Warehouse path planning method based on virtual reality

Abstract

The invention 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 a warehouse, sorting points are uniquely numbered and positioned, and data consistency and traceability are ensured; the path generation adopts a nearest neighbor principle and combines a node sequence and a numbering rule to obtain a preliminary reasonable path; evaluating the smoothness and coherence of the optimized path by using iterative local node exchange and the sum of corners; path overlapping is avoided through cross detection and a fallback mechanism, and the performability and safety are improved; a buffer point is introduced for a high rotation angle and projection correction is carried out when a boundary is exceeded, so that manual or robot following is facilitated and the picking efficiency is improved; and finally, visualization and interactivity of the path in virtual reality are realized through coordinate mapping, and the problems of disordered spatial modeling, unreasonable path, disordered crossing and unfriendly high corner in traditional path planning are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent path planning in virtual reality environment, in particular to a warehouse path planning method based on virtual reality. BACKGROUND

[0002] With the intelligentization and informatization development of modern logistics and warehousing industry, emerging technologies such as virtual reality (VR) have been gradually introduced into warehouse management and picking operation processes. Traditional warehouse path planning usually 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 techniques generally focus on physical efficiency indicators such as shortest distance and least time, and less consider the continuous visual experience of pickers in actual operation or virtual environment.

[0003] In recent years, virtual reality technology has been increasingly applied in intelligent warehousing. For example, pickers can simulate warehouse routes in advance through VR headsets to improve training efficiency and operational safety. Path planning in a virtual reality environment not only needs to ensure full coverage and accessibility of task points, but also needs to consider visual fluency and experience comfort in the virtual scene. However, in actual applications, existing shortest path-based paths often have many sharp turns, path intersections or mutations, etc., which cause pickers to frequently experience visual angle mutations, dizziness or spatial disorientation in the VR scene, seriously affecting immersive experience. In addition, existing methods mostly only optimize paths in physical space, lacking systematic consideration of virtual reality visual field characteristics and human-computer interaction behaviors. For how to ensure picking efficiency while reducing virtual environment corner turns, buffer visual field impact, and improving visual continuity, there is no mature and efficient solution. Especially in multi-target point, complex shelf layout, personnel training or auxiliary operation scenarios, how to achieve a picking path that not only meets the warehouse business logic but also is friendly to virtual reality experience has become a technical problem that the industry urgently needs to break through.

[0004] Therefore, the present application aims to provide a warehouse path planning method based on virtual reality, which effectively reduces sharp turns, mutations and intersections in the path, realizes spatial smoothing and visual continuity of the path, on the basis of ensuring effective coverage of all picking task points through systematic mathematical modeling and self-created path regularization algorithm. SUMMARY

[0005] The present application provides a warehouse path planning method based on virtual reality, which solves the problems mentioned in the background.

[0006] The present application provides the following technical solutions: a warehouse path planning method based on virtual reality, comprising:

[0007] S1, establish a two-dimensional rectangular coordinate system on the ground of a rectangular warehouse, determine the boundary range of the warehouse, determine the planar coordinates and unique numbers of a picking starting point and a plurality of picking target points, and form a path node set containing the starting point and all target points;

[0008] S2, sequentially starting from the starting point, gradually selecting according to the Euclidean distance between two nodes, preferentially selecting the unvisited target point closest to the current last node, and when there are distance ties, sequentially determining according to the planar coordinate order and the number order of the nodes to obtain an initial visiting order covering all target points;

[0009] S3, connecting adjacent nodes in the initial visiting order into a plurality of path segments, and calculating the turning angles between the path segments based on the directional relationship of adjacent path segments;

[0010] S4, taking the sum of the turning angles of all adjacent path segments in the entire path as the main evaluation index, and combining path length, node reaching order and other constraints to form a target function;

[0011] S5, attempting to exchange nodes at different positions in the current visiting order two by two, retaining the exchange and updating the sequence when the angle cost after the exchange is reduced, and abandoning the exchange when the cost is not reduced, obtaining a converged optimized path by a plurality of iterations, and determining a termination condition according to a preset maximum number of iterations and a continuous non-improvement threshold;

[0012] S6, performing line segment intersection detection on the path generated in the optimization process, and when any two segments are detected to strictly cross, reverting to the previous stable sequence and re-executing local exchange optimization;

[0013] S7, for high turning angle positions exceeding a preset threshold, inserting buffer points along adjacent path segments before and after the turning angle, and when the interpolation points exceed the boundary of the warehouse, projecting them into the boundary range and updating the node set and the visiting order;

[0014] S8, calculating the coordinate scaling relationship according to 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 containing the interpolation points.

[0015] Optionally, the establishment of a two-dimensional rectangular coordinate system on the ground of a rectangular warehouse, the determination of the boundary range of the warehouse, the determination of the planar coordinates and unique numbers of a picking starting point and a plurality of picking target points, and the formation of a path node set containing the starting point and all target points specifically include:

[0016] establishing a two-dimensional rectangular coordinate system on the ground of a rectangular warehouse , selecting any one vertex as the origin , connecting the two edges of the rectangular warehouse ground and the origin as the axis and the 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] wherein, is the path access sequence bit sequence;

[0028] S201, set the current path end point number as ;

[0029] S202, calculate the distance from all unvisited points: ; wherein, is the set of unvisited node identifiers remaining after the access position is completed;

[0030] S203, select the nearest point: ; wherein, is the global node identifier selected as the access position ;

[0031] If there are multiple minimum distances , select , i.e., according to the coordinates first, then , if still parallel, select the node number minimum;

[0032] S204, update the sequence ;

[0033] S205, update the set ;

[0034] Finally generate .

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

[0036] The two points , in the path constitute a segment vector , specifically:

[0037] ; wherein, is the displacement vector from the access position to the position ;

[0038] If , the two points coincide, set the segment as an invalid segment, and skip the included angle and interpolation calculation; wherein, is the two-norm;

[0039] If and , then:

[0040] ; wherein, 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 new angle cost: ;

[0052] If , receive exchange ; if , reject and keep unchanged;

[0053] Set the maximum number of consecutive failures to , and the maximum number of iterations to ;

[0054] If it is not updated for consecutive rounds or reaches the maximum number of rounds , terminate.

[0055] Optionally, the path generated in the optimization process is subjected to line segment intersection detection, and when any two segments are detected to strictly intersect, the previous stable sequence is rolled back and local exchange optimization is re-executed, specifically including:

[0056] Construct a vector cross product function ; wherein ; ;

[0057] Take any two paths, the end points of the first segment are , , and the end points of the second segment are , ;

[0058] The path segments intersect if and only if: and ;

[0059] If the path segment intersection condition is met, the path is rolled back to , and step S5 is re-executed from the point.

[0060] Optionally, for high corner positions exceeding a preset threshold, buffer points are inserted along adjacent path segments before and after the corner, and when the interpolation point exceeds the warehouse boundary, it is projected into the boundary range and the node set and access order are updated, specifically including:

[0061] Set the maximum allowed corner to ;

[0062] If there is a path segment , two buffer points are inserted;

[0063] Set the interpolation segment length at the th to ;

[0064] If or , do not interpolate;

[0065] Otherwise, set the first interpolation point before as and the last interpolation point after as

[0066] If or , project it to the boundary: ;

[0067] Construct ; where is the boundary projection function;

[0068] For , , apply and replace, respectively;

[0069] Construct a new node set ;

[0070] Between the original sequence , insert to form a new sequence .

[0071] Optionally, the coordinate scaling relationship is calculated according to the width and height of the virtual scene, the plane coordinates of each node in the final path are mapped into virtual scene coordinates, and the final path sequence containing interpolation points is output, specifically including:

[0072] Get the width of the scene, denoted as ; get the height of the scene, denoted as ; then the mapping coefficient is: , ; where is the direction mapping coefficient; is the direction mapping coefficient;

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

[0074] Output the path sequence ; where is the total number of final path nodes, including interpolation points.

[0075] ​The present application has the following beneficial effects:

[0076] 1. The actual warehouse space is mathematically modeled, a unified two-dimensional coordinate system and an effective boundary area are constructed, and all picking points and starting points are standardized as coordinate data. This full-warehouse-space digitization and full-process unique numbering design is different from the rough map marking or local coding method of many traditional warehouses. Through accurate collection of the coordinates of the starting point and each picking point, all subsequent path optimization and virtual mapping are based on the same reference, eliminating the deviation and inconsistency between actual operation and simulation. Not only does it improve the accuracy and reusability of the path algorithm, but it also lays the foundation for data docking between VR and the physical world. Compared with the prior art, it greatly avoids navigation failures caused by path point confusion, numbering conflicts or coordinate errors, and supports automation, digital twin and other high-level applications of the warehouse.

[0077] 2. The spatial distance greedy algorithm is adopted, without relying on any traditional TSP (Traveling Salesman Problem) or classic path library, a feasible initial path is quickly generated by gradually selecting the nearest unvisited point to the current location. This method has a simple structure and is efficient, and can adapt to various warehouse layouts and target point distributions, greatly reducing the pre-computation time of path generation, facilitating subsequent deep optimization. Not only does it ensure that the picking path does not have redundant returns, improving operational efficiency, but it also provides optimization space for further smoothing and regularization of the path. Compared with existing algorithms that only pursue the shortest path and ignore "a priori rationality", this method has a good initial solution, and also considers flexibility and subsequent plasticity, suitable for dynamic picking and large-scale warehouse scenarios.

[0078] 3. Through rigorous vector operations, discrete path nodes are converted into directed line segments, and further angle quantization is performed on the continuity of all path segments. This approach is different from traditional picking path solutions that only rely on node sequences without considering turning changes. By introducing a numerically safe angle calculation and boundary correction mechanism, the algorithm can automatically identify all sharp turns, sudden changes or discontinuities in the path, and provide accurate data for subsequent smoothing, interpolation and other operations. It injects continuity constraints into VR path planning, effectively reducing visual "jumps", directional confusion and dizziness during virtual walking. Compared with the prior art, this method no longer treats the path as a simple point set, but directly introduces human eye experience and spatial curvature into the core of the algorithm, achieving dual optimization of technology and experience.

[0079] 4、 The "global path smoothness" is converted into an optimized corner sum index. The total corner is used as the cost function, which is the only goal for all subsequent path adjustments, getting rid of the limitations of previous methods that only use distance and step count as the standard. It can actively avoid paths that are short in distance but have large sharp turns, resulting in a decline in user experience, and preferentially generate routes that are spatially continuous and have smooth turns. The introduction of this cost function provides a theoretical basis and evaluation system for path optimization in VR, filling the technical gap of existing methods that cannot quantitatively evaluate the continuity of experience, and significantly improving the usability and comfort of the final picking simulation.

[0080] 5、 A path adjustment strategy based on local exchange and global search is created. Through systematic exchange and iterative optimization of the order of any two picking points, the local extremum of the single greedy solution can be jumped out, and a better solution in terms of corner smoothness can be found. The optimization process can be strictly controlled by the number of failed rounds and the maximum number of steps, which not only ensures efficient convergence of the algorithm, but also prevents the path from being trapped in unnecessary repeated adjustments. Compared with existing technologies, this method does not need to enumerate all permutations globally (avoiding explosive calculations), and can significantly improve the smoothness of the path within a limited number of iterations, especially suitable for dynamic large-scale target point scenarios, saving a lot of time and cost for actual deployment.

[0081] 6、 The scheme specially designs a geometric intersection judgment and backtracking mechanism in the process of path generation and optimization, ensuring that the final output path is completely non-intersecting. Compared with the traditional method which only relies on distance and simple sorting, this step can actively detect and eliminate the intersection problem that may be caused by the adjustment of point order. It effectively prevents path crossing, overlapping and other abnormal situations in the VR virtual scene, greatly improving the reliability of picking guidance and the feasibility of real operation. This measure avoids the interference, visual disturbance and spatial confusion caused by intersection from the source, filling the gap in the continuity of three-dimensional space and the consistency of human-computer interaction in existing technologies.

[0082] 7、 The scheme innovatively sets up an acute corner detection and automatic buffer segment interpolation mechanism. Whenever the path has an acute turn greater than the set threshold, the algorithm will generate a transition interpolation point according to the length of the path segment on both sides, thereby significantly smoothing the corner change. Not only does it optimize the visual continuity in the VR scene, preventing users from experiencing spatial discontinuity at sharp turns, but it also ensures the natural and smooth movement of actual pickers, reducing safety hazards. Compared with existing technologies that only correct node order and cannot dynamically buffer changes, this method realizes spatial smoothing throughout the entire process, especially suitable for large scenes, complex shelf layouts, and virtual-real collaborative training scenarios.

[0083] 8、 The warehouse physical size is mapped with the consistency of virtual reality space, realizing the lossless conversion of the path from reality to virtual. All path points are strictly projected through the scale factor, ensuring that the picking trajectory in the virtual environment corresponds to the actual warehouse one by one. It ensures the bidirectional availability of virtual and real data, and provides a solid foundation for various scenarios such as warehouse digital twin, VR simulation and automatic driving. The mapping mechanism can adapt to warehouses of different scales and various VR platforms, greatly enhancing the universality and portability of the system, 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. BRIEF DESCRIPTION OF DRAWINGS

[0084] Figure 1 The flowchart of the present application is shown. DETAILED DESCRIPTION

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

[0086] Embodiment, refer to Figure 1 A warehouse path planning method based on virtual reality, comprising:

[0087] S1, a two-dimensional rectangular coordinate system is established on the ground of a rectangular warehouse, the boundary range of the warehouse is determined, the planar coordinates and unique numbers of a picking starting point and a plurality of picking target points are determined, and a path node set containing the starting point and all target points is formed;

[0088] S2, taking the starting point as the sequence starting point, selecting step by step according to the Euclidean distance between two nodes, preferentially selecting the unvisited target point closest to the current end node, and when there are distance ties, determining in turn according to the planar coordinate order and the number order of the nodes, to obtain an initial visiting order covering all target points;

[0089] S3, connecting adjacent nodes in the initial visiting order into a plurality of path segments, and calculating the turning angles between the path segments based on the directional relationship of adjacent path segments;

[0090] S4, taking the sum of the turning angles of each adjacent path segment in the whole path as the main evaluation index, and combining with the path length, node reaching order and other constraints to form a target function;

[0091] S5, in the current access order, the nodes at different positions are exchanged two by two, when the angle cost after the exchange is reduced, the exchange is kept and the sequence is updated, when the cost is not reduced, the exchange is abandoned, through multiple iterations, the maximum iteration number and the continuous improvement threshold are determined as the termination condition, and a convergent optimized path is obtained;

[0092] S6, the path generated in the optimization process is detected for line segment intersection, when any two segments are detected to strictly cross, the previous stable sequence is returned to and the local exchange optimization is re-executed;

[0093] S7, for high corner positions exceeding a preset threshold, buffer points are inserted along 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;

[0094] S8, the coordinate scaling relationship is calculated according to the width and height of the virtual scene, the plane coordinates of each node in the final path are mapped into virtual scene coordinates, and the final path sequence including interpolation points is output.

[0095] A complete warehouse path planning method based on virtual reality is proposed, which covers the whole process from warehouse space modeling, path point determination, path generation, path optimization, intersection judgment and correction, high corner smoothing to virtual reality output. By establishing a unified coordinate system on the ground of the warehouse, all picking points are uniquely numbered and located, ensuring the consistency and traceability of the data. In the path generation process, the nearest neighbor principle is adopted, supplemented by node order and numbering rules, to obtain a preliminary reasonable picking route. Further, through iterative local node exchange and angle sum evaluation, the smoothness and continuity of the path are continuously optimized. For the intersection between path segments, through intersection detection and automatic backtracking mechanism, the problem of path overlap in actual picking is effectively avoided, improving the executability and safety of the path. For the high corner problem, the method introduces interpolation buffer points, which not only makes the path easier to follow for humans or robots, but also improves the picking efficiency. Finally, through coordinate mapping, the results are used in virtual reality environment to realize the visualization and interactivity of the path. In summary, this series of steps are organically combined, solving the real problems of space modeling confusion, unreasonable path, intersection confusion and high corner unfriendliness in traditional warehouse path planning, improving the smoothness, accuracy of path execution and adaptability of virtual reality application, and is an important technical basis for the intelligent upgrading of warehouse logistics.

[0096] The two-dimensional rectangular coordinate system is established on the ground of the rectangular warehouse, the boundary range of the warehouse is determined, the plane coordinates and unique numbers of the picking starting point and a plurality of picking target points are determined, and a path node set containing the starting point and all target points is formed, which specifically includes:

[0097] Establish a two-dimensional rectangular coordinate system on the ground of the rectangular warehouse , select an arbitrary vertex as the origin , and connect the two edges of the rectangular warehouse ground with the origin as the axis and the axis respectively;

[0098] Obtain the length of the warehouse, denoted as , and the width of the warehouse, denoted as ;

[0099] Construct the warehouse coordinate domain ;

[0100] Unify the coordinate reference system of all path points and geometric operations, and clearly define the boundary set of the walkable space , provide a consistent physical basis (unit: meters) for subsequent distance, vector, and projection operations, and avoid path points falling into invalid areas;

[0101] Let the starting point number of the picker be , and the coordinate in the coordinate system be ; where is the coordinate of the starting point on the axis; is the coordinate of the starting point on the axis; uniquely determine the departure node in the global coordinate, as the 0th access position of the sequence, provide the starting point for initial path construction and subsequent segment vector calculation;

[0102] Obtain the total number of picking target points, denoted as , and the number of each target point as , , and the two-dimensional coordinate as ; where is the enumeration index, which does not represent the access order, but only indicates the enumeration order when defining the th storage location; is the number of the th picking target point; is the coordinate of the th picking target point; structure all storage locations in the form of coordinate points to form a set of to-be-visited, ensuring that each storage location can be uniquely indexed and calculated in both geometric and sequential aspects;

[0103] Construct the path node full set as ; establish a unified data container to centrally manage the starting point and storage points, facilitating subsequent algorithms to perform filtering, mapping, and updating at the set level.

[0104] The implementation of warehouse area modeling and path point initialization is refined. By establishing a rectangular coordinate system on the warehouse floor and selecting appropriate vertices as the origin, the warehouse physical space is abstracted into a unified data space. This step requires accurate acquisition of the actual length and width of the warehouse, and the space boundary is constructed based on this, effectively preventing subsequent path planning from exceeding the boundary, being invalid, and other problems. All path points, including the picking starting point and target point, are numbered and assigned coordinates, and are managed uniformly in a node set. This not only provides a clear data basis for path calculation, but also makes subsequent data calling and verification of each step more convenient and reliable. Through rigorous data initialization and physical boundary control, common problems in warehouse path algorithm such as spatial data confusion, non-uniform coordinate system, and non-standard node management are solved. This ensures the accurate operation of each subsequent algorithm, reduces the workload of algorithm development, system maintenance, and site adaptation, improves the engineering feasibility and generality of the scheme, and significantly improves the disadvantages of scattered data and non-standard space in existing technologies.

[0105] The starting point is selected as the sequence start, and the Euclidean distance between two nodes is selected step by step, and the nearest unvisited target point to the current end node is selected first. When there are distance ties, the node plane coordinates and number order are determined in turn to obtain the initial access order covering all target points, which specifically includes:

[0106] The Euclidean distance between any two points 、 is calculated ; wherein, and are the numbers of two path points; a unified geometric measure is provided for "neighbor selection" and "path cost" calculation; the measure is differentiable and dimensionless, which is convenient for subsequent angle cost docking;

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

[0108] Initialize the unvisited point set ; identify all nodes to be visited, and support selection of the optimal candidate from the remaining nodes in each round;

[0109] Steps S201 to S205 are performed in turn for to :

[0110] wherein, is the path access sequence bit sequence;

[0111] S201, set the current path end point number as ; provide geometric reference points for the "next node selection" in this round, and clarify the starting point of measurement;

[0112] S202, calculate the distance from all unvisited points: ; Wherein, is the unvisited point set after the current round of "next node selection"; After completion, the remaining unvisited node identification set; Quantify the spatial cost of the current end point to all candidate points, provide the basis for "nearest point" decision-making;

[0113] S203, select the nearest point: ; Wherein, is the global node identification of the selected access location ;

[0114] If there are multiple minimum distances , select , that is, according to the coordinates first, then , if still parallel, select the node number minimum;

[0115] Determine the node of the access location in this round, form a coherent initial trajectory skeleton;

[0116] S204, update sequence ;

[0117] S205, update set ;

[0118] Write the decision result into the sequence and eliminate it from the candidate set to ensure the uniqueness of access and the advancement of progress;

[0119] Finally generate ; output a geometrically reasonable path that can be used as the starting path for subsequent "corner smoothing optimization".

[0120] The generation of the initial path sequence is described in detail, and the recursive generation method of the path is determined. The specific method is to start from the starting point, and each time the nearest unvisited point to the current end node is selected first. If the distance is equal, further selection is made according to the size and number of the coordinates in order to ensure that the selection process is scientific and orderly. After each visit, the current path sequence and the unvisited point set are updated to ensure the uniqueness of the subsequent nodes and the integrity of the path. Finally, a preliminary coherent path skeleton covering all target points is obtained. Through the above detailed process, the problems of non-coherent path, node omission or redundancy, decision conflict and other problems in the traditional greedy method are effectively solved. At the same time, this method provides a reasonable starting point for subsequent angle smoothing and local optimization, avoiding the convergence difficulty or unsatisfactory final result caused by unordered initialization. The initial path generation is efficient and reasonable in structure, taking into account the actual path feasibility and subsequent optimization potential, especially suitable for warehouse picking scenarios with a large number of target points and complex distribution. Compared with the existing methods that only consider distance or manually set order, this method is more scientific and practical.

[0121] The connecting of adjacent nodes in the initial access sequence into several path segments, and the calculation of the turning angles between adjacent path segments based on the directional relationship of the adjacent path segments, specifically includes:

[0122] The two points in the path , The segment vector is composed of , specifically:

[0123] ; wherein, is the displacement vector from the access position to the position ; converting the discrete access sequence into a geometric kinematic quantity (displacement vector) provides a basic variable for angle, buffer length and intersection detection;

[0124] If , the two points coincide, and the segment is set as an invalid segment, and the included angle and interpolation calculation are skipped; wherein, is the two-norm; to avoid division by zero and meaningless angles; at the same time, it provides a target for "clearing zero segments" for subsequent exchange, improving the controllability of path quality;

[0125] If and , then:

[0126] ; wherein, is the turning angle from the th segment to the th segment;

[0127] wherein, the dot product ; The parameters defined in the interval , specifically:

[0128] ; quantifying the turning amplitude of consecutive segments, used as the core indicator of visual continuity; numerical protection, suppressing the out-of-domain input caused by floating-point error, ensuring robust calculation.

[0129] The algorithm focuses on the calculation method of path segment vectors and turning angles. By converting each pair of consecutive access points into displacement vectors, and then calculating the included angle between the two vectors segment by segment, the turning changes of the path are quantitatively managed. For the case of zero distance segment and node coincidence, the algorithm will automatically skip, avoiding the interference of invalid data on the overall calculation, while ensuring the accuracy of subsequent optimization and visualization. In addition, the algorithm also adopts the included angle interval restriction and numerical protection measures, effectively preventing the calculation instability problem caused by floating-point error. Through these detailed processing steps, the common data abnormalities, angle abnormalities and numerical inconsistencies in the process of converting path discrete sequence into kinematic trajectory are solved. On the one hand, it ensures that the basic data of subsequent algorithms such as path smooth optimization and intersection detection is always effective and reliable, and on the other hand, it makes the path itself have better continuity and navigability in the physical space. Compared with the traditional method which only considers the path length or does not do rigorous angle processing, this method ensures the robustness of the algorithm while providing a solid data foundation for further intelligent optimization of the path.

[0130] The sum of the turning measurements of each adjacent path segment in the entire path is taken as the main evaluation indicator, and it can be combined with path length, node arrival order and other constraints to form a target function, which specifically includes:

[0131] calculating the total turning angle ; accumulating local turning angles into a global indicator, which is used as the optimization target of path "visual smoothness", and evaluates the pros and cons of different access sequences;

[0132] Minimization target: selecting path number sequence to make minimum, that is ; explicitly solving the target - selecting the sequence with "minimum total turning angle" among all feasible access sequences, providing 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 a new angle cost: ;

[0140] If , receive exchange ; if , reject and keep unchanged; ensure that the objective function is monotonically non-increasing, avoid invalid oscillation, and improve convergence interpretability;

[0141] Let the upper limit of the maximum number of consecutive failures be , and the maximum iteration number be ;

[0142] If it is not updated for consecutive rounds or the maximum number of rounds is reached, terminate; endogenous definition of search upper limit and stopping condition according to problem size, to ensure that the algorithm will terminate and cover the upper bound of all one-exchange pairs.

[0143] The implementation of the local exchange type path optimization mechanism is clarified. The core approach is: each time, only the nodes at different positions in the current path are exchanged two by two, and the total sum of path corners is used as the standard for accepting new solutions. Only when the optimization result is improved, it is retained. The maximum number of iterations and the continuous improvement threshold are introduced to effectively prevent the algorithm from falling into an invalid loop or local extremum. This mechanism has the advantages of fast convergence speed, low implementation complexity, and controllable algorithm process. By constantly trying and selecting, the optimal or suboptimal path is gradually approached, effectively avoiding the problems of difficult implementation and slow convergence of traditional brute force or complex genetic algorithms. It not only ensures the efficiency of the algorithm execution, but also ensures that each exchange can bring actual improvement, improving the smoothness and rationality of the final path, meeting the real-time and batch automatic optimization requirements of the warehouse system.

[0144] The path generated in the optimization process is subjected to line segment intersection detection, and when any two segments are strictly crossed, the previous stable sequence is rolled back and the local exchange optimization is re-executed, specifically including:

[0145] A vector cross product function is constructed; wherein ; The relative orientation of two vectors is determined by the signed area of the directed area, which is used as the basic operator for line segment crossing test, and the calculation is simple and numerically stable;

[0146] Take any two end paths, the end points of the first segment are , , and the end points of the second segment are , ;

[0147] The path segments intersect if and only if: and ; algebraic judgment instead of geometric deduction, fast self-intersection screening, avoid line overlap insertion causing visual confusion and potential navigation ambiguity;

[0148] If the path segment intersection condition is met, the path is rolled back to , and step S5 is re-executed from the first point; the local strategy of point elimination intersection, the minimum range modification to preserve the smoothness.

[0149] Focus on the automation mechanism of path intersection judgment and correction. Through the vector cross product and other methods, it is quickly judged whether the strict intersection between path segments has occurred. Once the problem is found, it is immediately rolled back to the last non-intersecting state and restarted local optimization. This design effectively solves the problem of self-intersection and overlap of warehouse paths in high-density target point scenarios, avoiding operational confusion and navigation ambiguity in actual execution. Unlike existing methods that only focus on optimization goals but ignore path feasibility, this step provides a hard guarantee for the actual usability of the path result. It ensures that each planned path is truly reachable and free from intersection interference, effectively improving picking efficiency and system safety, and laying the foundation for the reliability and stability of actual deployment.

[0150] For high corner positions exceeding the preset threshold, buffer points are inserted along 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, including:

[0151] Let the maximum allowed corner be ;

[0152] If there is a path segment , two buffer points are inserted; only at "statistically significant large angles" to avoid excessive smoothing leading to redundant nodes and long walks;

[0153] Let the interpolation segment length at the first be ; 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 geometric structure too much;

[0154] If or , do not interpolate;

[0155] Otherwise, set the first interpolation point before the corner to , and set the second interpolation point after the corner to ; replace a sharp corner with two smooth transitions, reduce the change in turning speed at a single point, and alleviate Visual jitter and dizziness risk;

[0156] If or exceeds , it is projected onto the boundary:

[0157] Construct ; wherein, is the boundary projection function; the out-of-bound buffer points are projected back to the feasible region boundary to ensure that the path is completely executable and consistent with the real warehouse layout;

[0158] For , respectively apply and replace;

[0159] Construct a new node set ;

[0160] Between the original sequence , insert , form a new sequence ; Ensure that the new node has a globally unique identifier and is embedded in the original sequence to form a smooth path that can be directly rendered and executed.

[0161] A detailed mechanism for high corner buffer segment interpolation is proposed. When a corner greater than a preset threshold is encountered in the path, the system automatically inserts transition points before and after the key corner, and the interpolation points outside the warehouse boundary are automatically projected within the boundary range, and the node set and access sequence are updated simultaneously. In this way, the frequency of large-angle turning can be reduced while ensuring the main structure of the path, reducing the difficulty of personnel operation, equipment wear and tear, and even safety risks caused by sharp turns. The mechanism also reasonably limits the interpolation length, which can achieve smooth transition of the path, and also avoids node redundancy and long path. Compared with the traditional method of reducing corners by optimizing the order or manually processing high corners, the present invention can automatically, dynamically and accurately process high-risk corners, improving the friendliness and safety of the path.

[0162] The coordinate scaling relationship is calculated according to the width and height of the virtual scene, and the plane coordinates of each node in the final path are mapped to virtual scene coordinates to output the final path sequence containing interpolation points, which specifically includes:

[0163] Get the scene width, denoted as ; Get the scene height, denoted as ; Then the mapping coefficient is: , ; wherein, is the direction mapping coefficient; For Direction mapping coefficient; Establish linear scale of physical meters to virtual coordinate units, ensure that the graphic display is proportional to the real size mapping;

[0164] For any path point , get virtual coordinates: , ; Project the physical path accurately to plane coordinates, ensure consistency and interactivity of visualization and navigation guidance;

[0165] Output path sequence ; Wherein, is the total number of final path nodes, including interpolation points; The output coordinate sequence can be directly used for engine rendering and voice / arrow navigation to complete the complete link from "data-geometry-visual".

[0166] The physical path is accurately mapped to the virtual reality scene. By obtaining the actual size of the virtual scene, the appropriate coordinate scaling relationship is calculated, the spatial coordinates of each path node are linearly transformed into the corresponding points in the virtual environment, and the final path sequence containing all interpolation points is output. This not only guarantees the consistency of the path performance 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 docking, scale adjustment or detail loss between real physical path and digital virtual scene. It can make the path planning result be directly called by the virtual reality engine, and provide accurate data basis for warehouse digital twin, picking training simulation and other applications. Compared with traditional manual mapping or rough processing methods, the automation level and interaction experience of the system are improved.

[0167] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or action from another, and do not necessarily require or imply that there is any such actual relationship or order between these entities or actions. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0168] The above is only the preferred embodiment of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the technical principles of the present application, several improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A virtual reality-based warehouse path planning method, characterized by, The method comprises the following steps: S1, establishing a two-dimensional rectangular coordinate system on the ground of a rectangular warehouse, determining the boundary range of the warehouse, determining the planar coordinates and unique numbers of a picking starting point and a plurality of picking target points, and forming a path node set containing the starting point and all target points; S2, taking the starting point as the sequence start, selecting step by step according to the Euclidean distance between two nodes, preferentially selecting the unvisited target point closest to the current last node, and when there are equal distances, sequentially determining according to the order of the planar coordinates and the order of the numbers of the nodes to obtain an initial visiting order covering all target points; S3, connecting adjacent nodes in the initial visiting order into a plurality of path segments, and calculating the turning angles between the path segments based on the directional relationship of adjacent path segments; S4, taking the sum of the turning angles of all adjacent path segments in the entire path as the main evaluation index, and combining the path length, node arrival order and other constraints to form a target function; S5, attempting to exchange two nodes at different positions in the current visiting order, retaining the exchange and updating the sequence when the angle cost after the exchange is reduced, and abandoning the exchange when the cost is not reduced, through multiple iterations, determining the termination condition according to the preset maximum number of iterations and the continuous improvement threshold, and obtaining a converged optimized path; S6, performing line segment intersection detection on the path generated in the optimization process, and when any two segments are detected to strictly cross, returning to the previous stable sequence and re-executing local exchange optimization; S7, for high turning angle positions exceeding a preset threshold, inserting buffer points along adjacent path segments before and after the turning angle, and when the interpolation point exceeds the boundary of the warehouse, projecting it into the boundary range and updating the node set and the visiting order; S8, calculating the coordinate scaling relationship according to the width and height of the virtual scene, mapping the planar coordinates of each node in the final path to the virtual scene coordinates, and outputting the final path sequence containing the interpolation points.

2. The warehouse path planning method based on virtual reality according to claim 1, characterized in that, The method of establishing a two-dimensional rectangular coordinate system on the ground of a rectangular warehouse, determining the boundary range of the warehouse, determining the planar coordinates and unique numbers of a picking starting point and a plurality of picking target points, and forming a path node set containing the starting point and all target points, comprises the following steps: A two-dimensional rectangular coordinate system is established on the ground of the rectangular warehouse An arbitrary vertex is selected as the origin Two edges connecting the rectangular warehouse ground and the origin are selected as the axis and the axis respectively obtaining the length of the warehouse, denoted as obtaining the width of the warehouse, denoted as ​ Constructing warehouse coordinate domain ; Let the starting number of the picker be , the coordinates in the coordinate system be ; wherein, is the coordinate of the starting point in the axis; is the coordinate of the starting point in the axis; The total number of picking target points is obtained, denoted as Each target point is numbered as , , and has a two-dimensional coordinate ; wherein is an enumeration index, does not represent the access order, and only indicates the enumeration order when the th storage location is defined; is the number of the th picking target point; is the coordinate of the th picking target point; The set of path nodes is constructed as .

3. The virtual reality-based warehouse path planning method of claim 2, wherein, The method of taking the starting point as the sequence start, selecting step by step according to the Euclidean distance between two nodes, preferentially selecting the unvisited target point closest to the current last node, and when there are equal distances, sequentially determining according to the order of the planar coordinates and the order of the numbers of the nodes to obtain an initial visiting order covering all target points, comprises the following steps: Euclidean distance between any two points , ; where, and are the indices of the two path points respectively;​ Initialization of waypoint number sequence ; Initializing a set of unvisited points ; To To Steps S201 to S205 are executed in this order: wherein is a path access sequence bit sequence; S201、set the current path end point number as ; S202, calculate the distance from all unvisited points: ; wherein, is the unvisited node identifier set remaining after completion of the visit to the visited location is the unvisited node identifier set remaining after completion of the visit to the visited location S203, selecting the nearest point: ; wherein, is a global node identification selected as the access location ; If there are multiple minimum distances , select , i.e. by coordinates first, then , if still tied, select the one with the smallest node number; S204, updating the sequence ; S205, updating the set ; final generation .

4. The warehouse path planning method based on virtual reality according to claim 3, characterized in that, The method of connecting adjacent nodes in the initial visiting order into a plurality of path segments, and calculating the turning angles between the path segments based on the directional relationship of adjacent path segments, comprises the following steps: Two consecutive points in the path , Construct segment vector Specifically: ; wherein, is an access location to a location displacement vector; If , the two points coincide, set the segment as an invalid segment, and skip the angle and interpolation calculation; wherein, is a two-norm; If and then: ; wherein, is the corner of the segment to segment; where the dot product ; The parameters are limited in the interval , in particular: 。 5. The virtual reality-based warehouse path planning method of claim 4, wherein, The method of taking the sum of the turning angles of all adjacent path segments in the entire path as the main evaluation index, and combining the path length, node arrival order and other constraints to form a target function, comprises the following steps: calculating total rotation ; Minimization goal: select path number sequence Make Min, i.e. .

6. The virtual reality-based warehouse path planning method of claim 5, wherein, The method of attempting to exchange two nodes at different positions in the current visiting order, retaining the exchange and updating the sequence when the angle cost after the exchange is reduced, and abandoning the exchange when the cost is not reduced, through multiple iterations, determining the termination condition according to the preset maximum number of iterations and the continuous improvement threshold, and obtaining a converged optimized path, comprises the following steps: Let the current path number sequence be denoted as , and the angle cost be ; wherein is the path number sequence under the th iteration; is the number of iterations; is the path corner angle under the th iteration; For any , Steps S501 to S502 are sequentially executed to perform path switching: wherein, is a sequence of interchangeable waypoint numbers positions; S501, perform number exchange: ; wherein the function represents exchanging the numbers in the th position of the sequence and the th position, resulting in a new sequence ; S502, calculate a new angle cost: ; If , then receive exchange ; if , then reject and keep unchanged; The maximum number of rounds of consecutive failures is set to , and the maximum number of iteration rounds is ; If consecutive The wheel is not updated or reaches the maximum number of wheels then terminate.

7. The virtual reality-based warehouse path planning method of claim 6, wherein, The path generated in the optimization process is subjected to line segment intersection detection, and when strict intersection of any two segments is detected, the previous stable sequence is rolled back and the local exchange optimization is re-executed, specifically including: Constructing vector cross product functions ; wherein, ; ; Take any two end paths, the first segment endpoint is , , the second segment endpoint is , ; The path segments intersect if and only if: and ; If the path segment intersection condition is satisfied, the path is backed up to Step S5 is executed again from the first point. Step S5 is executed again from the first point.

8. The virtual reality-based warehouse path planning method of claim 7, wherein, For high corner positions exceeding a preset threshold, buffer points are inserted along adjacent path segments before and after the corner, and when an interpolation point exceeds the warehouse boundary, it is projected within the boundary range and the node set and access order are updated, specifically including: Let the maximum allowable rotation angle be ; If there is a path segment then insert two buffer points; Set the interpolation segment length to ; If or then no interpolation. Otherwise, set the first interpolation point at , and the second interpolation point at ; If or exceeds , project it onto the border: Constructing ; wherein, is a boundary projection function; to , respectively apply and replace; Constructing a new set of nodes ; between the original sequences , forming a new sequence .​ 9. The virtual reality-based warehouse path planning method of claim 8, wherein, The coordinate scaling relationship is calculated according to the width and height of the virtual scene, the plane coordinates of each node in the final path are mapped into virtual scene coordinates, and the final path sequence including interpolation points is output, specifically including: acquiring a scene width, denoted as ; acquiring a scene height, denoted as ; then the mapping coefficient is: , ; wherein, is a direction mapping coefficient; is a direction mapping coefficient; For any waypoint , obtain virtual coordinates: , ; Output path sequence ; wherein is the total number of final path nodes, including interpolation points.

Images