A dynamic risk assessment clustering unmanned aerial vehicle path planning method and system

Abstract

The application belongs to the technical field of unmanned aerial vehicle path planning, and discloses a dynamic risk assessment clustering unmanned aerial vehicle path planning method and system, which comprises the following steps: generating a risk sub-region and calculating a risk weight, running an improved D-star algorithm to obtain an initial path in a constructed grid map, simplifying the path by using a path compression algorithm, deleting redundant nodes, and generating a new path segment by performing path collision detection, and obtaining a final optimized path by curvature constraint, risk weight coupling and smoothing processing according to the new path segment. The application optimizes the D-star path planning algorithm, improves the path planning efficiency and safety, and has good performance in narrow and concave areas.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle path planning, in particular to a dynamic risk assessment clustering unmanned aerial vehicle path planning method and system. BACKGROUND

[0002] With the wide application of unmanned aerial vehicle technology in various fields, path planning is directly related to the operation efficiency and flight safety of unmanned aerial vehicles. Traditional algorithms are difficult to achieve efficient collaborative optimization of path length, smoothness and obstacle avoidance safety. Therefore, how to improve planning efficiency while ensuring obstacle avoidance safety and achieve path planning with reliability and practicality has become a key technical bottleneck that needs to be broken through in the current unmanned aerial vehicle field.

[0003] As a classic dynamic path planning method, D-star algorithm is widely used in unmanned aerial vehicle navigation in complex environments due to its environmental adaptability and real-time re-planning capability. However, it still has significant limitations: first, the calculation burden of node expansion and update in large-scale maps is heavy and inefficient; second, it is easy to fall into local optimum when facing narrow paths or concave regions; third, it only focuses on dynamic obstacle avoidance and path feasibility, without considering unmanned aerial vehicle motion constraints and path smoothness. * RRT algorithm effectively limits the sampling space through ellipse pruning technology, thereby accelerating the convergence of the optimal path. However, ellipse pruning technology has not been widely applied in dynamic path planning algorithms. SUMMARY

[0004] To solve the above problems, the present application provides a dynamic risk assessment clustering unmanned aerial vehicle path planning method and system, which optimizes the D-star path planning algorithm, improves the path planning efficiency and safety, and performs well in narrow and concave regions.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a dynamic risk assessment clustering unmanned aerial vehicle path planning method, comprising the following steps:

[0007] A two-dimensional map is constructed based on the grid method, and the grid state is defined. The ellipse parameters are initialized, the grid risk density score is calculated, and the risk sub-regions are generated based on the score through density clustering. Then, the average risk density of the risk sub-regions, the average path curvature, and the total repulsive force are calculated to obtain the risk weight.

[0008] Based on the eight-neighborhood ray expansion, candidate nodes are generated. The search direction priority is determined according to the angle difference between the current node and the target point, and the ellipse pruning parameters are adaptively adjusted. The total cost function is calculated, and the node with the minimum total cost function is expanded. After the target point is expanded, the initial path is generated by backtracking.

[0009] The initial path is compressed to obtain a simplified path based on an obstacle-free straight path, a preset interval threshold is set, secondary compression is performed through redundancy filtering, and a compressed path is obtained; sampling detection is performed and it is judged whether to add a safety point according to a risk weight of a risk sub-region where the compressed path is located; when an obstacle is detected, the risk sub-region where the obstacle is located is located, the risk weight of the region is updated, it is judged whether to trigger local re-planning, and a new path segment is generated;

[0010] According to the new path segment, a final optimized path is obtained through curvature constraint, risk weight coupling and smoothing processing.

[0011] As an optional implementation, the formula for calculating the grid risk density score is:

[0012] ; wherein, is the current grid for calculating the risk density, is the grid in the neighborhood, is the set of grids within the risk perception range, is the obstacle indication function; the formula for calculating the average risk density of the risk sub-region is: ; wherein, is the average risk density, is the total number of grids in the risk sub-region, is the risk sub-region; the formula for the average path curvature is: ; wherein, is the average path curvature, is the number of path segments in the risk sub-region, is the curvature value of the i th path segment; the total repulsive force is the vector sum of all individual repulsive forces, and the formula for calculating the individual repulsive force is: ; wherein, is the current grid of the unmanned aerial vehicle, is the individual obstacle grid, is the set composed of the grids of the obstacles, is the repulsive force strength coefficient, is the repulsive force decay coefficient.

[0013] As an optional implementation, the obtained risk weight is represented as:

[0014] ; wherein, is the average risk density, is the average path curvature, is the overall repulsion strength of the region; the overall repulsion strength of the region is the average value of the total repulsive force of all grids in the risk sub-region.

[0015] As an alternative embodiment, the ellipse pruning parameter is adaptively adjusted based on the overall repulsion strength of the region, denoted as:

[0016] ; wherein, is the semi-minor axis of the adaptively adjusted ellipse, is the semi-minor axis of the initialized ellipse parameter, is the repulsion strength coefficient, is the overall repulsion strength of the region.

[0017] As an alternative embodiment, the node total cost function is denoted as:

[0018] ;

[0019] wherein, is the actual cost, is the heuristic estimated cost, is the ellipse region penalty cost, is the clustering weight penalty term, denoted as:

[0020] ; wherein, is the weight coefficient, is the risk weight.

[0021] As an alternative embodiment, the curvature constraint and risk weight coupling is denoted as:

[0022] ; wherein, is the maximum allowed curvature, is the minimum turning radius, is the risk weight to which the new path segment belongs;

[0023] If the new path segment passes through two risk sub-regions whose risk weights satisfy , then 2 offset control points are added at the boundaries of the two risk sub-regions, denoted as:

[0024] ; wherein, , are the end points of the new path segment in the two risk sub-regions, respectively, is the offset distance of the control point, defined as is the normal vector of the connecting line.

[0025] In a second aspect, the present application provides a dynamic risk assessment clustering UAV path planning system, comprising the following modules:

[0026] The clustering module is configured to: construct a two-dimensional map based on the raster method, define the raster state, initialize the ellipse parameters, calculate the raster risk density score, generate risk sub-regions through density clustering based on this score, and then calculate the average risk density, average path curvature, and total repulsion of the risk sub-regions to obtain the risk weight.

[0027] The initial path generation module is configured to: generate candidate nodes based on eight-neighbor ray expansion, determine the priority of the search direction according to the angle difference between the current node and the target point, adaptively adjust the ellipse pruning parameters, calculate the total cost function of the nodes and expand the node with the smallest total cost function, and backtrack to generate the initial path after the target point has been expanded.

[0028] The compression and detection module is configured to: compress the initial path based on the barrier-free straight path to obtain a simplified path, preset the spacing threshold, and perform secondary compression through redundant filtering to obtain the compressed path; perform sampling detection and determine whether to add a safety point based on the risk weight of the risk sub-region where the compressed path is located; when an obstacle is detected, locate the risk sub-region where it is located, update the risk weight of the region, determine whether to trigger local replanning, and generate a new path segment.

[0029] The path optimization module is configured to obtain the final optimized path based on the new path segment through curvature constraints, risk weight coupling, and smoothing.

[0030] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0031] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0032] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] This invention employs multi-dimensional risk clustering, integrating parameters such as obstacle risk density and path curvature to accurately delineate high-risk areas, unlike traditional density clustering algorithms that divide regions based on only a single density index. Elliptical pruning adaptively adjusts the range according to environmental risk, avoiding the expansion of invalid nodes; priority ranking of search directions reduces redundant exploration, significantly lowering the computational burden in large-scale maps and improving path planning efficiency.

[0035] This invention introduces a magnetic field repulsion model, with multi-directional force coupling to help escape the local optimum of the concave region; at the same time, safety points are added at the boundary of the risk area to guide it to travel safely along the narrow channel; the safety points across the region further avoid the risk of boundary collision, which solves the path planning bottleneck of traditional algorithms in complex terrain and ensures the safety and feasibility of the path.

[0036] This invention employs a non-uniform rational B-spline with risk weights during the smoothing process, dynamically adjusting the maximum allowable curvature and control point weights based on the risk weights. Simultaneously, path compression involves removing redundant nodes twice, retaining key turning points. The generated path satisfies motion constraints such as the minimum turning radius of the UAV, avoiding sharp turns; paths in high-risk areas are more stable, while paths in low-risk areas are shorter, balancing smoothness and length optimization, reducing UAV energy consumption and mechanical wear, and improving flight stability.

[0037] This invention correlates elliptical pruning parameters with regional repulsion strength, and dynamically adapts the ellipse size to environmental risks, avoiding excessive pruning that could lead to path infeasibility. A fault-tolerant penalty mechanism prevents insufficient pruning, balancing search efficiency and path integrity, thus solving the problem of fixed parameters in traditional elliptical pruning.

[0038] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is a flowchart of the dynamic risk assessment clustering UAV path planning method provided in Embodiment 1 of the present invention;

[0041] Figure 2 The experimental results of the traditional D-star algorithm are shown in the figure.

[0042] Figure 3 The experimental results of the improved D-Star algorithm provided in Embodiment 1 of the present invention are shown in the figure.

[0043] Figure 4 Repulsion diagram for safety passage;

[0044] Figure 5 Repulsion diagram for unsafe passages;

[0045] Figure 6 This is a diagram of the trajectory of the force field motion.

[0046] Figure 7 This is a diagram showing the motion trajectory of the concave region. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0050] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment provides a dynamic risk assessment clustering UAV path planning method, including the following steps:

[0053] S1: Construct a two-dimensional map based on the raster method, define the raster state, initialize the ellipse parameters, and calculate the raster risk density score. Based on this score, generate risk sub-regions through density clustering, and then calculate the average risk density, average path curvature, and total repulsion of the risk sub-regions to obtain the risk weight.

[0054] In step S1, a two-dimensional map is constructed using a grid method. Specifically, the physical space is divided using a two-dimensional grid, with a grid size of s (unit: meters, set according to the accuracy of the UAV sensor, such as 0.5m × 0.5m).

[0055] The defined grid states include free space (0), obstacles (1), starting point (S), and target point (T), and the data structures are initialized, including:

[0056] Both the OpenList (storing nodes to be expanded) and the CloseList (storing processed nodes) are empty sets; the initial cost G of the starting point S is 0, and it is added to the OpenList. This step corresponds to... Figure 1The first box in "Environment Modeling" is the input basis for the entire process, and all subsequent path planning steps are based on this grid map.

[0057] In step S1, the ellipse parameters are initialized, with the start and end points as the foci of the ellipse, where the semi-focal length is... semi-major shaft Semi-short shaft The coordinates of the center point of the ellipse are Specifically:

[0058] Let the starting coordinates be ( , ), target point coordinates ( , ),calculate:

[0059] Half focal length ;

[0060] Semi-major axis a = 2c (ensuring the ellipse covers the basic path range from the starting point to the target point);

[0061] Semi-short shaft ;

[0062] center point of the ellipse ;

[0063] Equation of an ellipse: [(x- ) / a]²+ [(y- ) / b]²≤1;

[0064] And define the initial repulsive force guiding direction. ,in Let the angle with the elliptical axis be defined, and initialize the overall repulsive strength of the region. .

[0065] In step S1, for each grid u(x,y), its neighborhood N( (centered on u) Raster risk density score within the radius of all graticles:

[0066] ;in, , For grid, For grid and Euclidean distance (unit: number of grids × s); This is the obstacle indicator function (1 for obstacles, 0 otherwise);

[0067] The risk perception radius is expressed as: ,in, Given the minimum turning radius of the drone, it can be seen that the risk perception radius is related to the minimum turning radius of the drone and the sensor resolution, and this is to ensure accurate identification in narrow passages.

[0068] according to Risk sub-regions are generated through density clustering, specifically:

[0069] Set risk threshold Filter out ≥ The grid.

[0070] High-risk rasters are aggregated using a density-based clustering method:

[0071] Clustering neighborhood radius =2×s (ensuring adjacent high-risk rasters are grouped into the same region), minimum number of rasters included (MinPts=3) (avoiding isolated high-risk rasters forming invalid regions), generating a set of risk sub-regions { } (k=1,2,...,n, each (A set of consecutive high-risk grids).

[0072] For each risk sub-region The formula for its average risk density is:

[0073] ;in, For average risk density, This represents the total number of grid cells within the risk sub-region. This is a risk sub-region; the formula for its average path curvature is:

[0074] ;in, The average path curvature This represents the number of path segments within the risk sub-region. Let be the curvature value of the i-th path segment, expressed as: ;in, The path turning angle, The length of the line segment.

[0075] In step 1, a magnetic field repulsion model is introduced, and the total repulsion force is obtained through initialization and calculation; this includes:

[0076] Repulsion parameter settings:

[0077] 1. Repulsion intensity coefficient : Positively correlated with the size of the drone fuselage.

[0078] 2. Repulsive force attenuation coefficient : =1 / (2· ),in The risk perception radius (correlated with the drone's minimum turning radius and sensor resolution) ensures significant repulsive force at close range and rapid attenuation at long range.

[0079] Calculation of repulsive force for a single obstacle grid: For the grid U where the UAV is currently located ( , ), and a single obstacle grid V ( , The vector repulsive force F( , The calculation is as follows:

[0080] Unit repulsion direction vector: ;in, The direction is outward along the line connecting the drone and the obstacle.

[0081] Magnitude of a single repulsive force: ;in, The grid where the drone is currently located. For a single obstacle grid, This represents the set of all the grid cells identified as obstacles after a two-dimensional map is constructed using the grid method. The repulsive force intensity coefficient, The repulsive force attenuation coefficient is... The Euclidean distance between the two grid cells is expressed in meters. The unit repulsion direction vector extends outward along the line connecting the UAV and the obstacle, i.e.:

[0082] ;

[0083] When there are n obstacle grids around the drone At that time, the total repulsive force is the vector sum of the individual repulsive forces:

[0084] ;

[0085] Total repulsive force direction This represents the vector and direction angle of the repulsive force of all obstacles within the area.

[0086] In this embodiment, to avoid falling into the concave trap of the algorithm, inspired by magnetic fields and based on the engineering requirements of UAV path planning, the core characteristic of magnetic fields being "like poles repel each other" is functionally reconstructed. A magnetic field model is introduced to quantify the repulsive force of obstacles to measure connectivity. The repulsive force between the two decreases with increasing distance and increases with increasing obstacle density. The force of a real magnetic field is directional.

[0087] Concave Region Escape Mechanism: The grid U inside a concave obstacle will be repulsed by multiple obstacles on the concave edge. The direction of the total repulsive force after superposition is... Pointing towards the concave opening (outwards), it forces the drone to move outwards from the area.

[0088] Narrow passage guidance mechanism: The grid U in the narrow passage is repulsed by the obstacles on both sides. The repulsive force components perpendicular to the passage cancel each other out, and the components along the passage direction are superimposed to form a "guiding force" that guides the drone to pass through the passage.

[0089] Actual channel width D ≥ ( (s is the grid size), the vertical components of the repulsive forces exerted on the drone by the obstacles on both sides cancel each other out, while the horizontal components are superimposed to form a "guiding force" along the channel direction. Figure 4 (the resultant force arrow in the image), at this time The drone can travel safely along the passage, and the combined force guides the drone to pass smoothly.

[0090] Actual channel width D < The obstacles on both sides are too close together, and the vertical component of the repulsive force cannot be completely canceled out, so the resultant force points outward from the channel. Figure 5 (The resultant force arrows at both ends of the diagonal direction), at this time > This prevents drones from entering and avoids the risk of collision.

[0091] unsafe passage The value is significantly increased and is directly used to calculate the risk weight, so that safe channels correspond to low-risk weights and unsafe channels correspond to high-risk weights.

[0092] Upon entering a narrow passage, the drone experiences slight resistance due to the force fields of obstacles on both sides. At this point, the path planning module triggers a speed adjustment, causing the drone to decelerate and reduce inertia. Figure 6 As shown.

[0093] As the drone traverses the passage, the force lines gradually follow the passage's direction, transforming the resistance into 'auxiliary thrust'. At this point, the drone accelerates, quickly passing through the narrow area, such as... Figure 7 As shown.

[0094] In this embodiment, the micro-grid repulsion is transformed into macro-regional features, and for each risk sub-region set { Calculate the overall repulsion strength of the region: in for{ The total number of grid cells within} Let be the magnitude of the total repulsive force on grid u. A larger value indicates a denser concentration of obstacles and stronger spatial constraints (such as concave areas or narrow passages), which is directly used for risk weighting. calculate.

[0095] The final risk weights are expressed as follows:

[0096] ;in, For average risk density, The average path curvature The overall repulsion strength of the region.

[0097] S2: Generate candidate nodes based on eight-neighbor ray expansion, determine the priority of the search direction according to the angle difference between the current node and the target point, and adaptively adjust the ellipse pruning parameters. Calculate the total cost function of the nodes and expand the node with the smallest total cost function until the target point is expanded. Then backtrack to generate the initial path.

[0098] In step S2, this embodiment improves the traditional D-star algorithm. This embodiment utilizes the improved D-star algorithm by first modifying the traditional eight-neighbor expansion mode into an eight-neighbor ray expansion mode, and then generating candidate nodes based on the eight-neighbor ray expansion, specifically as follows:

[0099] Eight-neighbor ray expansion: Sets the minimum step size for the ray pattern. =1, maximum step size =5; Rays are emitted from the current node (x,y) in 8 neighboring directions (0°, 45°, 90°, ..., 315°), moving step grids each time to generate candidate nodes. If a candidate node exceeds the map boundary or falls into an obstacle grid, the expansion in that direction is terminated; otherwise, the candidate node is added to the OpenList.

[0100] Then, the search direction priority is determined based on the angle difference between the current node and the target point. Specifically: Step 1: Determine the reference direction: Calculate the direction of the line connecting the current node (x, y) to the target point T. Step 2: Calculate the angle difference: For each extension direction (e.g., 0° corresponds to) =0), calculate the angle difference Δ =| - Step 3: Sorting rules: Sort by Δ Sort by size from smallest to largest, prioritizing expansion in directions with smaller angle differences (closer to the target direction); if the current region... If the value is ≥0.7, the sorting restrictions are relaxed, and candidate nodes in all 8 directions are retained.

[0101] In step S2, based on the overall repulsion strength of the region The adaptive adjustment of ellipse pruning parameters is specifically based on the semi-minor axis b in the initialized ellipse parameters in S1.

[0102] ;in, The semi-minor axis of the ellipse after adaptive adjustment. To initialize the semi-minor axis of the ellipse parameters, The repulsive force intensity coefficient, The overall repulsion strength of the region.

[0103] Based on cluster regions The ellipse size is adjusted adaptively, and nodes outside the circle are pruned. A minimum threshold for 'b' is set to ensure that the semi-minor axis 'b' does not shrink indefinitely, avoiding situations where feasible nodes cannot be found due to an excessively small ellipse. To prevent overly strict pruning constraints, the ellipse in this embodiment is a fault-tolerant ellipse, allowing for the expansion of certain nodes outside the circle when necessary, but penalizing them by lowering their priority.

[0104] In this embodiment, the pruning rule is as follows: for a candidate node (x, y), if it does not satisfy the equation of the ellipse, it is marked as an "outer node of the ellipse". This can be distinguished using the simplest binary method. A penalty cost P is set for nodes inside the circle. e A penalty cost P is imposed on the outside of the circle if the value is 0. e The value is 1.

[0105] For each candidate node, check if it is an obstacle or already in the CloseList; otherwise, continue calculating the cost. The total cost function for a node is expressed as:

[0106] ;

[0107] in, The actual cost from the starting point to the current node (cumulative distance traveled, in meters); This is used for heuristic cost estimation (Euclidean distance from the current node to the target point). The penalty cost is applied to the elliptical region; The clustering weight penalty term is represented as:

[0108] ;in, These are the weighting coefficients. Risk weights.

[0109] The larger the value, the denser the obstacles in the area and the stronger the repulsive force (such as concave areas or narrow passages). Increase the path cost in such areas to guide the algorithm to prioritize areas with lower repulsion and higher security.

[0110] If the candidate node If a value is less than an existing value in the OpenList, then update and reorder the OpenList; select The node with the smallest value is selected as the next expanded node and added to CloseList.

[0111] When the target point T is added to the CloseList, the parent node pointers of the nodes in the CloseList are traced back to generate the initial path P=[S,p1,p2,...,T].

[0112] Table 1, obtained from the simulation results of this embodiment, is as follows:

[0113] Table 1

[0114]

[0115] In path planning scenarios, the improved D-Star algorithm outperforms the traditional D-Star algorithm, such as... Figure 2 and 3 As shown, its running time is optimized by 75%, the number of turns is reduced by 37.5%, and the number of expanded nodes is optimized by 72.7%; moreover, the total path cost is reduced by 11.5%. The simulation results demonstrate that the improved D-Star algorithm significantly improves running efficiency, path smoothness, and path optimization, enabling it to plan better paths more efficiently.

[0116] S3: Based on the barrier-free straight path, the initial path is compressed to obtain a simplified path. A preset spacing threshold is used for secondary compression through redundant filtering to obtain a compressed path. According to the risk weight of the risk sub-region where the compressed path is located, sampling detection is performed and it is determined whether to add a safety point. When an obstacle is detected, the risk sub-region where it is located is located, the risk weight of the region is updated, it is determined whether to trigger local replanning, and a new path segment is generated.

[0117] In step S3, path compression aims to streamline the initial path by removing redundant nodes while retaining key steering information, laying the foundation for subsequent smoothing.

[0118] The initial path is compressed based on an accessible straight path, specifically as follows: First, the initial path node sequence is obtained through reverse search, that is, tracing back from T to S to obtain the node sequence P={ , ,..., }( =S, =T); then for three consecutive nodes , , Use Bresenham algorithm to determine arrive Does the straight line pass through an obstacle?

[0119] If there are no obstacles, then delete. The updated sequence is [ , ,...];

[0120] If there are obstacles, then leave it as is. Continue testing the next group of nodes;

[0121] Repeat the above process until there are no redundant nodes to delete, thus obtaining a simplified path.

[0122] A preset spacing threshold is used to perform secondary compression through redundant filtering, resulting in a compressed path, specifically:

[0123] First, preset the spacing threshold setting: =1.5 Then perform redundancy filtering: calculate The Euclidean distance d between adjacent nodes in the interval, if d < If the change in path direction between two nodes is less than 5° (non-critical turning point), then delete the latter node; retain critical turning points (direction change ≥ 5°) to obtain the final compressed path.

[0124] Based on the risk weight of the risk sub-region where the compressed path is located, sampling detection and determination of whether to add a safety point are performed. Specifically, the sampling density is first allocated: for path segments in the compressed path ( , If the area is ≤0.4 (low risk), number of sampling points N=2L (L is the path segment length, unit: meters, e.g., L=10 meters, then 20 sampling points); if ≥0.7 (high risk), number of sampling points N=3L (rounded up, e.g., if L=10 meters, then 30 points are sampled).

[0125] Then, collision detection is performed. For each sampling point (x, y), the distance between it and the nearest obstacle is calculated. :

[0126] like < =1.5 If the path segment is marked as "dangerous" and needs to be replanned, it is otherwise considered safe.

[0127] Then if the path segment is compressed ( , Crossing two areas and , and | - If the value is ≥0.3 (significant risk difference), then a safety point will be added.

[0128] Safepoint generation:

[0129] Midpoint M of the boundary (( + ) / 2, ( + ) / 2);

[0130] Low-weight side safety point From M to Directional offset δ = 0.5 × ;

[0131] High-weighted side safety point From M to Directional offset δ = 0.5 × ;

[0132] If there is an obstacle in the direction of offset, switch to the opposite direction to offset, ensuring... and satisfy ;

[0133] Will and Insert compressed path segment ( , In ), the update path is ( , , , ).

[0134] When an obstacle is detected, its risk sub-region is located, the risk weight of that region is updated, it is determined whether local replanning is triggered, and a new path segment is generated. Specifically:

[0135] Obstacle localization: After the sensor detects a new obstacle, it maps it to a grid v(x,y) to determine the area to which it belongs. Weight update: For Recalculate (Incorporating new obstacles) =1) and If | - A local replanning is triggered if the value is ≥0.3 or if a new obstacle falls directly into the original planned path segment. Local replanning scope: Only updates... and its adjacent cluster regions (sharing boundaries) The node cost is ±1), and OpenList assigns a priority coefficient β=1+0.5 to nodes in these regions. (High-risk nodes are expanded first). Cost update: Recalculate the F-value of affected nodes (key update). and ), generate a new path segment to replace the original dangerous segment, and combineFigure 4 The security assessment of the channel.

[0136] S4: Based on the new path segment, the final optimized path is obtained through curvature constraints, risk weight coupling, and smoothing.

[0137] In step S4, a coupling mechanism between risk weights and curvature constraints is introduced to smoothly optimize the path, ensuring that the final path meets the kinematic characteristics of the UAV (such as turning radius limitations). Specifically, this includes:

[0138] Maximum permissible curvature Adjusted to:

[0139] ;in, For the maximum permissible curvature, Minimum turning radius, This represents the risk weight of the new path segment. High-weight areas. Tighten curvature constraints To avoid sharp turns by drones; to relax constraints in low-weight areas to optimize path length.

[0140] If the risk weight of the new path segment is satisfied If there are two risk sub-regions, then two offset control points are added at the boundary between the two risk sub-regions, as shown below:

[0141] ;in, , These are the endpoints of the new path segment in the two risk sub-regions, respectively. The offset distance of the control point. for The normal vector of the line connecting them.

[0142] This embodiment uses non-uniform rational B-splines for path smoothing, and achieves adaptive optimization of the curve shape by adjusting the control point weights, specifically:

[0143] For path segments within high-weight clusters, increase the control point weights and reduce curve offset, using the following formula:

[0144] ;

[0145] For path segments within low-weight clusters, reduce the weights to optimize path length:

[0146] ; Based on the weights.

[0147] The equation for the NURBS curve is:

[0148] ;

[0149] in, These are the points on the B-spline curve corresponding to the parameter u. It is a 3rd order spline basis function. It is the i-th control point, and the denominator is the weight normalization term to ensure the geometric meaning of the curve coordinates.

[0150] In summary, this embodiment significantly improves the overall performance of UAV path planning by combining dynamic risk assessment clustering with the improved D-Star algorithm: The search range is dynamically adjusted using risk weights for clustered regions, offering particularly significant advantages in large-scale map scenarios; a magnetic field repulsion model is introduced, where multi-directional repulsion in concave regions superimposes to form an "escape thrust," and repulsion on both sides of narrow channels synthesizes a "channel-guided force." Pruning, combined with repulsion constraints, focuses on effective directions and eliminates ineffective ones. By allowing the ellipse size and fault tolerance mechanism to directly respond to the strength and direction of the repulsion, it achieves "where the repulsion points, the ellipse reserves space for that direction"; coupled with an improved non-uniform rational B-spline smoothing algorithm to dynamically adjust control point weights, turning angles are reduced, lowering UAV energy consumption and mechanical losses, achieving a synergistic optimization of planning efficiency, flight safety, and path quality in complex environments.

[0151] Example 2

[0152] This embodiment provides a dynamic risk assessment clustering UAV path planning system, including the following modules:

[0153] The clustering module is configured to: construct a two-dimensional map based on the raster method, define the raster state, initialize the ellipse parameters, calculate the raster risk density score, generate risk sub-regions through density clustering based on this score, and then calculate the average risk density, average path curvature, and total repulsion of the risk sub-regions to obtain the risk weight.

[0154] The initial path generation module is configured to: generate candidate nodes based on eight-neighbor ray expansion, determine the priority of the search direction according to the angle difference between the current node and the target point, adaptively adjust the ellipse pruning parameters, calculate the total cost function of the nodes and expand the node with the smallest total cost function, and backtrack to generate the initial path after the target point has been expanded.

[0155] The compression and detection module is configured to: compress the initial path based on the barrier-free straight path to obtain a simplified path, preset the spacing threshold, and perform secondary compression through redundant filtering to obtain the compressed path; perform sampling detection and determine whether to add a safety point based on the risk weight of the risk sub-region where the compressed path is located; when an obstacle is detected, locate the risk sub-region where it is located, update the risk weight of the region, determine whether to trigger local replanning, and generate a new path segment.

[0156] The path optimization module is configured to obtain the final optimized path based on the new path segment through curvature constraints, risk weight coupling, and smoothing.

[0157] It should be noted that the above modules correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above modules and their corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules can be executed in a computer system as part of the system.

[0158] In further embodiments, the following is also provided:

[0159] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0160] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0161] A computer-readable storage medium for storing computer instructions that, when executed by a processor, perform the method of Embodiment 1.

[0162] The method in Example 1 can be directly executed by a hardware processor, or it can be executed by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0163] A computer program product includes a computer program that, when executed by a processor, implements the method in Embodiment 1.

[0164] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0165] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0166] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0167] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this embodiment.

[0168] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A dynamic risk assessment clustering method for UAV path planning, characterized in that, include: A two-dimensional map is constructed based on the raster method, and the raster state is defined. Ellipse parameters are initialized and the raster risk density score is calculated. Based on this score, risk sub-regions are generated by density clustering. Then, the average risk density, average path curvature, and total repulsion of the risk sub-regions are calculated to obtain the risk weight. Candidate nodes are generated based on eight-neighbor ray expansion. The priority of the search direction is determined according to the angle difference between the current node and the target point. The ellipse pruning parameters are adaptively adjusted. The total cost function of the nodes is calculated and the node with the smallest total cost function is expanded until the target point is expanded. Then the initial path is backtracked to generate the path. A simplified path is obtained by compressing the initial path based on an accessible straight path. A preset spacing threshold is then used for secondary compression through redundant filtering to obtain the compressed path. Based on the risk weight of the risk sub-region where the compression path is located, sampling detection and determination are performed to determine whether to add a safety point; When an obstacle is detected, its risk sub-region is located, the risk weight of that region is updated, it is determined whether local replanning is triggered, and a new path segment is generated. Based on the new path segment, the final optimized path is obtained through curvature constraints, risk weight coupling, and smoothing.

2. The dynamic risk assessment clustering UAV path planning method as described in claim 1, characterized in that, The formula for calculating the grid risk density score is as follows: ;in, For the current raster used to calculate risk density, For the grid within the neighborhood, A grid set within the risk perception range. Here is the obstacle indication function; the formula for calculating the average risk density of the risk sub-region is: ;in, For average risk density, This represents the total number of grid cells within the risk sub-region. For the risk sub-region; the formula for the average path curvature is: ;in, The average path curvature This represents the number of path segments within the risk sub-region. Let be the curvature value of the i-th path segment; the total repulsive force is the vector sum of all individual repulsive forces, and the formula for calculating the individual repulsive force is: ;in, The grid where the drone is currently located. For a single obstacle grid, A set of grids representing obstacles. The repulsive force intensity coefficient, This is the repulsive force attenuation coefficient.

3. The dynamic risk assessment clustering UAV path planning method as described in claim 1, characterized in that, The resulting risk weights are expressed as follows: ;in, For average risk density, The average path curvature The overall repulsion strength of the region is the average value of the total repulsion force of all grids within the risk sub-region.

4. The dynamic risk assessment clustering UAV path planning method as described in claim 3, characterized in that, The elliptical pruning parameters are adaptively adjusted based on the overall repulsion strength of the region, as follows: ;in, The semi-minor axis of the ellipse after adaptive adjustment. To initialize the semi-minor axis of the ellipse parameters, The repulsive force intensity coefficient, The overall repulsion strength of the region.

5. The dynamic risk assessment clustering UAV path planning method as described in claim 1, characterized in that, The total cost function of the node is expressed as: ； in, For the actual cost; For heuristic cost estimation; The penalty cost is applied to the elliptical region; The clustering weight penalty term is represented as: ;in, These are the weighting coefficients. Risk weights.

6. The dynamic risk assessment clustering UAV path planning method as described in claim 1, characterized in that, The curvature constraint and risk weight coupling are expressed as follows: ;in, For the maximum permissible curvature, Minimum turning radius, The risk weight of the new path segment; If the risk weight of the new path segment is satisfied If there are two risk sub-regions, then two offset control points are added at the boundary between the two risk sub-regions, as shown below: ;in, , These are the endpoints of the new path segment in the two risk sub-regions, respectively. The offset distance of the control point. for The normal vector of the line connecting them.

7. A dynamic risk assessment clustering UAV path planning system, characterized in that, include: The clustering module is configured to: construct a two-dimensional map based on the raster method, define the raster state, initialize the ellipse parameters, calculate the raster risk density score, generate risk sub-regions through density clustering based on this score, and then calculate the average risk density, average path curvature, and total repulsion of the risk sub-regions to obtain the risk weight. The initial path generation module is configured to: generate candidate nodes based on eight-neighbor ray expansion, determine the priority of the search direction according to the angle difference between the current node and the target point, adaptively adjust the ellipse pruning parameters, calculate the total cost function of the nodes and expand the node with the smallest total cost function, and backtrack to generate the initial path after the target point has been expanded. The compression and detection module is configured to: compress the initial path based on the unobstructed straight path to obtain a simplified path, preset the spacing threshold, and perform secondary compression through redundant filtering to obtain the compressed path; Based on the risk weight of the risk sub-region where the compression path is located, sampling detection and determination are performed to determine whether to add a safety point; When an obstacle is detected, its risk sub-region is located, the risk weight of that region is updated, it is determined whether local replanning is triggered, and a new path segment is generated. The path optimization module is configured to obtain the final optimized path based on the new path segment through curvature constraints, risk weight coupling, and smoothing.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images