence speed, low implementation complexity, and controllable algorithm process. Through continuous trial and error and selection, it gradually approaches the optimal or near-optimal path, effectively avoiding the problems of traditional brute-force exhaustive search or complex genetic algorithms, which are difficult to implement and have slow convergence. It ensures both the high efficiency of the algorithm and that each swap brings actual improvement, enhancing the smoothness and rationality of the final path, and meeting the practical needs of real-time and batch automatic optimization in warehousing systems.

[0144] The process of performing line segment intersection detection on the paths generated during optimization, and when a strict intersection is detected between any two segments, reverting to the previous stable sequence and re-performing local exchange optimization, specifically includes:

[0145] Constructing the cross product function ;in, ; The relative orientation of two vectors is determined by the directed area sign, which serves as the basic operator for line segment crossing tests. It is simple to calculate and numerically stable.

[0146] Taking any two ends of the path, the endpoints of the first segment are... , The endpoint of the second segment is , ;

[0147] Path segments intersect if and only if: and Algebraic notation is used to replace geometric inference, enabling rapid screening of self-fertilization and avoiding... Overlapping and intersecting lines cause visual confusion and potential navigation ambiguity;

[0148] If the path segment intersection condition is met, then backtrack to the previous path. Then from the first Re-execute step S5; use a local strategy to eliminate intersections at specific points, making minimal modifications to preserve the existing smoothness.

[0149] This system focuses on an automated mechanism for path intersection detection and correction. Using methods such as vector cross products, it quickly determines whether strict intersections have occurred between path segments. If a problem is detected, it immediately reverts to the previous non-intersecting state and restarts local optimization. This design effectively solves the problems of self-intersection and overlap that easily occur in warehouse paths under high-density target point scenarios, avoiding operational chaos and navigation ambiguity during actual execution. Unlike existing methods that only focus on optimization objectives while ignoring path feasibility, this step provides a hard guarantee for the actual usability of the path results. It ensures that every planned path is truly reachable and free from intersection interference, effectively improving picking efficiency and system security, laying the foundation for the reliability and stability of actual deployment.

[0150] For high-angle locations exceeding a preset threshold, buffer points are inserted along adjacent path segments before and after the angle. When an interpolation point exceeds the warehouse boundary, it is projected onto the boundary range, and the node set and access order are updated. Specifically, this includes:

[0151] Let the maximum allowable turning angle be ;

[0152] If a path segment exists If so, insert two buffer points; intervene only at "statistically significant large angles" to avoid excessive smoothing leading to redundant nodes and lengthy walks;

[0153] Let the first The length of the interpolation segment is Ensure that the buffer segment does not exceed half the length of the adjacent shortest segment, which can significantly reduce the angle without changing the overall geometry too much;

[0154] like or If , then no interpolation is performed;

[0155] Otherwise, set the first The interpolation point is , set the Post-interpolation point Replacing a sharp angle with two smooth transitions reduces the change in steering speed at a single point and alleviates the impact. visual tremors and the risk of dizziness;

[0156] like or Exceeding Then project it onto On the boundary:

[0157] Build ;in, For boundary projection function; project out-of-bounds buffer points back to the feasible region boundary to ensure that the path is fully executable and consistent with the actual warehouse layout;

[0158] right , Apply respectively And replace;

[0159] Build a new set of nodes ;

[0160] In the original sequence Insert between , forming a new sequence Ensure that new nodes have a globally unique identifier and are embedded in the original sequence, forming a smooth path that can be directly rendered and executed.

[0161] A sophisticated interpolation mechanism for high-angle buffer sections is proposed. When a corner exceeding a preset threshold is encountered in the path, the system automatically inserts transition points before and after the critical corner. If the interpolation point exceeds the warehouse boundary, it is automatically projected onto the boundary area, and the node set and access sequence are updated synchronously. This reduces the frequency of large-angle turns while maintaining the main path structure, minimizing operational difficulties, equipment wear and tear, and even safety risks caused by sharp turns. The mechanism also reasonably limits the interpolation length, achieving smooth path transitions while avoiding node redundancy and path length. Compared to traditional methods that rely solely on optimizing the sequence to reduce corners or manually handling high-angle turns, this invention can automatically, dynamically, and accurately handle high-risk corners, improving path friendliness and safety.

[0162] The process involves calculating the coordinate scaling relationship based on the width and height of the virtual scene, mapping the planar coordinates of each node in the final path to virtual scene coordinates, and outputting the final path sequence including interpolation points. Specifically, this includes:

[0163] Get Scene width, denoted as ; obtain Scene height, denoted as The mapping coefficients are: , ;in, for Direction mapping coefficient; for Orientation mapping coefficient; establish a linear ratio from physical meters to virtual coordinate units to ensure that the graphic display is proportionally mapped to the actual size;

[0164] For any path point Obtain virtual coordinates: , Precisely project the physical path to Planar coordinates ensure consistency and interactivity between visualization and navigation guidance;

[0165] Output path sequence ;in, This represents the total number of nodes in the final path, including interpolation points; the output can be directly used... The engine renders the coordinate sequence with voice / arrow navigation, completing the entire link from "data - geometry - visualization".

[0166] This system achieves precise mapping from physical paths to virtual reality scenes. By acquiring the actual dimensions of the virtual scene and calculating appropriate coordinate scaling relationships, the spatial coordinates of each path node are linearly transformed into corresponding points in the virtual environment, outputting a final path sequence containing all interpolated points. This not only ensures the consistency between the path representation in the virtual reality system and the actual warehouse scene but also facilitates subsequent path visualization, virtual navigation, and human-computer interaction. This step directly solves the problems of difficulty in connecting real physical paths with digital virtual scenes, scale misalignment, or loss of detail. The path planning results can be directly called by the virtual reality engine, providing a precise data foundation for applications such as warehouse digital twins and picking training simulations. Compared to traditional manual mapping or coarse processing methods, this improves the system's automation level and interactive experience.

[0167] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0168] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A warehouse path planning method based on virtual reality, characterized in that, include: S1. Establish a two-dimensional rectangular coordinate system on the rectangular warehouse floor to determine the warehouse boundary, the picking start point, and the planar coordinates and unique numbers of several picking target points, forming a path node set containing the start point and all target points; S2. Starting from the start point, select nodes step by step according to the Euclidean distance between each pair of nodes, prioritizing the unvisited target point closest to the current end node. When there are nodes with the same distance, determine them sequentially according to the planar coordinate order and number order to obtain an initial access order covering all target points; S3. Connect adjacent nodes in the initial access order into several path segments, and calculate the turning angle between each segment based on the directional relationship between adjacent path segments; S4. Use the sum of the turning angles of all adjacent path segments on the entire path as an evaluation index to construct an objective function, and use minimizing the objective function as the path optimization objective; S5. In the current access order, for different positions... The nodes are swapped pairwise. If the angle cost of the swap is reduced, the swap is retained and the sequence is updated. If the cost is not reduced, the swap is abandoned. Through multiple rounds of iteration, the termination condition is determined based on the preset maximum number of iterations and the continuous unimproved threshold to obtain a converged optimized path. S6. Line segment intersection detection is performed on the path generated during the optimization process. When any two segments are detected to have a strict intersection, the process reverts to the previous stable sequence and re-executes the local swap optimization. S7. For high corner positions that exceed the preset threshold, buffer points are inserted along the adjacent path segments before and after the corner. When the interpolation point exceeds the warehouse boundary, it is projected into the boundary range and the node set and access order are updated. S8. The coordinate scaling relationship is calculated based on the width and height of the virtual scene. The planar coordinates of each node in the final path are mapped to the virtual scene coordinates, and the final path sequence including the interpolation points is output.

2. The warehouse path planning method based on virtual reality according to claim 1, characterized in that, The process of establishing a two-dimensional Cartesian coordinate system on the rectangular warehouse floor to determine the warehouse boundary, the picking start point, and the planar coordinates and unique numbers of several picking target points, forming a path node set containing the start point and all target points, specifically includes: establishing a two-dimensional Cartesian coordinate system on the rectangular warehouse floor. Choose any vertex as the origin. The two sides connecting the rectangular warehouse floor to the origin are respectively taken as shaft and Axis; Get the length of the warehouse, denoted as Get the width of the warehouse, denoted as . Construct the warehouse coordinate domain Let the starting number of the picker be... The coordinates in the coordinate system are ;in, Starting point The coordinates of the axis; Starting point The coordinates of the axis; obtain the total number of picking target points, denoted as . Each target point is numbered as ， Two-dimensional coordinates are ;in, The index is for enumeration and does not represent the order of access; it only indicates the order of access. The enumeration order when defining each storage location; For the first The number of each picking target point; For the first The coordinates of each picking target point; construct the complete set of path nodes as follows: 。 3. The warehouse path planning method based on virtual reality according to claim 2, characterized in that, The process begins with the starting point as the sequence and proceeds step-by-step by selecting nodes based on their pairwise Euclidean distances. Priority is given to selecting unvisited target points that are closest to the current last node. When distances are tied, the nodes are determined sequentially by their planar coordinates and node numbers to obtain an initial visiting order covering all target points. Specifically, this includes calculating the distances between any two points... 、 Euclidean distance between ;in, and These are the numberings of the two path points; initialize the path point numbering sequence. Initialize the set of unvisited points. ;right to Steps S201 to S205 are executed sequentially: where, S201. Let the current path end point number be... S202. Calculate the distance to all unvisited points: ;in, For accessing the location After completion, identify the remaining unvisited nodes; S203, select the nearest point: ;in, To be selected as the access location The global node identifier; if multiple minimum distances exist ,choose That is, according to coordinates First, then If they are still tied, select the one with the smallest node number; S204, Update the sequence. S205, Update Set ; final generation 。 4. The warehouse path planning method based on virtual reality according to claim 3, characterized in that, The process of connecting adjacent nodes in the initial access order into several path segments and calculating the turning angle between each segment based on the directional relationship between adjacent path segments specifically includes: two consecutive points in the path 、 Construct segment vector Specifically: ;in, For access location Arrive at the location The displacement vector; if If the two points coincide, the segment is considered invalid, and the angle and interpolation calculations are skipped; where, It is a norm 2; if and ,but: ;in, For the first Duan Dao The corner of the segment; where the dot product ； parameters Limited to the range Inside, specifically: 。 5. The warehouse path planning method based on virtual reality according to claim 4, characterized in that, The objective function is constructed using the sum of the turning angles of all adjacent path segments along the entire path as the evaluation index, and minimizing the objective function is used as the path optimization objective. Specifically, this includes: calculating the total turning angles. Minimize objective: Select path number sequence make Minimum, that is 。 6. The warehouse path planning method based on virtual reality according to claim 5, characterized in that, The process involves attempting pairwise swaps between nodes at different positions in the current access order. If the angle cost of the swap decreases, the swap is retained and the sequence is updated; otherwise, the swap is abandoned. Through multiple iterations, a termination condition is determined based on a preset maximum number of iterations and a threshold for continuous unimproved paths, resulting in a converged optimized path. Specifically, this includes: recording the current path number sequence as... The angle cost is ;in, For the first The path numbering sequence under round iteration; This represents the number of iterations. For the first Sum of path turns in round iterations; for any 、 Path switching is performed sequentially from steps S501 to S502: Position the interchangeable pathpoint number sequence; S501, Perform number swapping: ; where, function Indicates will In the hierarchy With the By swapping the position numbers, a new sequence is obtained. S502, Calculate the cost of the new angle: ;like Then receive the exchange ;like Then refuse and remain Unchanged; let the maximum number of consecutive failures be capped at 1. The maximum number of iteration rounds is If continuous Wheel not updated or maximum number of wheels reached Then it will terminate.

7. A warehouse path planning method based on virtual reality according to claim 6, characterized in that, The optimization process involves performing line segment intersection detection on the generated paths. When any two segments are detected to have a strict intersection, the process reverts to the previous stable sequence and re-executes the local exchange optimization. Specifically, this includes constructing a vector cross product function. ;in, ； Take any two ends of the path, the endpoints of the first segment are... 、 The endpoint of the second segment is 、 Path segments intersect if and only if: and If the path segment intersection condition is met, then backtrack to the previous path. Then from the first Click to re-execute step S5.

8. The warehouse path planning method based on virtual reality according to claim 7, characterized in that, For high-angle locations exceeding a preset threshold, buffer points are inserted along adjacent path segments before and after the angle. When an interpolated point exceeds the warehouse boundary, it is projected onto the boundary range, and the node set and access order are updated. Specifically, this includes setting the maximum allowed angle as... If a path segment exists Then insert two buffer points; let the first one be... The length of the interpolation segment is ;like or If the condition is met, then no interpolation is performed; otherwise, the first value is set. The interpolation point is , set the Post-interpolation point ;like or Exceeding Then project it onto On the boundary: Construction ;in, For boundary projection functions; ， Apply respectively Replace; construct a new set of nodes. In the original sequence Insert between , forming a new sequence 。 9. A warehouse path planning method based on virtual reality according to claim 8, characterized in that, The process involves calculating the coordinate scaling relationship based on the width and height of the virtual scene, mapping the planar coordinates of each node in the final path to virtual scene coordinates, and outputting the final path sequence including interpolation points. Specifically, this includes: obtaining... Scene width, denoted as ; obtain Scene height, denoted as The mapping coefficients are: ， ;in, for Direction mapping coefficient; for Direction mapping coefficients; for any path point Obtain virtual coordinates: ， Output path sequence ;in, This represents the total number of nodes in the final path, including interpolation points.

Images