A face-centered cubic lattice-based unmanned aerial vehicle three-dimensional path planning method

Abstract

The application discloses a kind of unmanned vehicle three-dimensional path planning method based on face-centered cubic lattice, belong to unmanned vehicle path planning technical field. Obtain three-dimensional airspace, start and end point and obstacle information, carry out discretization modeling to three-dimensional airspace using face-centered cubic lattice, lattice parameter is adaptively determined according to the shape envelope of unmanned vehicle, utilize lattice index constraint to generate legal node, execute deterministic mapping processing to the start and end point of non-lattice position;Global path search is completed using FCC-A*, and global discrete path is obtained by implementing node and path edge double-layer collision detection;Newly appeared obstacle is perceived by means of look-ahead window during flight, and local bounding box is constructed to execute local re-planning, complete path splicing and configure multi-level abnormal degradation strategy;The application can output the discrete flight path with higher safety according to the non-orthogonal topology of face-centered cubic 12-neighborhood, can respond to the sudden obstacle in flight, and is suitable for urban low altitude, emergency rescue and other operation scenarios.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning technology, and specifically to a UAV three-dimensional path planning method based on a face-centered cubic lattice. Background Technology

[0002] 3D path planning for unmanned aerial vehicles (UAVs) is a core technology for low-altitude urban transportation, logistics delivery, emergency rescue, and security patrol. The discrete path output by the path planning algorithm directly determines the feasibility and safety of the UAV flight. Currently, mainstream engineering solutions generally use Cartesian orthogonal grids, i.e., cubic voxels, to complete 3D airspace environment modeling. Search or sampling algorithms such as Rapidly Expanding Random Trees (RRT) are used to generate UAV flight paths. Traditional 6-connected orthogonal grids only support node expansion in six directions (positive and negative) along the X, Y, and Z axes. This modeling method has a simple data structure and low program implementation threshold, and is widely used in small to medium-sized simple airspace scenarios. However, the motion direction of 6-connected grids is strictly constrained by the coordinate axes, and the path generation process is prone to generating a large number of stepped discrete paths with right-angle turns, resulting in a large cumulative turning angle of the overall path. To meet the actual kinematic constraints of UAVs, such discrete paths cannot be directly delivered to the flight control system for execution. An additional trajectory smoothing and time parameterization post-processing module must be configured. This module relies on B-spline curves, polynomial interpolation, gradient optimization, and other methods to perform secondary processing on the discrete path points. The post-processing stage will bring additional computational resource consumption. On UAV platforms with limited onboard computing power, this will reduce the computing power occupied by other task modules and adversely affect the real-time performance of path planning.

[0003] To mitigate the stepped path defects caused by 6-connected orthogonal grids, existing technologies extend grid connectivity to 26-connected full neighborhoods. This significantly increases the number of search directions available to nodes, reducing right-angle turns in the path. While the 26-connectivity scheme can improve the morphology of discrete paths to some extent, the number of candidate branches for each extended node increases from 6 to 26. This leads to a significant increase in computational overhead during node expansion, including neighborhood coordinate generation, obstacle collision detection, open set maintenance, and memory read / write access. When the planned spatial area is large or the grid resolution is high, the massive number of candidate neighborhood nodes continuously amplifies the computational burden of the algorithm, directly reducing the response performance of real-time path planning on the airborne end. Meanwhile, some existing non-orthogonal grid studies employ hexagonal voxelization planning, utilizing the isotropic nature of hexagonal structures in the horizontal plane to optimize planar path search efficiency. However, this approach is essentially a 2.5D columnar expansion scheme, optimizing only the horizontal dimension while still using orthogonal columnar constraints in the vertical direction, failing to achieve complete three-dimensional topology adaptation. In urban low-altitude operation environments with dense high-rise buildings and frequent take-off and landing maneuvers, vertical motion constraints will expose obvious limitations and make it difficult to adapt to the actual use needs of UAV full three-dimensional spatial maneuvering operations.

[0004] During actual drone flight operations, onboard sensors continuously detect sudden obstacles and temporary no-fly zones not detected during the global planning phase. Local replanning capability in dynamic obstacle scenarios is an essential function of any path planning system. In traditional orthogonal grid systems, once a pre-planned global path is blocked by a newly detected obstacle, limited by grid axial movement rules, local replanning can only generate a box-shaped detour trajectory with right-angle turns. This detour trajectory suffers from long detour distances and large turning angles, requiring significant spatial and temporal costs for drones to avoid sudden risks. For dynamic obstacles, existing technologies employ two approaches: the first is to directly abandon the original global path and re-perform a global path search across the entire airspace. This method is computationally intensive, and re-searching the entire airspace when the obstacle only affects a small segment of the path results in a waste of computational resources. The second approach only performs local replanning, but the inherent limitations of orthogonal grid topology lead to poor geometric quality of the local detour path.

[0005] Existing 3D path planning technology has an irreconcilable trade-off between discrete path turning characteristics, algorithm search computational overhead, and dynamic obstacle local replanning geometric efficiency. It is difficult to simultaneously take into account path morphology, computational overhead, and dynamic obstacle avoidance capabilities, and cannot well adapt to the 3D flight operation requirements of UAVs in complex urban low-altitude environments. Summary of the Invention

[0006] To address the aforementioned technical problems, this application discloses a three-dimensional path planning method for unmanned aerial vehicles (UAVs) based on face-centered cubic lattices, comprising the following steps:

[0007] S1. Acquire 3D spatial data, starting node, target node, and obstacle information;

[0008] S2. Discretize the three-dimensional spatial domain according to the face-centered cubic lattice rule to obtain a set of discrete lattice nodes, and map obstacle information to the discrete lattice nodes, marking impassable nodes; each central node of the face-centered cubic lattice is configured with twelve equidistant neighboring nodes, which are distributed in the diagonal directions of the three coordinate planes; determine the actual step size of lattice adjacency based on the UAV's outer envelope size, obtain the single-axis coordinate increment based on the actual step size, and set the safety margin used for collision detection and local bounding boxes;

[0009] S3. Adopts a face-centered cubic lattice topology. The algorithm performs a global path search to obtain a collision-free global three-dimensional discrete flight path; The algorithm is based on the A* search framework. It expands the neighborhood according to the twelve neighboring nodes of the face-centered cubic lattice. During the neighborhood expansion, it performs double-layer collision detection between nodes and path edges, calculates the cost of adjacent edges, and uses three-dimensional Euclidean distance as a heuristic to estimate the cost.

[0010] S4. Control the UAV to fly along a global three-dimensional discrete flight path, and continuously detect whether new obstacle areas appear ahead of the flight through the forward-looking window;

[0011] S5. When a collision risk is detected, determine the reconnection node, construct a local bounding box, and execute the operation inside the local bounding box. Local path search yields a local detour path. This local detour path is then concatenated with the global 3D discrete flight path to obtain an updated collision-free 3D discrete flight path. If the local search fails, the local bounding box is expanded for retry. If the retry still fails, a global replanning is triggered. If the global replanning still fails, a planning failure is output and a hover command is issued.

[0012] S6. If no collision risk is detected, the global three-dimensional discrete flight path without collision is directly output.

[0013] Preferably, in step S2, the three-dimensional spatial domain is discretized according to the face-centered cubic lattice rule, including determining the neighborhood displacement relationship of the face-centered cubic lattice, screening the legal nodes of the lattice, and mapping the physical coordinates to the lattice index coordinates.

[0014] The set of neighborhood basis vectors of the face-centered cubic lattice is:

[0015]

[0016] in, For single-axis coordinate increment parameters, each set of positive and negative values ​​is independently taken to generate all neighborhood displacements, and all twelve neighborhood nodes corresponding to the center node are generated from the neighborhood basis vector.

[0017] The preferred set of legal nodes for a face-centered cubic lattice is given by the formula: In the formula, For lattice index integers, For single-axis coordinate increment parameters;

[0018] When discretizing a three-dimensional spatial domain, traverse the integer indices. ,according to Valid FCC lattice nodes are obtained through constraint screening, invalid nodes that do not meet the constraints are removed, and a mapping relationship is established between lattice index coordinates and Cartesian physical coordinates.

[0019] Preferably, step S2 further includes lattice node mapping processing for the starting point and the target point; when the Cartesian coordinates of the original starting position or the original target position do not coincide with any FCC lattice node, the following processing is performed:

[0020] Enumerate all candidate lattice nodes corresponding to the surrounding cells of the original position, calculate the Euclidean distance between each candidate lattice node and the original position, and sort the candidate lattice nodes in ascending order of distance.

[0021] The candidate lattice nodes are subjected to node safety envelope collision detection and collision detection between the candidate nodes and the original positions in the sorting order.

[0022] Select the first candidate lattice node that passes all tests and use it as the starting node or the target node after mapping, respectively; if there is no candidate lattice node that meets the conditions, output that the mapping failed.

[0023] Preferably, the specific process in S2 for determining the actual step size of lattice adjacency and the single-axis coordinate increment parameter based on the UAV's external envelope size is as follows:

[0024] Obtain the outer envelope dimensions of the UAV, including the propellers and payload, in three orthogonal directions. , , Calculate the envelope feature size Set the actual step size of lattice adjacency. The single-axis coordinate increment parameters satisfy ;

[0025] Within the complete execution cycle of a single global path search or a single local path search, and Keep it fixed; when the drone's shape envelope changes, recalculate it before the next path planning initiation. and Update the FCC lattice node set.

[0026] Preferably, the node and path edge dual-layer collision detection includes a node collision determination process and a path edge collision determination process.

[0027] Node collision determination: Calculate the minimum circumscribed sphere radius of the UAV's external envelope. Set a safety margin The radius of the safe enclosing sphere is obtained. ; Construct a sphere with the lattice node to be detected as the center and a radius of ; The safety bounding sphere; if the safety bounding sphere intersects with the obstacle region, then the lattice node is marked as a non-passable node;

[0028] Path edge collision determination: Using the connecting line segment between two adjacent lattice nodes to be detected as the axis, a radius equal to... The capsule-shaped sweep body; if the capsule-shaped sweep body intersects with the obstacle area, then the path edge of that section is determined to be impassable.

[0029] Preferably, in S3 The algorithm performs a global path search, including initialization of open and closed sets, iterative expansion of nodes, neighborhood verification and filtering, cost value update, and search termination judgment.

[0030] The algorithm evaluation function is ;in To reach the node from the starting node after mapping The cumulative cost of adjacent edge Euclidean length; For nodes The heuristic cost of the 3D Euclidean distance to the mapped target node is given by the following formula: In the formula For the current node Cartesian physical coordinates These are the Cartesian physical coordinates of the target node after mapping;

[0031] Open sets adopt the following approach: This is a priority queue for sorting key values. During each node expansion, the twelve neighboring nodes are traversed, filtering out neighboring nodes that are within the closed set, impassable due to node collisions, or impassable due to path edge collisions. The remaining valid neighboring nodes are then processed. Value, parent node, Values ​​are updated and the open set is maintained.

[0032] Preferably, the construction of the look-ahead window and the collision detection execution process in S4 are as follows: reading the look-ahead detection length of the airborne sensor. The actual step size of lattice adjacency is denoted as Number of forward window nodes ;

[0033] When the UAV flies along a global three-dimensional discrete flight path, the index of the current path point of the UAV is read in real time, and the continuous path points from the current index are selected on the global path. For each path point, collision detection is performed sequentially on all path nodes and path edges between adjacent nodes within that range. When any node or path edge is detected to intersect with a newly appearing obstacle area, a collision risk is determined, triggering a local replanning process.

[0034] Preferably, the specific process for determining the reconnection node in S5 is as follows:

[0035] Obtain the index of the first path point that collides within the lookout window as the starting index of the affected path segment. Traverse the global path backward to determine the index of the end of the path segment where consecutive collisions occur. ;

[0036] From the index Begin by selecting candidate path points one by one along the global path toward the target node;

[0037] For each candidate path point, perform node collision detection and path edge collision detection between the candidate path point and the previous path point in sequence; determine the first candidate path point that passes both the node and the path edge detection as the reconnection node; if no candidate path point that meets the conditions is found after traversing to the global path endpoint, output that there is no available reconnection node.

[0038] Preferably, the S5 step of constructing a local bounding box, performing a local search, and performing path concatenation degradation processing specifically includes:

[0039] Using the lattice node and reconnection node corresponding to the current position of the UAV as the diagonal vertex, a cuboid is generated with an aligned axis. The safety margins of the cuboid are then extended outwards in the X, Y, and Z coordinate directions. This yields the local bounding box used in the first local search.

[0040] Using the local bounding box as a constraint in the search space, call... The algorithm performs a local path search with the current location node of the drone as the search starting point and the reconnection node as the search ending point;

[0041] When a feasible local detour path is found through local path search, the local detour path and the original global path after the reconnected node are deduplicated to complete the path splicing, and collision detection is performed again on the path edges at the splicing position.

[0042] When the first local path search fails, the safety margin for outward expansion is replaced with... Reconstruct the local bounding box and perform the local search again; if the search still fails, perform the global search. Replan; if global replanning still fails, output a planning failure status and issue a hover command to the flight controller.

[0043] Compared with the prior art, the technical solution of this application has the following technical effects:

[0044] This invention uses a face-centered cubic lattice to complete three-dimensional spatial discrete modeling, which can provide multiple sets of non-axial diagonal motion directions under the condition of a limited number of neighborhood nodes, enriching the selectable motion directions in the discrete space; the distance between each neighborhood node is kept consistent under the topological structure, ensuring uniform edge cost during path search, reducing path distortion caused by differences in discrete directions, and providing a reliable spatial topological basis for subsequent generation of discrete paths.

[0045] This invention establishes a complete dual-layer collision detection mechanism, simultaneously determining node safety and verifying path edge collisions. It can not only filter out impassable nodes within obstacle zones but also identify invalid path edges where nodes are safe but their connections traverse obstacles. This two-dimensional filtering of unsafe paths improves the safety and reliability of the output discrete paths and reduces the risk of hidden collisions.

[0046] This invention is based on an improved A* search framework adapted to face-centered cubic lattice topology. It expands nodes according to the lattice's own neighborhood rules and completes path solving with corresponding cost calculation and heuristic evaluation methods. It reuses the mature main search framework and only adjusts the neighborhood expansion and evaluation-related logic, making it easy to port and deploy on airborne equipment. It can stably output a sequence of collision-free discrete path points that meet topological constraints.

[0047] This invention sets up a mapping process from start and end positions to lattice nodes. For start and end points that are not on lattice nodes, available nodes are selected according to distance and security verification rules. This can be compatible with input coordinates at any location in real-world scenarios, avoids being constrained by the position of lattice nodes, and allows path planning to adapt to any actual take-off point and target point of the UAV, thus broadening the practical application scenarios of the solution.

[0048] This invention constructs a hierarchical global-local replanning processing mechanism. It relies on a forward-looking window to perceive newly appearing obstacles during flight, limits the local bounding box to carry out local path search, and configures complete path splicing and degradation processing logic. It searches the local space affected by obstacles without repeatedly recalculating the entire airspace, and can update the flight path in a timely manner to cope with the changing working conditions of the environment during flight.

[0049] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0050] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0052] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0053] Figure 1 This is a schematic diagram of the overall process of the path planning method based on face-centered cubic lattice in an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the spatial distribution of the central node and 12 neighboring nodes of a face-centered cubic lattice according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram illustrating the mapping and transformation from a continuous three-dimensional spatial domain to an FCC discrete lattice space according to an embodiment of the present invention;

[0056] Figure 4 This is an embodiment of the present invention. A schematic diagram of the internal execution flow of the global path search algorithm;

[0057] Figure 5 This is a schematic diagram of the complete local replanning process during flight, according to an embodiment of the present invention.

[0058] Figure 6 This is a schematic diagram illustrating the principle of the local bounding box and detour path when encountering dynamic obstacles according to an embodiment of the present invention;

[0059] Figure 7 This is a graph showing the comparison of path length changes between the two algorithms under multiple test cases in an embodiment of the present invention.

[0060] Figure 8This is a graph showing the cumulative steering angle changes of the two algorithms under multiple test cases in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0062] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0063] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0064] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0065] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0066] It should also 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.

[0067] Example 1

[0068] This embodiment mainly describes a three-dimensional path planning method for unmanned aerial vehicles (UAVs) based on face-centered cubic lattices, such as... Figure 1 As shown, it includes the following steps:

[0069] S1. Acquire 3D spatial data, starting node, target node, and obstacle information;

[0070] S2. Discretize the three-dimensional spatial domain according to the face-centered cubic lattice rule to obtain a set of discrete lattice nodes, and map obstacle information to the discrete lattice nodes, marking impassable nodes; each central node of the face-centered cubic lattice is configured with twelve equidistant neighboring nodes, which are distributed in the diagonal directions of the three coordinate planes; determine the actual step size of lattice adjacency based on the UAV's outer envelope size, obtain the single-axis coordinate increment based on the actual step size, and set the safety margin used for collision detection and local bounding boxes;

[0071] S3. Adopts a face-centered cubic lattice topology. The algorithm performs a global path search to obtain a collision-free global three-dimensional discrete flight path; the FCC-A algorithm is based on The search framework expands the neighborhood according to the twelve neighborhood nodes of the face-centered cubic lattice. During the neighborhood expansion, a two-layer collision detection is performed between the node and the path edge to calculate the cost of the adjacent edge. The three-dimensional Euclidean distance is used as a heuristic to estimate the cost.

[0072] S4. Control the UAV to fly along a global three-dimensional discrete flight path, and continuously detect whether new obstacle areas appear ahead of the flight through the forward-looking window;

[0073] S5. When a collision risk is detected, determine the reconnection node, construct a local bounding box, and execute the operation inside the local bounding box. Local path search yields a local detour path. This local detour path is then concatenated with the global 3D discrete flight path to obtain an updated collision-free 3D discrete flight path. If the local search fails, the local bounding box is expanded for retry. If the retry still fails, a global replanning is triggered. If the global replanning still fails, a planning failure is output and a hover command is issued.

[0074] S6. If no collision risk is detected, the global three-dimensional discrete flight path without collision is directly output.

[0075] This invention addresses the limitation of orthogonal grid motion directions in 3D spatial path planning by employing a face-centered cubic (FCC) lattice structure to discretize a continuous 3D spatial domain. In this FCC lattice topology, each central node corresponds to twelve equidistant neighboring nodes. These twelve neighboring nodes are distributed diagonally across the xy, xz, and yz coordinate planes, forming a non-orthogonal spatial adjacency topology. This allows for multiple sets of discrete motion directions while controlling the total number of neighboring nodes, providing a discrete spatial basis for subsequent path search. Figure 2 As shown.

[0076] The current complete configuration of the UAV is used as the basis for determining the actual adjacency step size of the FCC lattice. The UAV's outer envelope needs to include the UAV fuselage, propeller protection structure, and the overall outer contour of the currently mounted equipment. , and Define the envelope feature size as follows: (This refers to the maximum envelope size of the UAV, including its propeller protection structure, and currently mounted in the three orthogonal directions X, Y, and Z). Take the actual adjacency step size between adjacent nodes of FCC. According to the definition of neighborhood basis vectors in this application, the single-axis coordinate increment parameter... Equivalent to .

[0077] When a drone undergoes a change in shape or its mounting equipment is replaced, the lattice parameters... and The parameters are adjusted synchronously with the drone's external envelope dimensions. To reduce the computational complexity of the algorithm implementation, parameter calculations are limited to the period before each global planning initiation or each local replanning initiation; within a single complete... During the search cycle, and The value remains constant. If the UAV's shape envelope changes during the path planning algorithm, it will not perform online scale changes on the FCC lattice topology currently performing the search calculation. Instead, it will wait until the next path planning task starts to recalculate. , Simultaneously update the FCC lattice node set, obstacle mapping relationship, and search path data.

[0078] A preferred implementation: This implementation directly uses the envelope feature size of the UAV's overall shape to determine the lattice step size. Alternatively, the specific value of the step size can be manually specified based on the complexity of the operational airspace environment and the planned resolution requirements. This alternative method only changes the discrete granularity of the lattice space and does not alter the 12-neighborhood topology of the face-centered cubic lattice itself.

[0079] The neighborhood basis vectors are defined with any FCC lattice node as the spatial center node, and the set of discrete displacement vectors corresponding to the twelve neighboring nodes of that center node is denoted as . The sign of each vector within the set can be independently determined, thereby generating all twelve neighborhood displacement directions.

[0080]

[0081] gather The Euclidean norm of every neighborhood basis vector is equal, and the actual step size between adjacent FCC lattice nodes is denoted as . All twelve neighboring nodes are exactly the same Euclidean distance from the central node, ensuring that the cost of each adjacent edge remains consistent during the path search process and eliminating the inherent cost deviation between different expansion directions.

[0082] A preferred embodiment: This embodiment uses only the above 12 sets of displacement vectors to construct the neighborhood relationship of the face-centered cubic lattice, without introducing the neighborhood displacement vectors corresponding to the body-centered cubic lattice, diamond lattice, and hexagonal close-packed lattice, and excludes other lattice topologies.

[0083] Spatial discretization mapping generates a large number of candidate lattice nodes from the continuous three-dimensional spatial domain to be used for path planning, strictly following the FCC lattice topology rules. Figure 3 As shown, a deterministic two-way mapping relationship is established between lattice index integer coordinates and Cartesian physical space coordinates, enabling mutual conversion between index coordinates and actual spatial points.

[0084] Given a starting and target location for a task, when the original starting or target coordinates do not fall exactly on an FCC lattice node, there are two processing modes. The first mode maps the original coordinates to available adjacent lattice nodes that satisfy all safety constraints, using these mapped nodes as the actual starting and ending points for the path search. The second mode retains the original Cartesian space target coordinates without mapping. The search process internally uses a distance threshold as the termination condition. Two processing modes can be selected based on task requirements.

[0085] A preferred implementation: prioritize performing node mapping processing on amorphous points; in scenarios where high accuracy of the target point location is required, the original point can be retained, and a distance threshold can be used as the search termination condition;

[0086] For obstacle mapping and collision detection rules, the minimum circumscribed sphere radius of the UAV's shape envelope is obtained. Set a unified safety margin The radius of the safety bounding sphere used in the collision detection process was further calculated. .

[0087] Node-level collision determination: Using the Cartesian coordinates of each lattice node as the center of a sphere, a radius equal to... The safe bounding sphere is used to traverse all obstacle regions in the environment. If the safe bounding sphere corresponding to a certain lattice node intersects with the obstacle space region, the lattice node is marked as an impassable node and will not be selected in subsequent search processes.

[0088] Path edge collision determination: For any two adjacent lattice nodes, a collision path is constructed with a radius equal to the line segment connecting the two points as the central axis. The capsule-shaped sweep body; if the capsule-shaped sweep body and the obstacle area intersect spatially, then the path edge of that segment is marked as an impassable path edge.

[0089] By using collision detection at both the node and path edge levels, two types of unsafe objects can be filtered out simultaneously. One type is an unpassable node that is itself within the influence range of an obstacle, and the other type is an invalid path edge where both endpoint nodes are determined to be safe, but the line connecting the two points will cross the obstacle area.

[0090] A preferred implementation: This scheme uses a safe bounding sphere combined with a capsule-shaped sweep body to complete collision detection; when the input environment map is in an occupied grid format, discrete sampling can also be performed along the path edge segments using a sampling interval no greater than half the voxel side length, and the collision detection effect of the sampled point collision detection is approximately equivalent to that of the capsule sweep body.

[0091] Based on the FCC lattice space topology obtained through discrete modeling, the standard is reused. The overall framework of heuristic search, tailored to the topological characteristics of face-centered cubic lattices, modifies the rules for expanding neighboring nodes and the logic for calculating the single-step cost of adjacent edges. (Nodes) The evaluation function is ,in This represents the search progress from the starting node to the current node. The cumulative path cost Represents the current node The heuristic cost estimation to the target node in this scheme The calculation is performed using three-dimensional Euclidean distance.

[0092] When the cost of each adjacent edge in the FCC lattice is directly taken as the Euclidean length of the corresponding edge, the target point corresponds to an exact FCC lattice node, and reaching the target node is used as the standard termination condition, this 3D Euclidean distance heuristic function will not overestimate the true shortest path cost from the node to the target node, and can solve for the shortest path in the FCC discrete graph. If a distance threshold is used to terminate the search early, a weighted heuristic function is used, or other approximate solution strategies are employed, the algorithm output only represents feasible paths that meet the corresponding termination conditions, and cannot unconditionally guarantee that the output path is the globally optimal path.

[0093] The algorithm ultimately outputs a collision-free discrete pathpoint sequence, which represents only a set of spatial points and inherently satisfies the kinematic constraints of the UAV. For example... Figure 4 As shown, The specific steps for global path search are as follows:

[0094] S31 Initialization Settings: Initialize the open set and the closed set. The open set stores lattice nodes waiting to be expanded, and the closed set stores lattice nodes that have completed the expansion process. Add the starting node of the mapped path to the open set, assigning it the actual cumulative cost and total evaluation cost, and simultaneously calculating the heuristic estimated cost from the starting node to the target node. The open set is implemented using a priority queue data structure, with the queue prioritized by node... The total evaluation cost is used as the sorting key to ensure that the algorithm can quickly extract the node to be expanded with the minimum total cost in each round.

[0095] S32 Current Node Selection: In each iteration, the node with the smallest total evaluation cost is selected from the open set as the current expansion node for this round. This node is then removed from the open set and simultaneously added to the closed set.

[0096] S33 Termination Condition Judgment: A termination condition judgment is performed. If the current extended node is the target lattice node after mapping, the search is considered successful, and the process is reversed along the parent node pointer to reconstruct the global discrete path. If the original target position does not fall on a lattice node, the Euclidean distance between the current node and the original target's Cartesian coordinates is checked to see if it is less than or equal to a pre-set termination threshold. If the conditions are met, the search can be terminated early. In this case, the output path is a feasible path that meets the distance threshold constraint.

[0097] The neighboring nodes are traversed, and the 12 sets of neighboring basis vectors of the FCC lattice are read. The coordinates of all twelve neighboring nodes corresponding to the current extended node are calculated in turn. Multi-dimensional validity checks are performed on each calculated neighboring node. If a neighboring node already exists in the closed set, or a neighboring node is marked as impassable, or a candidate path edge between the current node and the neighboring node is determined to be impassable, then the neighboring node is skipped and no further update logic is performed on the neighboring node.

[0098] S34 Cost Calculation and Update: Calculate the single-step movement cost from the current expanded node to neighboring nodes. The single-step cost is uniformly taken as the Euclidean norm corresponding to the neighborhood basis vector. ; to accumulate the current node's data The value is added to the cost of the single-step movement to obtain the temporary cumulative cost of the neighboring node.

[0099] If the neighboring node is not in the open set, or if the newly calculated temporary cumulative cost is less than the old cost already saved by the neighboring node. If the value is not specified, then update the parent node of the neighboring node to point to the current expanding node in this round, and set the neighboring node's value to the current expanding node. The value is replaced with the temporary cumulative cost, and the total evaluation cost of the neighboring node is recalculated. If the neighboring node was not originally in the open set, then add the neighboring node to the open set.

[0100] The above steps of selecting the current node, judging the termination condition, traversing the neighboring nodes, calculating the cost and updating are repeated in a loop. There are two cases when the loop ends: the first case is that the target node is found and the reconstructed global discrete path is output; the second case is that the open set becomes an empty set and it is determined that there is no feasible path under the condition of the empty domain.

[0101] A preferred implementation: This algorithm uses only three-dimensional Euclidean distance as the heuristic function, and does not use Manhattan distance, Chebyshev distance, or weighted heuristic functions; this algorithm is based on a standard adapted to FCC topology. The framework implementation does not use Dijkstra, D Lite, RRT, or various swarm intelligence algorithms as the main search framework.

[0102] During actual flight operations, onboard sensors continuously perceive the environment, detecting new obstacles and temporary no-fly zones not identified during the global planning phase. To address this, this application employs a hierarchical global-local replanning strategy. This strategy only re-searches the path within the local spatial area affected by the obstacle, avoiding recalculation of the entire three-dimensional airspace, thus reducing computational overhead during the replanning phase. This part involves reactive local replanning based on a real-time updated map. This implementation does not handle continuously moving obstacles. To handle scenarios with continuously moving obstacles, additional obstacle state prediction and path time parameterization modules are required. Figure 5 As shown, the specific execution steps of hierarchical global-local replanning are as follows:

[0103] S51 Forward Collision Detection: Representing the look-ahead detection length achievable by airborne sensors, the actual step size between adjacent nodes in the FCC lattice is... The number of waypoints included in the look-ahead window As the drone flies along the pre-planned global path, it continuously reads the index of the current path point and updates the subsequent points along the global path. Collision detection is performed on all lattice nodes and all adjacent path edges within the range of each path point; once an intersecting node or path edge is detected, the first collision path segment is identified, and the local replanning process is immediately triggered.

[0104] S52 Reconnection Node Determination: The index of the first path point where a collision occurs, detected by the look-ahead window, is denoted as... Continue traversing the global path to identify the end indices of consecutive path segments affected by obstacles. From the index The search for reconnection nodes begins at the current location. The total number of path points in the global path is denoted as [value missing]. Candidate starting position The candidate path point is validated. If the candidate node itself is impassable, or the path edge between the node and its upstream node is impassable, the process continues along the original global path towards the target point, selecting subsequent path points one by one. Collision detection is performed on each node and its corresponding path edge, and the first path point that passes all detections is selected as the reconnection node. If no path point meeting the conditions is found after traversing to the end of the global path, the process returns a state that no reconnection node is available.

[0105] S53 Local Bounding Box Construction: Obtain the lattice node coordinates corresponding to the current position of the UAV and the coordinates of the already determined reconnected nodes. Using these two points as diagonal vertices, generate an axis-aligned cuboid to obtain the minimum bounding area. Expand the cuboid outwards with a safety margin in the X, Y, and Z coordinate directions. ,in The axis-aligned cuboid obtained after expansion serves as the spatially defined region for the first local path search, and subsequent local paths will follow suit. Searches are only allowed to expand nodes within the bounding box.

[0106] A preferred implementation: This implementation uses an axis-aligned cuboid as the bounding box for local search, without using other geometric shapes such as spheres or cylinders to limit the search space; this solution only relies on the obstacle detection results of the look-ahead window to trigger the local replanning process, without using replanning triggering logic based on relative speed, collision time, or moving obstacle prediction.

[0107] S54 Local Path Search: Called within the constructed local bounding box space. The algorithm sets the lattice node corresponding to the drone's current position as the search starting point and the reconnected node as the search ending point. During the search process, only nodes falling within the bounding box and whose nodes and path edges have all passed collision detection are allowed to generate local detour paths. The termination threshold for the local search is set to... If the endpoint is a precise FCC lattice node, then reaching the target node is preferred as the search termination condition.

[0108] The S55 continues to fly along the original global path. When no obstacles are detected within the forward window and there is no risk of collision, the UAV continues to perform flight operations according to the pre-planned global discrete path point sequence, traversing each path point in turn, maintaining the established flight attitude and speed to complete the route.

[0109] While continuing to fly along the original global path, the system will not stop the environmental perception and risk identification process. It will keep the forward window collision detection function working, continuously perform collision checks on the lattice nodes and adjacent path edges within the subsequent flight path range, and monitor the changes in the airspace environment ahead of the flight in real time. Once a new obstacle or a path segment with a collision risk is detected, the subsequent replanning process will be triggered immediately.

[0110] S56 Path Concatenation and Anomaly Handling: At the path point data level, the local detour path obtained by solving is concatenated with the original global path after reconnecting nodes. After concatenation, a collision detection check is performed again on the path edges at the concatenation position.

[0111] S57 If the first local path search within the bounding box fails, the safety margin for expanding the bounding box outwards is increased from... Enlarged to The local bounding box is reconstructed, and a local path search is performed again. If the second local search after enlarging the bounding box still fails, a complete global path search is triggered. Replanning; if global replanning still fails to find a feasible path, output a planning failure status and issue a hovering command to the UAV flight control system to ensure flight safety.

[0112] A preferred implementation: This implementation outputs a reactive discrete path for newly detected static obstacles. This embodiment does not include functional modules related to continuous moving obstacle trajectory prediction, multi-time-layer path search, multi-UAV collaborative planning, and spatiotemporal conflict resolution.

[0113] To implement the three-dimensional spatial discretization process programmatically, this embodiment represents the FCC lattice nodes using half-spaced integer coordinates and introduces an index parity constraint to filter out invalid nodes that do not belong to the FCC lattice. The formula is as follows: , All are integers. The uniaxial coordinate increment parameters output by the adaptive lattice parameter calculation stage

[0114] For the adaptively determined single-axis coordinate increment parameters The twelve neighborhood displacements of any FCC lattice center node are all taken from the set Each group within the set takes independent values ​​for positive and negative signs, generating all twelve displacement directions; the Euclidean length of each adjacent edge within the lattice is equal to And in this embodiment, let The value of is equal to the feature size of the drone's outer envelope. During program execution, for each set of index coordinates, the following checks are performed sequentially: valid index verification, map boundary range verification, node safety envelope collision verification, and path edge collision verification, to filter out the valid nodes that can participate in the path search.

[0115] When the original starting point or original target position of the task does not fall on any FCC lattice node, the point mapping process is executed. First, the FCC cell where the original point is located and all candidate lattice nodes corresponding to the surrounding adjacent cells are enumerated; the Euclidean distance between each candidate lattice node and the original point is calculated, and the candidate nodes are sorted in ascending order of distance; strictly following the sorting order, node safety envelope collision detection is performed on the candidate nodes in turn, and collision detection of the connecting edges between the candidate nodes and the original point is also performed; the first candidate lattice node that passes all checks is selected as the starting point node or target node after the mapping is completed.

[0116] If two candidate lattice nodes are exactly equal in distance from the original point, then the lattice index is used. Candidate nodes are selected based on lexicographical priority; if no lattice node that can pass all security checks is found after enumerating all surrounding candidate nodes, a mapping failure status is output.

[0117] A preferred implementation: The above is the deterministic point mapping logic, which strictly follows the complete process of distance sorting, double security verification, and lexicographical tie handling; in other embodiments, the nearest filtering strategy can be replaced, but node security envelope verification and path edge collision verification must be performed simultaneously.

[0118] The evaluation function used by the algorithm is , Represents the path from the starting node to the node after mapping. The cumulative Euclidean length of adjacent edges. Representative node The three-dimensional Euclidean distance pointing to the mapped target node is given by the following formula:

[0119]

[0120] In the formula For nodes Cartesian physical coordinates These are the Cartesian physical coordinates of the target node after mapping is complete.

[0121] When the target is an exact FCC lattice node, the adjacent edge cost is Euclidean length, and reaching the target node is used as the standard termination condition, this heuristic function is an acceptable heuristic function and will not overestimate the actual shortest cost of the remaining paths. When a weighted heuristic function or a distance threshold is used to terminate the search early, the output path can only be described as a feasible discrete path that satisfies the termination condition, and the global optimum is no longer guaranteed.

[0122] Suppose that the discrete path point sequence contains, in sequence to total There are several path points; the total path length is defined as the sum of the Euclidean distances between any two adjacent path points in the sequence; the cumulative turning angle is defined as the sum of the angles between two adjacent path segments at all internal path point positions. If comparative experiments between different algorithms are required, the experiments must use the exact same continuous three-dimensional spatial environment, the same start and end points, consistent UAV shape envelope parameters, and a unified lattice step size. and a unified safety margin The metrics that need to be recorded during the experiment include the total number of algorithm expansion nodes, the total number of collision detection executions, the algorithm running time, and the peak memory usage during program execution.

[0123] A preferred implementation method is to record all the above indicators in a simulation comparison experiment scenario; under the actual operating conditions of the airborne UAV, some indicators can be selectively recorded for operation status output based on the airborne computing power conditions.

[0124] The complete closed-loop execution process of local replanning is as follows: First, using a look-ahead window, collision detection is performed on path nodes and edges along the current global path to identify the first collision-involved path segment. Second, the process traverses the original global path towards the target point, filtering candidate reconnection nodes and selecting those that simultaneously meet the collision detection requirements for both nodes and path edges. Third, combining the UAV's current position, reconnection nodes, and obstacle influence range, a local bounding box with a safety margin is constructed, and the process runs under the constraints of this bounding box. Local path search; fourth step, perform deduplication and splicing operation on the obtained local detour path and the remaining segments of the original global path, and perform collision review and verification on the path edges at the splicing position again, and output the updated three-dimensional discrete path point sequence in which all nodes and path edges have passed the collision detection.

[0125] Several abnormal operating conditions may occur during local replanning operations, including not finding available reconnection nodes, and local... If the open set is empty and the path edge collision detection at the concatenation position fails, the algorithm handles these anomalies according to a fixed degradation order: first, increase the outward margin of the local bounding box, reconstruct the bounding box, and retry the local search; if the retry fails, perform a complete global search. Replanning; if global replanning still fails to find a feasible path, output a planning failure status and issue a hovering command to the flight control system, such as... Figure 6 As shown.

[0126] A preferred implementation: This embodiment adopts the above-mentioned fixed three-level exception degradation logic; in other embodiments, the specific value of the bounding box magnification factor can be adjusted, but the three-level processing hierarchy of local retry, global replanning, and hovering after failure is still retained.

[0127] This embodiment is based on a face-centered cubic lattice. The path planning scheme, based on a 12-directional equidistant neighborhood topology, breaks away from the limitation of traditional orthogonal grids that only allow passage along the coordinate axes. It can generate three-dimensional discrete paths with fewer detours and lower turning frequency, effectively reducing the burden of UAV heading adjustment and reducing flight energy consumption. At the same time, in conjunction with a local replanning mechanism, it can quickly respond to sudden obstacles during flight and ensure the safety of low-altitude flight routes.

[0128] Example 2, to objectively characterize the face-centered cubic lattice-based crystal structure proposed in this application. The actual output performance of the UAV 3D path planning method was verified through comparative simulation in a closed 3D simulation environment. Simulation experiments tested both global path planning performance and local replanning performance. All simulation experiments were conducted under unified hardware operating environment, identical 3D airspace map, consistent start and end points, identical UAV shape envelope parameters, and equal safety margins to ensure the comparability of experimental results between different algorithms. The simulation environment constructed a 20m×20m×20m cubic 3D operating airspace. Multiple sets of suspended rectangular obstacles were placed within the airspace at different height levels to simulate real-world conditions such as protruding building structures and temporary no-fly zones in urban low-altitude environments. The obstacles were randomly distributed in 3D space, with some obstacles spanning multiple height planes, thus constructing a complex 3D passage scenario with numerous detour requirements. In the simulation process, the control group used the traditional 6-connected orthogonal cubic grid A* algorithm. Both algorithms adopted the same safety distance setting and collision detection logic, with only the discrete space topology and neighborhood expansion rules being different. Multiple quantitative indicators, such as total path length, cumulative path turning angle, algorithm operation time, number of expanded nodes, extra detour distance generated by replanning, and extra turning angle, were recorded and statistically analyzed to evaluate the geometric characteristics of the path and the computational cost of the algorithm.

[0129] During the global path planning simulation test, fixed starting point coordinates and fixed target coordinates were selected within a 20m×20m×20m 3D obstacle simulation environment. The starting point and target were positioned diagonally opposite each other in the simulation airspace, and the path between the two points needed to bypass multiple suspended rectangular obstacles within the airspace. Under the conditions of this simulation map, starting and ending point configuration, discrete resolution, and algorithm implementation, a traditional 6-connected grid... The algorithm outputs a global path with a total length of 58.00m and a cumulative turning angle of 3960°. The path contains numerous orthogonal right-angle turns, which will cause the UAV to frequently adjust its flight heading during flight. This application... The algorithm outputs a global path length of 41.01m, a significant reduction, which is 29.3% shorter than the control group. The cumulative turning angle decreases to 1020°, a 74.2% reduction in the cumulative turning index compared to the control group. It should be noted that the aforementioned reduction and decrease percentages are limited to the specific simulation conditions of this experiment. Influenced by obstacle arrangement, the relative positions of the start and end points, and the lattice step size parameter, the percentage values ​​obtained in this experiment cannot be directly applied or extrapolated to other operational scenarios; they only represent the algorithm's performance under this specific simulation environment.

[0130] like Figure 7 As shown, the solid line represents the path length variation of a traditional 6-connected raster A*, and the dashed line represents... The path length of the algorithm changes. The trend of the curve shows that across multiple test cases... The corresponding curve is located in a 6-connected grid. Below the curve, it is reflected that in scenarios with multiple obstacle distributions, the overall spatial span of the path obtained based on the face-centered cubic lattice topology solution is smaller. This allows for full utilization of the diagonal directions provided by the lattice for passage, avoiding the detours caused by orthogonal grids requiring travel along coordinate axes. For example... Figure 8 As shown, the curve of the 6-connected grid A* remains in a relatively high angle range overall. The curve values ​​are significantly reduced, and the curve fluctuation amplitude is smaller, which is intuitively reflected. The output path has fewer turning frequencies, and the overall path is smoother.

[0131] When conducting local replanning simulation verification, the aforementioned 20m×20m×20m three-dimensional simulation airspace is reused. First, the global planning algorithm is run to generate an initial collision-free global path. Then, a sudden obstacle area is artificially added in the middle of the initial global path to simulate the actual situation where the onboard sensors of the UAV detect new unknown obstacles during flight. The obstacle will cut off part of the original global path, thereby triggering the local replanning process within the algorithm. During the simulation experiment, the traditional grid... and this application All models were configured with local replanning logic, and the same look-ahead window parameters and local bounding box expansion margins were set. The replanning operation time, the additional detour distance generated after replanning, and the additional turning angle introduced by the local detour path were recorded to complete the comparative analysis of the local replanning dimension.

[0132] Under the local replanning simulation conditions of this group, the traditional grid A* takes 1.25ms to complete a single local replanning operation. The additional detour distance after replanning compared to the original global path is 4.00m, and the additional turning angle caused by the local detour is 180°. This application The time taken to complete a single local replanning operation is 1.48ms, which is also within the millisecond range of computation time in this simulation scenario. It did not achieve a shorter replanning computation time; the computational overhead was on the same order of magnitude as the traditional grid scheme. The additional detour distance generated after local replanning is 2.83m, which is 29.2% shorter than the control group. The additional turning angle introduced by local detour is only 60°, which is 66.7% less than the control group. It can be seen that the proposed solution does not have a significant advantage in terms of computation time in local replanning scenarios. Its core advantage lies in the geometric indicators of the local detour path. When encountering sudden obstacles that require detour, it can generate local detour routes with lower detour degree and fewer turning actions, reducing the frequency of UAV heading adjustments.

[0133] Analyzing the overall simulation results The reason why the computation time did not decrease significantly is that the face-centered cubic lattice has 12 neighborhood directions. Each round of node expansion requires more neighborhood node traversal and collision verification calculations, which increases the computational load to some extent. However, the overall computational load remains at the millisecond level, which is sufficient to meet the real-time replanning requirements of the UAV onboard unit. The simulation results of global path and local replanning both reflect that, based on the non-orthogonal topology of the 12 equidistant neighborhoods of the face-centered cubic lattice, the algorithm can output discrete paths with less spatial detours and significantly reduced turning actions. For the UAV platform, a reduced turning angle means fewer heading adjustments during flight, which can reduce the attitude adjustment burden of the flight control system and reduce the energy consumption caused by frequent speed changes during flight. It should be noted again that all the above simulation values ​​and percentages are based on the specific simulation environment of this group. When the obstacle layout, start and end points, lattice step size, and safety margin change, the various indicators and comparison ratios will change accordingly. Therefore, the quantitative results of this group of simulations cannot be directly generalized to other unknown simulations or real-world operation scenarios.

[0134] This embodiment compares with the traditional 6-connected grid. The global path length and cumulative turning angle of this scheme are significantly optimized, and the extra detour distance and extra turning angle under local replanning are greatly reduced. Although the increase in the number of neighborhoods slightly increases the replanning time, it still remains in the millisecond level, which can meet the real-time computing requirements of airborne systems, and verifies the significant advantages of this scheme in path geometry characteristics.

[0135] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A three-dimensional path planning method for unmanned aerial vehicles (UAVs) based on face-centered cubic lattices, characterized in that, Includes the following steps: S1. Acquire 3D spatial data, starting node, target node, and obstacle information; S2. Discretize the three-dimensional spatial domain according to the face-centered cubic lattice rule to obtain a set of discrete lattice nodes, and map obstacle information to the discrete lattice nodes, marking impassable nodes; each central node of the face-centered cubic lattice is configured with twelve equidistant neighboring nodes, which are distributed in the diagonal directions of the three coordinate planes; determine the actual step size of lattice adjacency based on the UAV's outer envelope size, obtain the single-axis coordinate increment based on the actual step size, and set the safety margin used for collision detection and local bounding boxes; S3. Adopting a face-centered cubic lattice topology. The algorithm performs a global path search to obtain a collision-free global three-dimensional discrete flight path; The algorithm is based on The search framework expands the neighborhood according to the twelve neighborhood nodes of the face-centered cubic lattice. During the neighborhood expansion, a two-layer collision detection is performed between the node and the path edge to calculate the cost of the adjacent edge. The three-dimensional Euclidean distance is used as a heuristic to estimate the cost. S4. Control the UAV to fly along a global three-dimensional discrete flight path, and continuously detect whether new obstacle areas appear ahead of the flight through the forward-looking window; S5. When a collision risk is detected, determine the reconnection node, construct a local bounding box, and execute the operation inside the local bounding box. Local path search yields a local detour path. This local detour path is then concatenated with the global 3D discrete flight path to obtain an updated collision-free 3D discrete flight path. If the local search fails, the local bounding box is expanded for retry. If the retry still fails, a global replanning is triggered. If the global replanning still fails, a planning failure is output and a hover command is issued. S6. If no collision risk is detected, the global three-dimensional discrete flight path without collision is directly output.

2. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 1, characterized in that, In S2, the three-dimensional spatial domain is discretized according to the face-centered cubic lattice rule, including determining the neighborhood displacement relationship of the face-centered cubic lattice, screening the legal nodes of the lattice, and mapping the physical coordinates to the lattice index coordinates. The set of neighborhood basis vectors of the face-centered cubic lattice is: in, For single-axis coordinate increment parameters, each set of positive and negative values ​​is independently taken to generate all neighborhood displacements, and all twelve neighborhood nodes corresponding to the center node are generated from the neighborhood basis vector.

3. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 2, characterized in that, The set of legal nodes in a face-centered cubic lattice is given by the formula: In the formula, For lattice index integers, For single-axis coordinate increment parameters; When discretizing a three-dimensional spatial domain, traverse the integer indices. ,according to Valid FCC lattice nodes are obtained through constraint screening, invalid nodes that do not meet the constraints are removed, and a mapping relationship is established between lattice index coordinates and Cartesian physical coordinates.

4. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 3, characterized in that, S2 further includes lattice node mapping processing for the starting point and the target point; when the Cartesian coordinates of the original starting position or the original target position do not coincide with any FCC lattice node, the following processing is performed: Enumerate all candidate lattice nodes corresponding to the surrounding cells of the original position, calculate the Euclidean distance between each candidate lattice node and the original position, and sort the candidate lattice nodes in ascending order of distance. The candidate lattice nodes are subjected to node safety envelope collision detection and collision detection between the candidate nodes and the original positions in the sorting order. Select the first candidate lattice node that passes all tests and use it as the starting node or the target node after mapping, respectively; if there is no candidate lattice node that meets the conditions, output that the mapping failed.

5. A three-dimensional path planning method for unmanned aerial vehicles based on a face-centered cubic lattice according to claim 1 or 3, characterized in that, The specific process in S2 for determining the actual step size of lattice adjacency and the single-axis coordinate increment parameter based on the UAV's external envelope size is as follows: Obtain the outer envelope dimensions of the UAV, including the propellers and payload, in three orthogonal directions. , , Calculate the envelope feature size Set the actual step size of lattice adjacency. The single-axis coordinate increment parameters satisfy ; Within the complete execution cycle of a single global path search or a single local path search, and Keep it fixed; when the drone's shape envelope changes, recalculate it before the next path planning initiation. and Update the FCC lattice node set.

6. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 5, characterized in that, The dual-layer collision detection of nodes and path edges includes a node collision determination process and a path edge collision determination process. Node collision determination: Calculate the minimum circumscribed sphere radius of the UAV's external envelope. Set a safety margin The radius of the safe enclosing sphere is obtained. ; Construct a sphere with the lattice node to be detected as the center and a radius of ; The safety bounding sphere; if the safety bounding sphere intersects with the obstacle region, then the lattice node is marked as a non-passable node; Path edge collision determination: Using the connecting line segment between two adjacent lattice nodes to be detected as the axis, a radius equal to... The capsule-shaped sweep body; if the capsule-shaped sweep body intersects with the obstacle area, then the path edge of that section is determined to be impassable.

7. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 1, characterized in that, In S3 The algorithm performs a global path search, including initialization of open and closed sets, iterative expansion of nodes, neighborhood verification and filtering, cost value update, and search termination judgment. The algorithm evaluation function is ;in To reach the node from the starting node after mapping The cumulative cost of adjacent edge Euclidean length; For nodes The heuristic cost of the 3D Euclidean distance to the mapped target node is given by the following formula: In the formula For the current node Cartesian physical coordinates These are the Cartesian physical coordinates of the target node after mapping; Open sets adopt the following approach: This is a priority queue for sorting key values. During each node expansion, the twelve neighboring nodes are traversed, filtering out neighboring nodes that are within the closed set, impassable due to node collisions, or impassable due to path edge collisions. The remaining valid neighboring nodes are then processed. Value, parent node, Values ​​are updated and the open set is maintained.

8. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 1, characterized in that, The process of constructing the look-ahead window and performing collision detection in S4 is as follows: reading the look-ahead detection length of the airborne sensor. The actual step size of lattice adjacency is denoted as Number of forward window nodes ; When the UAV flies along a global three-dimensional discrete flight path, the index of the current path point of the UAV is read in real time, and the continuous path points from the current index are selected on the global path. For each path point, collision detection is performed sequentially on all path nodes and path edges between adjacent nodes within that range. When any node or path edge is detected to intersect with a newly appearing obstacle area, a collision risk is determined, triggering a local replanning process.

9. The UAV three-dimensional path planning method based on face-centered cubic lattice according to claim 8, characterized in that, The specific process for determining the reconnection node in S5 is as follows: Obtain the index of the first path point that collides within the lookout window as the starting index of the affected path segment. Traverse the global path backward to determine the index of the end of the path segment where consecutive collisions occur. ; From the index Begin by selecting candidate path points one by one along the global path toward the target node; For each candidate path point, perform node collision detection and path edge collision detection between the candidate path point and the previous path point in sequence; determine the first candidate path point that passes both the node and the path edge detection as the reconnection node; if no candidate path point that meets the conditions is found after traversing to the global path endpoint, output that there is no available reconnection node.

10. A three-dimensional path planning method for unmanned aerial vehicles based on a face-centered cubic lattice according to claim 9, characterized in that, The S5 process, which involves constructing local bounding boxes, performing local searches, and performing path concatenation and degradation processing, specifically includes: Using the lattice node and reconnection node corresponding to the current position of the UAV as the diagonal vertex, a cuboid is generated with an aligned axis. The safety margins of the cuboid are then extended outwards in the X, Y, and Z coordinate directions. This yields the local bounding box used in the first local search. Using the local bounding box as a constraint in the search space, call... The algorithm performs a local path search with the current location node of the drone as the search starting point and the reconnection node as the search ending point; When a feasible local detour path is found through local path search, the local detour path and the original global path after the reconnected node are deduplicated to complete the path splicing, and collision detection is performed again on the path edges at the splicing position. When the first local path search fails, the safety margin for outward expansion is replaced with... Reconstruct the local bounding box and perform the local search again; if the search still fails, perform the global search. Replan; if global replanning still fails, output a planning failure status and issue a hover command to the flight controller.

Images