Unmanned aerial vehicle cluster cooperative operation path planning method based on distributed computing power

Abstract

The application discloses a UAV cluster cooperative operation path planning method based on distributed computing power, comprising the following steps: constructing a time-space safety corridor composed of convex polyhedrons for each UAV; generating an initial connected trajectory in the corridor and performing invariant reparameterization based on curvature and torsion, and optimizing and solving in the invariant space; representing the result as a curvature continuous spline curve, constructing an objective function containing bending energy and path cost with control points and node time as variables, imposing multiple constraints and establishing complementary curvature constraints; minimizing the objective function under the constraints to obtain a smooth flight path; locally re-optimizing when an obstacle or near collision is detected; edge nodes are responsible for constraint synchronization, and each UAV independently optimizes and rolls up the flight path. The application combines distributed computing power cooperation, geometric invariant reparameterization and complementary curvature constraints, and improves the stability and energy utilization efficiency of path planning.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a path planning method for collaborative operation of unmanned aerial vehicle (UAV) swarms based on distributed computing power. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarm collaborative operations have become an important research direction in the fields of intelligent equipment and air-ground collaborative control in recent years. With the widespread application of UAVs in tasks such as logistics transportation, disaster inspection, emergency rescue, and regional mapping, swarm flight scheduling has become a key technical means to improve mission efficiency and operational coverage.

[0003] In existing technologies, UAV path planning mostly adopts centralized control or single-unit autonomous planning methods. Centralized path planning is usually calculated by the ground control center to obtain the cluster trajectory. Although it can obtain the globally optimal solution, it is prone to delay and failure problems when communication links are limited or the mission environment changes dynamically. Single-unit autonomous planning relies on local sensors and computing power, which can achieve local obstacle avoidance to a certain extent, but lacks global coordination and is prone to path conflicts or uneven resource allocation.

[0004] For multi-objective mission planning of UAV swarms, researchers have proposed various methods based on graph search, dynamic programming, reinforcement learning, and convex optimization. For example, path search based on A* or D* algorithms can quickly find the shortest path in discrete space, but it is difficult to balance dynamic constraints and energy costs. Continuous trajectory optimization methods based on Bézier curves, B-splines, or polynomial curves can generate smooth trajectories, but when the constraints are complex or there is multi-UAV interaction, the computational load increases significantly and the real-time performance is insufficient.

[0005] To address computational latency issues, some solutions have introduced distributed computing architectures, allocating trajectory optimization tasks to edge nodes or onboard processing units. However, most existing distributed planning methods still represent trajectories using traditional time or arc-length parameterization, lacking modeling of geometric invariants (such as curvature and torsion). This results in a highly correlated optimization space and limited convergence speed. Furthermore, existing multi-UAV collaborative constraints often rely solely on position or distance thresholds, failing to establish a curvature-level group coordination mechanism, which can easily lead to local clustering or trajectory intersections in complex scenarios.

[0006] Therefore, how to provide a method for collaborative operation path planning of drone swarms based on distributed computing power is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a path planning method for collaborative operation of UAV swarms based on distributed computing power. This invention establishes a two-layer solution mechanism under a distributed computing architecture, dividing trajectory planning into collaborative execution by edge nodes and the airborne terminal: edge nodes are responsible for the synchronous solution of constraints and data exchange among multiple UAVs, while the airborne terminal independently completes spline optimization and constraint correction, forming a parallel computing structure to improve real-time performance and scalability. A trajectory reparameterization method based on geometric invariants is adopted to convert the trajectory from spatial coordinates into an invariant form of curvature and torsion, performing optimization within the invariant space to reduce parameter correlation and enhance curvature continuity and dynamic executability. Furthermore, complementary curvature constraints between adjacent UAVs are introduced, making the curvature changes of local trajectory segments inversely correlated, achieving streamline collaboration among the swarm and reducing trajectory aggregation and intersection. The constructed objective function comprehensively considers bending energy, path length, velocity and acceleration smoothness, and energy consumption, and combines normalization and dynamic weight adjustment mechanisms to achieve an adaptive balance between energy utilization and trajectory smoothness.

[0008] The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to embodiments of the present invention includes the following steps:

[0009] Based on the obstacles, no-fly zones, and terrain boundaries of the operating area, a spatiotemporal safety corridor composed of multiple convex polyhedra is constructed for each drone;

[0010] Within the spatiotemporal safety corridor, an initial connectivity trajectory is generated based on the take-off and landing points and the mission points;

[0011] The initial connected trajectory is reparameterized based on geometric invariants, and the trajectory is represented as a function of curvature and torsion. Optimization is then performed in the invariant space.

[0012] The reparameterized trajectory is represented as a curvature-continuous spline curve. The spline satisfies second-order continuity at position, velocity, and acceleration. The spline control points and node times are used as optimization variables.

[0013] Construct an objective function that includes a bending energy term and a path cost term, and impose constraints on the spline curve, wherein complementary curvature constraints are established in the spline parameter matrices corresponding to adjacent UAVs;

[0014] Minimize the objective function under the constraints to obtain a curvature continuous spline track that satisfies the complementary curvature constraints;

[0015] When a moving obstacle or a near collision between adjacent UAVs is detected, the node time of the corresponding corridor sub-segment is locally re-optimized to obtain an optimized trajectory.

[0016] During the optimization and local re-optimization process, the edge computing nodes are responsible for the synchronous solution of constraints and data exchange among multiple UAVs, while each UAV independently performs local spline optimization and constraint correction.

[0017] The optimized flight path is issued and executed, and is updated in a rolling manner according to a sliding time window when the environment changes or energy thresholds are triggered.

[0018] Furthermore, the construction of the spatiotemporal security corridor includes:

[0019] Obtain the digital elevation model of the work area and obstacle boundary information, and perform spatial discretization processing on obstacles and no-fly zones;

[0020] Discretize the planning period on the time axis with a preset step size to form a discrete time layer;

[0021] Within each discrete time layer, a spatial convex polyhedron is constructed using the convex hull generation method, with the predicted range of the UAV's current position as the center.

[0022] Based on the predicted points set of UAV flight paths, the coordinates of the polyhedron vertices are determined using the spatial density clustering results within adjacent time layers.

[0023] By connecting convex polyhedra in adjacent time layers in chronological order, a continuous spatiotemporal corridor sequence is formed along the time direction;

[0024] Perform a geometric intersection test on the spatiotemporal corridor sequence to remove polyhedra that overlap with obstacles or no-fly zones;

[0025] Adjusting polyhedron boundaries based on path connectivity detection.

[0026] Furthermore, the generation of the initial connected trajectory includes:

[0027] Within the feasible area of ​​the spatiotemporal safety corridor, determine the spatial coordinates of the takeoff point, mission point, and landing point;

[0028] A path search method based on feasible corridor topology is adopted to establish a connectivity graph between continuous convex polyhedra;

[0029] The path search method employs parallel heuristic search, maintaining a corridor connectivity graph at edge nodes and synchronizing the search results of each UAV.

[0030] Using the takeoff point as the starting node and the mission point and landing point as the target nodes, a shortest path search is performed in the connected graph to obtain an initial path sequence composed of the center points of several convex polyhedra.

[0031] The initial path sequence is spliced ​​and smoothed to ensure that the turning angle, heading rate of change, and altitude rate of change of adjacent line segments all meet the dynamic constraints of the UAV, thus forming an initial connected trajectory.

[0032] Furthermore, the reparameterization process based on geometric invariants includes:

[0033] Obtain the discrete coordinate point sequence of the initial connected trajectory, and calculate the curvature and torsion of adjacent trajectory segments;

[0034] Using curvature and torsion as invariant parameters, a parameterized function describing the geometry of the trajectory is constructed;

[0035] By using integral mapping, a correspondence is established between invariant parameters and spatial coordinates, and the objective function is solved in the invariant space with curvature and torsion as optimization variables.

[0036] The integral mapping adopts the discrete integral approximation method, which realizes the mapping of curvature and torsion to spatial coordinates by piecewise cumulative summation;

[0037] After the solution is completed, the optimization results are mapped back to the spatial coordinate system according to the integral relationship to obtain the corresponding curvature continuous spline trajectory.

[0038] Furthermore, the objective function comprising the bending energy term and the path cost term is composed of a weighted sum of the following terms:

[0039] The bending energy term is used to characterize the degree of curvature change of a spline curve.

[0040] The track length term is used to characterize the path arc length;

[0041] Velocity smoothing term and acceleration smoothing term are used to constrain the rate of change of velocity and acceleration;

[0042] Energy consumption item, used to characterize thrust output or electrical energy consumption;

[0043] After normalization, each cost is combined according to a preset weight coefficient to form a dimensionless comprehensive optimization objective.

[0044] The weighting coefficients for each cost are dynamically adjusted based on the drone's computing power, battery power, and communication status.

[0045] Furthermore, the constraints imposed on the spline curve include:

[0046] The corridor constraint restricts the position of the spline at each discrete time point to be within the spatiotemporal safety corridor. The corridor constraint is achieved by performing half-space detection on the spline sampling points. Each polyhedron is described by a set of planar equations.

[0047] Velocity constraint: limits the velocity of the spline at any node to a set upper limit.

[0048] Acceleration constraints and jerk constraints limit the acceleration and jerk of the UAV to a preset range;

[0049] Safety distance constraints limit the minimum distance between drones to a safety threshold.

[0050] The time window constraint limits the arrival time of the drone at the mission point to a specified time interval.

[0051] Furthermore, the constraints applied to the spline curve also include complementary curvature constraints, the establishment process of which includes:

[0052] Obtain the spline parameter matrix of adjacent UAVs within the same time period, and discretize the local curvature distribution of their respective splines;

[0053] A complementary relationship constraint is established between the discretized curvature distributions, so that in the flight path segment where the local curvature of any UAV increases, the curvature of the corresponding adjacent UAV decreases by a preset inverse coefficient;

[0054] The inverse coefficient is dynamically set based on the distance between UAVs and communication delay, and the complementary curvature data is updated synchronously through edge nodes;

[0055] In the spline optimization process, the complementary relationship is incorporated into the objective function in the form of linear or nonlinear constraints.

[0056] Furthermore, the process of minimizing the objective function under constraints and performing local re-optimization includes:

[0057] An iterative optimization method is used to jointly solve for the spline control points and node times to obtain a curvature continuous spline track that satisfies the constraints.

[0058] Local re-optimization is performed within a preset time window, and the optimization scope is limited to the corridor sub-segment where conflicts are detected and its adjacent time layer.

[0059] When a moving obstacle or a near-collision between adjacent UAVs is detected, the corresponding corridor segment is determined, and the node time of the segment is used as a local optimization variable. The objective function is then re-solved while maintaining the continuity of adjacent segments to obtain the optimized trajectory.

[0060] Furthermore, the steps for issuing and executing the optimized flight path include:

[0061] The optimized trajectory parameters for each UAV are sent to the airborne control unit in segments, and trajectory tracking control is executed according to the time sequence.

[0062] When environmental parameter changes or energy thresholds are detected, the status information is re-acquired, and the aforementioned optimization steps are called within the sliding time window for rolling updates to generate new feasible tracks and replace the original tracks.

[0063] Furthermore, the scrolling update within the sliding time window includes:

[0064] Real-time monitoring of the drone's pose, remaining energy, communication link quality, and external environmental parameters;

[0065] When any parameter is detected to exceed a preset threshold, a rolling update process is triggered;

[0066] During the rolling update process, the spline optimization and constraint correction are re-executed on the affected corridor segments using the most recent optimization result as the initial condition;

[0067] During the rolling update process, the edge computing node is responsible for the aggregation of the states of multiple drones and the synchronization of unified constraints. Each drone performs local optimization and returns the update results.

[0068] The updated flight path is distributed to the corresponding drone via edge computing nodes.

[0069] The beneficial effects of this invention are:

[0070] This invention divides the trajectory solving process into two levels: edge nodes and airborne terminals. The edge computing nodes are responsible for the synchronous solution of constraints and data exchange among multiple UAVs, while each UAV terminal independently performs local spline optimization and constraint correction, forming a parallel computing structure. This effectively reduces the load on the central node and improves the real-time performance and scalability of path planning.

[0071] By converting the trajectory from a spatial coordinate representation to an invariant representation of curvature and torsion, and performing optimization in the invariant space, parameter correlation can be significantly reduced, the objective function can have better convergence characteristics, and the optimization results are more stable in terms of curvature continuity and dynamic executability.

[0072] The curvature changes of adjacent UAVs in local flight segments are inversely correlated. When one UAV performs a large turn, the neighboring UAVs automatically adjust their own flight path curvature to achieve streamline coordination among the group, reduce the risk of trajectory aggregation and intersection, and thus improve the safety and space utilization of swarm flight.

[0073] The objective function comprehensively considers factors such as bending energy, path length, speed and acceleration smoothness, and energy consumption. Through normalization and dynamic weight adjustment mechanisms, it achieves a trade-off between energy and smoothness under different flight states, thereby enhancing the adaptability of the planning results. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1 This is a flowchart of the UAV swarm collaborative operation path planning method based on distributed computing power proposed in this invention;

[0076] Figure 2 This is a schematic diagram illustrating the principle of constructing a spatiotemporal safety corridor using the UAV swarm collaborative operation path planning method based on distributed computing power proposed in this invention.

[0077] Figure 3 This diagram illustrates the local re-optimization and sliding time window rolling update of the UAV swarm collaborative operation path planning method based on distributed computing power proposed in this invention. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0079] refer to Figure 1 - Figure 3 The method for collaborative operation path planning of UAV swarms based on distributed computing power includes the following steps:

[0080] Based on the obstacles, no-fly zones, and terrain boundaries of the operating area, a spatiotemporal safety corridor composed of multiple convex polyhedra is constructed for each drone;

[0081] Within the spatiotemporal safety corridor, an initial connectivity trajectory is generated based on the take-off and landing points and the mission points;

[0082] The initial connected trajectory is reparameterized based on geometric invariants, and the trajectory is represented as a function of curvature and torsion. Optimization is then performed in the invariant space.

[0083] The reparameterized trajectory is represented as a curvature-continuous spline curve. The spline satisfies second-order continuity at position, velocity, and acceleration. The spline control points and node times are used as optimization variables.

[0084] Construct an objective function that includes a bending energy term and a path cost term, and impose constraints on the spline curve, wherein complementary curvature constraints are established in the spline parameter matrices corresponding to adjacent UAVs;

[0085] Minimize the objective function under the constraints to obtain a curvature continuous spline track that satisfies the complementary curvature constraints;

[0086] When a moving obstacle or a near collision between adjacent UAVs is detected, the node time of the corresponding corridor sub-segment is locally re-optimized to obtain an optimized trajectory.

[0087] During the optimization and local re-optimization process, the edge computing nodes are responsible for the synchronous solution of constraints and data exchange among multiple UAVs, while each UAV independently performs local spline optimization and constraint correction.

[0088] The optimized flight path is issued and executed, and is updated in a rolling manner according to a sliding time window when the environment changes or energy thresholds are triggered.

[0089] In this embodiment, the construction of the spatiotemporal safety corridor includes:

[0090] The digital elevation model and obstacle boundary information of the work area are obtained. The resolution of the digital elevation model is set to 1 meter and a regular grid structure is adopted. The obstacle boundary information is generated based on the lidar point cloud and geographic information database. After three-dimensional meshing, it is unified to the global coordinate system. Spatial discretization processing is performed on the obstacles and no-fly zones, and the continuous area is divided into voxel units with a voxel side length of 2 meters.

[0091] Discretize the planning period on the time axis with a preset step size to form a discrete time layer. Each time layer represents a static spatial snapshot within a fixed time interval.

[0092] Within each discrete time layer, with the current location prediction range of the UAV as the center, the spatial location of the UAV in the future is predicted based on its current spatial location, velocity, acceleration and attitude angle. The predicted location is calculated by adding the product of the velocity and the time step to the current location, plus half of the acceleration and the square of the time step. Based on these predicted locations, a set of predicted points containing multiple sampling points is generated.

[0093] Based on the obtained set of predicted points, a spatial convex polyhedron is constructed by generating the convex hull. Using the set of predicted points as input, the minimum convex hull that can completely enclose all predicted points is calculated, thereby determining the vertex coordinates and boundary plane parameters of the polyhedron. Each generated convex polyhedron represents the spatial range in which the UAV can safely fly within this time layer.

[0094] Based on the UAV trajectory prediction point set, the spatial density clustering results in adjacent time layers are used to determine the coordinates of the polyhedron vertex. The prediction points are projected onto a three-dimensional grid, the point density in each grid cell is counted, and points with a density not less than 80% of the maximum density are selected as candidate points for the polyhedron vertex. Candidate points in adjacent time layers are matched according to the spatial nearest neighbor principle to determine the corresponding vertex of the convex polyhedron in the continuous time layer.

[0095] The convex polyhedra in adjacent time layers are connected in chronological order to form a continuous spatiotemporal corridor sequence along the time direction. A geometric intersection test is performed on the spatiotemporal corridor sequence to remove polyhedra that overlap with obstacles or no-fly zones. The boundary equations of each convex polyhedron are substituted into the grid point coordinates of the obstacle. If more than 5% of the grid points are found to be inside the polyhedron, the polyhedron is considered to overlap with the obstacle.

[0096] Adjusting polyhedron boundaries based on path connectivity detection.

[0097] In this embodiment, the generation of the initial connected trajectory includes:

[0098] Within the feasible area of ​​the spatiotemporal safety corridor, the spatial coordinates of the takeoff point, mission point, and landing point are determined. The takeoff point and landing point are determined based on the spatial deployment of the operation mission, and the mission point is set according to the operation objective or the aerial survey coverage area.

[0099] A path search method based on feasible corridor topology is adopted to establish a connectivity graph between continuous convex polyhedra. Each convex polyhedron is regarded as a node in the graph structure, and polyhedra with spatial intersection between adjacent time layers are regarded as connectable nodes.

[0100] The path search method adopts a parallel heuristic search, maintains a corridor connectivity graph on the edge nodes and synchronizes the search results of each UAV. The path search starts from the polyhedron node where the take-off point is located, takes the polyhedrons where the mission point and landing point are located as target nodes, and searches for feasible paths layer by layer along the node connectivity edges. During the search process, the optimal channel is selected according to the path cost function, which simultaneously considers spatial distance, turning change and connectivity index.

[0101] The initial path sequence is spliced ​​and smoothed. The splicing process connects the center points of adjacent polyhedra in time sequence to form initial track segments, so that the turning angle, heading rate of change and altitude rate of change of adjacent segments all meet the dynamic constraints of the UAV, thus forming an initial connected trajectory.

[0102] In this embodiment, the reparameterization process based on geometric invariants includes:

[0103] Obtain the discrete coordinate point sequence of the initial connected trajectory, and calculate the curvature and deflection of adjacent trajectory segments. Curvature is used to characterize the degree of bending of the trajectory at that position, and deflection is used to characterize the degree of spatial twisting of the trajectory.

[0104] By using curvature and torsion as invariant parameters, the initial connected trajectory is transformed from a coordinate sequence into a function with curvature and torsion as independent variables;

[0105] By establishing a correspondence between invariant parameters and spatial coordinates through integral mapping, and performing objective function solution with curvature and torsion as optimization variables in invariant space, the optimization process takes curvature continuity and torsion smoothness as constraints, aims to reduce overall bending energy and spatial deviation, and iteratively adjusts the rate of change of curvature and torsion to meet the requirements of smoothness and stability in continuous time period.

[0106] The integral mapping adopts the discrete integral approximation method, which realizes the mapping of curvature and torsion to spatial coordinates by piecewise cumulative summation;

[0107] After the solution is completed, the optimization results are mapped back to the spatial coordinate system according to the integral relationship to obtain the corresponding curvature continuous spline trajectory.

[0108] After completing the curvature and torsion-based reparameterization, the curvature continuous spline optimization and constraint application process is performed. The reparameterized trajectory is expressed as a spline curve. The spline curve is defined by several control points and node times, with the spatial coordinates of each control point and the time node constituting the optimization variables.

[0109] The objective function consists of a weighted combination of multiple cost terms, including a bending energy term to characterize the geometric smoothness of the trajectory, a trajectory cost term to measure the path length, a velocity and acceleration smoothness term to reflect motion smoothness, and an energy cost term to control energy usage efficiency.

[0110] ;

[0111] Indicates the trajectory in arc length parameter curvature at that point Represents torsion The derivative of the velocity vector with respect to the arc length is used to describe the rate of change of velocity. This represents the derivative of the acceleration vector with respect to the arc length, used to describe the smoothness of acceleration. This represents the total arc length of the trajectory. , , , , The weighting coefficients are normalized and adjusted according to the different smoothness and energy requirements of the task. , , These are the normalized reference length, velocity, and acceleration constant, respectively, used to maintain the dimensionless nature of all terms.

[0112] In this embodiment, the constraints applied to the spline curve include:

[0113] The corridor constraint restricts the position of the spline at each discrete time point to be within the spatiotemporal safety corridor. The corridor constraint is achieved by performing half-space detection on the spline sampling points. Each polyhedron is described by a set of planar equations.

[0114] Velocity constraint: limits the velocity of the spline at any node to a set upper limit.

[0115] Acceleration constraints and jerk constraints limit the acceleration and jerk of the UAV to a preset range;

[0116] Safety distance constraints limit the minimum distance between drones to a safety threshold.

[0117] The time window constraint limits the arrival time of the drone at the mission point to a specified time interval.

[0118] In this embodiment, the constraints applied to the spline curve also include complementary curvature constraints, and the establishment process includes:

[0119] Obtain the spline parameter matrix of adjacent UAVs within the same time period, and discretize the local curvature distribution of their respective splines;

[0120] A complementary relationship constraint is established between the discretized curvature distributions, so that in the flight path segment where the local curvature of any UAV increases, the curvature of the corresponding adjacent UAV decreases by a preset inverse coefficient;

[0121] The inverse coefficient is dynamically set based on the distance between UAVs and communication delay, and the complementary curvature data is updated synchronously through edge nodes;

[0122] In the spline optimization process, the complementary relationship is incorporated into the objective function in the form of linear or nonlinear constraints.

[0123] After applying all constraints, the objective function is minimized. During the solution process, the spline control points and node times are iteratively adjusted to converge the objective function value to the optimum while satisfying all constraints. After optimization, a spline trajectory with continuous curvature, smooth torsion, and meeting the requirements of multi-machine collaboration is obtained.

[0124] The consistency of the optimization results was checked. By comparing the curves of curvature and torsion changes before and after optimization, it was confirmed that their continuity and smoothness met the preset standards.

[0125] In this embodiment, the process of minimizing the objective function under constraints and performing local re-optimization includes:

[0126] An iterative optimization method is used to jointly solve for the spline control points and node times to obtain a curvature continuous spline track that satisfies the constraints.

[0127] Local re-optimization is performed within a preset time window, and the optimization scope is limited to the corridor sub-segment where conflicts are detected and its adjacent time layer.

[0128] When a moving obstacle or a near-collision between adjacent UAVs is detected, the corresponding corridor segment is determined, and the node time of the segment is used as a local optimization variable. The objective function is then re-solved while maintaining the continuity of adjacent segments to obtain the optimized trajectory.

[0129] In this embodiment, the step of issuing and executing the optimized trajectory includes:

[0130] The optimized trajectory parameters for each UAV are sent to the airborne control unit in segments, and trajectory tracking control is executed according to the time sequence.

[0131] When environmental parameter changes or energy thresholds are detected, the status information is re-acquired, and the aforementioned optimization steps are called within the sliding time window for rolling updates to generate new feasible tracks and replace the original tracks.

[0132] In this embodiment, the scrolling update within the sliding time window includes:

[0133] During flight, the drone continuously collects information on its own position, speed, remaining energy, communication link status, and external obstacles.

[0134] When any parameter is detected to exceed the preset threshold, a rolling update process is triggered, and the scope of local re-optimization is limited to the corridor sub-segment where the conflict occurs and its adjacent time layer.

[0135] During the local re-optimization process, the spline parameters in the unaffected areas remain unchanged, and only the spline control points and node times in the conflict areas are used as optimization variables. The optimization objective is the same as that in the global planning, which is to minimize the objective function value based on the smoothing principle of curvature and torsion. However, the boundary conditions of the optimization area are fixed as the endpoints of adjacent trajectory segments.

[0136] After receiving the local states of each drone, the edge computing node is responsible for synchronizing constraint updates and data distribution. If multiple drones are detected to be in conflict in the same area, the complementary curvature constraint relationship of each drone is coordinated through a unified constraint synchronization mechanism.

[0137] The updated flight path is distributed to the corresponding drone via edge computing nodes.

[0138] During long-duration flights, a sliding time window mechanism is employed for rolling updates to address continuous environmental changes and variations in energy consumption. The length of the time window is dynamically set based on the mission cycle and communication frequency. At the beginning of each time window, the system re-acquires flight status, obstacle information, and energy levels, and uses the most recent optimization result as initial conditions to re-execute the spline optimization and constraint correction process.

[0139] When communication delays or data loss are detected, the edge computing nodes determine the validity of each drone's status based on the latest synchronization timestamp. If data lag occurs, the previous optimization result is retained, and the update is postponed to the next window period to ensure trajectory continuity and system stability.

[0140] Example 1:

[0141] To verify the feasibility of this invention, the UAV swarm collaborative operation path planning method based on distributed computing power was applied to a power line inspection task in a mountainous area. In this embodiment, the operation area includes complex terrain, elevation differences, and multiple obstacles, and a swarm of several UAVs equipped with edge computing units is used. Each UAV can independently perform spline optimization and constraint correction, while the edge computing nodes are responsible for global constraint synchronization and rolling update management.

[0142] Based on the regional digital elevation model and obstacle boundary information, a spatial mesh for the operational area is generated. Edge computing nodes perform discretization on the time axis, forming a multi-layered spatiotemporal safety corridor. A feasible flight path for each UAV is established using a convex hull generation method, and geometric intersection and connectivity checks are performed.

[0143] An initial connected trajectory is constructed using the take-off and landing points of each UAV and the inspection task points as input. After smoothing, the trajectory is input to the UAV for invariant reparameterization and spline optimization based on curvature and torsion. During this process, edge computing nodes synchronize the complementary curvature constraints of each UAV and dynamically adjust the constraint coefficients according to distance and communication latency.

[0144] When a moving obstacle or a potential conflict between adjacent drones is detected, local re-optimization is triggered. The affected drones only need to readjust the spline control points and nodes in the corresponding corridor sub-segment, while the edge nodes are responsible for the synchronous solution and trajectory coordination of the conflict segment.

[0145] Throughout the inspection process, edge computing nodes periodically perform rolling updates using a sliding time window approach, correcting flight paths in real time. After each update cycle, multi-aircraft status information is re-aggregated and constraint corrections are performed, thereby achieving continuous adaptation of flight paths and maintaining group stability.

[0146] To verify the performance of this invention, a comparative experiment was conducted between the method of this invention and a traditional centralized path planning method under the same operating conditions. The experiment used a swarm of six multi-rotor UAVs to perform inspection and obstacle avoidance tasks within a power line inspection area of ​​approximately 3 kilometers. The experimental environment included multiple obstacles, terrain undulations, and communication delay interference.

[0147] Table 1. Parameter settings for the control experiment

[0148] Number of drones 6 6 Length of working area Approximately 3km Approximately 3km Number of edge nodes 1 1 central control server Communication delay Approximately 120ms Approximately 120ms Optimize refresh cycle 2s (Dynamic) 10s (fixed) Trajectory Representation Curvature-torsion spline 3D coordinate spline Optimization range Local window optimization Global single-step solution Complementary constraint mechanism Enable Not enabled

[0149] Under the same hardware conditions, the experimental results of the two methods were statistically analyzed in terms of computation time, trajectory smoothness, communication latency adaptability, energy utilization, and cluster safety distance maintenance rate.

[0150] Table 2. Performance comparison results between the method of the present invention and the control method.

[0151] Average path planning time s 4.86 1.72 ↓64.6% Curvature continuity error (mean square deviation) 1 / m² 0.031 0.009 ↓70.9% Control stability under communication delay % 82.3 97.5 ↑18.5% Cluster safety distance maintenance rate % 91.4 99.2 ↑8.5% Energy utilization rate % 84.7 94.1 ↑11.1% Central node CPU utilization % 92.8 28.3 ↓69.5%

[0152] As shown in Table 2, the method of this invention reduces the average path planning time by approximately 65% ​​compared to traditional methods, lowers the trajectory curvature continuity error by approximately 71%, improves flight control stability under communication delay conditions by nearly 20%, maintains a cluster safety distance rate close to 100%, and significantly reduces the load on the central node. This indicates that the present invention, through a combination of edge computing and geometric invariant optimization, achieves a comprehensive improvement in the real-time performance, smoothness, and multi-UAV collaborative performance of path planning.

[0153] To ensure the objectivity of the results, this experiment was repeated ten times, and the data in the table are the average values. All results are used only to illustrate the technical effects of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0154] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power, characterized in that, Includes the following steps: Based on the obstacles, no-fly zones, and terrain boundaries of the operating area, a spatiotemporal safety corridor composed of multiple convex polyhedra is constructed for each drone; Within the spatiotemporal safety corridor, an initial connectivity trajectory is generated based on the take-off and landing points and the mission points; The initial connected trajectory is reparameterized based on geometric invariants, and the trajectory is represented as a function of curvature and torsion. Optimization is then performed in the invariant space. The reparameterized trajectory is represented as a curvature-continuous spline curve. The spline satisfies second-order continuity at position, velocity, and acceleration. The spline control points and node times are used as optimization variables. Construct an objective function that includes a bending energy term and a path cost term, and impose constraints on the spline curve, wherein complementary curvature constraints are established in the spline parameter matrices corresponding to adjacent UAVs; Minimize the objective function under the constraints to obtain a curvature continuous spline track that satisfies the complementary curvature constraints; When a moving obstacle or a near collision between adjacent UAVs is detected, the node time of the corresponding corridor sub-segment is locally re-optimized to obtain an optimized trajectory. During the optimization and local re-optimization process, the edge computing nodes are responsible for the synchronous solution of constraints and data exchange among multiple UAVs, while each UAV independently performs local spline optimization and constraint correction. The optimized trajectory is issued and executed, and is updated in a rolling time window when environmental changes or energy thresholds are triggered. The constraints applied to the spline curve also include complementary curvature constraints, the establishment process of which includes: Obtain the spline parameter matrix of adjacent UAVs within the same time period, and discretize the local curvature distribution of their respective splines; A complementary relationship constraint is established between the discretized curvature distributions, so that in the flight path segment where the local curvature of any UAV increases, the curvature of the corresponding adjacent UAV decreases by a preset inverse coefficient; The inverse coefficient is dynamically set based on the distance between UAVs and communication delay, and the complementary curvature data is updated synchronously through edge nodes; In the spline optimization process, the complementary relationship is incorporated into the objective function in the form of linear or nonlinear constraints.

2. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The construction of the spatiotemporal security corridor includes: Obtain the digital elevation model of the work area and obstacle boundary information, and perform spatial discretization processing on obstacles and no-fly zones; Discretize the planning period on the time axis with a preset step size to form a discrete time layer; Within each discrete time layer, a spatial convex polyhedron is constructed using the convex hull generation method, with the predicted range of the UAV's current position as the center. Based on the predicted points set of UAV flight paths, the coordinates of the polyhedron vertices are determined using the spatial density clustering results within adjacent time layers. By connecting convex polyhedra in adjacent time layers in chronological order, a continuous spatiotemporal corridor sequence is formed along the time direction; Perform a geometric intersection test on the spatiotemporal corridor sequence to remove polyhedra that overlap with obstacles or no-fly zones; Adjusting polyhedron boundaries based on path connectivity detection.

3. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The generation of the initial connected trajectory includes: Within the feasible area of ​​the spatiotemporal safety corridor, determine the spatial coordinates of the takeoff point, mission point, and landing point; A path search method based on feasible corridor topology is adopted to establish a connectivity graph between continuous convex polyhedra; The path search method employs parallel heuristic search, maintaining a corridor connectivity graph at edge nodes and synchronizing the search results of each UAV. Using the takeoff point as the starting node and the mission point and landing point as the target nodes, a shortest path search is performed in the connected graph to obtain an initial path sequence composed of the center points of several convex polyhedra. The initial path sequence is spliced ​​and smoothed to ensure that the turning angle, heading rate of change, and altitude rate of change of adjacent line segments all meet the dynamic constraints of the UAV, thus forming an initial connected trajectory.

4. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The reparameterization process based on geometric invariants includes: Obtain the discrete coordinate point sequence of the initial connected trajectory, and calculate the curvature and torsion of adjacent trajectory segments; Using curvature and torsion as invariant parameters, a parameterized function describing the geometry of the trajectory is constructed; By using integral mapping, a correspondence is established between invariant parameters and spatial coordinates, and the objective function is solved in the invariant space with curvature and torsion as optimization variables. The integral mapping adopts the discrete integral approximation method, which realizes the mapping of curvature and torsion to spatial coordinates by piecewise cumulative summation; After the solution is completed, the optimization results are mapped back to the spatial coordinate system according to the integral relationship to obtain the corresponding curvature continuous spline trajectory.

5. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The objective function, which includes the bending energy term and the path cost term, is composed of a weighted sum of the following terms: The bending energy term is used to characterize the degree of curvature change of a spline curve. The track length term is used to characterize the path arc length; Velocity smoothing term and acceleration smoothing term are used to constrain the rate of change of velocity and acceleration; Energy consumption item, used to characterize thrust output or electrical energy consumption; After normalization, each cost is combined according to a preset weight coefficient to form a dimensionless comprehensive optimization objective. The weighting coefficients for each cost are dynamically adjusted based on the drone's computing power, battery power, and communication status.

6. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The constraints applied to the spline curve include: The corridor constraint restricts the position of the spline at each discrete time point to be within the spatiotemporal safety corridor. The corridor constraint is achieved by performing half-space detection on the spline sampling points. Each polyhedron is described by a set of planar equations. Velocity constraint: limits the velocity of the spline at any node to a set upper limit. Acceleration constraints and jerk constraints limit the acceleration and jerk of the UAV to a preset range; Safety distance constraints limit the minimum distance between drones to a safety threshold. The time window constraint limits the arrival time of the drone at the mission point to a specified time interval.

7. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The process of minimizing the objective function under constraints and performing local re-optimization includes: An iterative optimization method is used to jointly solve for the spline control points and node times to obtain a curvature continuous spline track that satisfies the constraints. Local re-optimization is performed within a preset time window, and the optimization scope is limited to the corridor sub-segment where conflicts are detected and its adjacent time layer. When a moving obstacle or a near-collision between adjacent UAVs is detected, the corresponding corridor segment is determined, and the node time of the segment is used as a local optimization variable. The objective function is then re-solved while maintaining the continuity of adjacent segments to obtain the optimized trajectory.

8. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The steps for issuing and executing the optimized flight path include: The optimized trajectory parameters for each UAV are sent to the airborne control unit in segments, and trajectory tracking control is executed according to the time sequence. When environmental parameter changes or energy thresholds are detected, the status information is re-acquired, and the aforementioned optimization steps are called within the sliding time window for rolling updates to generate new feasible tracks and replace the original tracks.

9. The method for collaborative operation path planning of unmanned aerial vehicle (UAV) swarms based on distributed computing power according to claim 1, characterized in that, The scrolling updates within the sliding time window include: Real-time monitoring of the drone's pose, remaining energy, communication link quality, and external environmental parameters; When any parameter is detected to exceed a preset threshold, a rolling update process is triggered; During the rolling update process, the spline optimization and constraint correction are re-executed on the affected corridor segments using the most recent optimization result as the initial condition; During the rolling update process, the edge computing node is responsible for the aggregation of the states of multiple drones and the synchronization of unified constraints. Each drone performs local optimization and returns the update results. The updated flight path is distributed to the corresponding drone via edge computing nodes.

Images