Unmanned aerial vehicle task execution method and system based on two-stage task allocation algorithm

Abstract

The invention relates to an unmanned aerial vehicle task execution method and system based on a two-stage task allocation algorithm. The method comprises the following steps: obtaining load information of a plurality of unmanned aerial vehicles; selecting a task allocation method according to the load information; when the task allocation method is selected as a first-level task allocation method, performing multi-thread parallel computing by adopting a nearest neighbor search method and a Hungary matching algorithm to match the formation position and the path planning scheme of each unmanned aerial vehicle; when the task allocation method is selected as a second-level task allocation method, matching the formation position and the path planning scheme of each unmanned aerial vehicle by adopting a graph neural network algorithm; generating an unmanned aerial vehicle task according to the matched formation position and the path planning scheme; according to the invention, a two-stage task allocation algorithm is adopted to allocate unmanned aerial vehicle formation tasks for conditions with small load difference and large load difference; the unmanned aerial vehicle formation tasks are allocated by adopting a proper algorithm, so that the unmanned aerial vehicle task allocation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) cluster control, and in particular to a UAV task execution method and system based on a two-level task allocation algorithm. Background Art

[0002] Large-scale drone swarm missions typically involve two tasks: swarm task allocation and swarm path planning. Currently, swarm formation transformations in drone swarm applications primarily take the form of complex offline simulations and real-time fixed formation transformations. Complex formation transformations and task allocation in real-time missions have long been challenging issues within the industry. While this task can be broken down into three phases: task allocation, environmental perception, and path planning, the high computational complexity and time required generally hinder its application in real-world scenarios.

[0003] The development trend of drone swarm technology is towards multi-task collaboration. Multiple drones in a swarm can achieve parallel sensing responses through the complementary combination of heterogeneous sensors. They can also use actuators to perform subtasks separately, achieving distributed execution of the overall task. If some drones fail, others can take over and complete the intended tasks, making the swarm system highly fault-tolerant and robust. Consequently, this technology has attracted widespread attention. Integrating task allocation and path planning into drone swarm systems could be a new application of current drone swarm technology.

[0004] EGO-Swarm is a widely used swarming solution. Its core concept is to achieve safe and efficient coordinated flight of drone swarms in dense obstacle environments through a lightweight trajectory optimization algorithm and a distributed perception framework. However, this technology requires high-precision onboard sensors (such as lidar), which is costly. In extremely dynamic environments, if obstacles are moving too fast (approaching the drone's speed limit), obstacle avoidance may fail.

[0005] For the heterogeneous UAV task allocation problem, planning and solving are usually used to abstract it into the knapsack problem or the traveling salesman problem. However, its solution is complex and has a limited scope. The TSP (knapsack problem) is an NP-hard problem, and the calculation time of its optimal solution increases exponentially with the increase in the number of cities. Even if algorithms such as dynamic programming are used, the time complexity is O(n 2 2 n ), when the number of drones is large, the amount of calculation is huge and it is difficult to get the answer within a reasonable time.

[0006] Although drone swarm technology has been studied for a long time, the following bottlenecks still exist:

[0007] 1. Large number: Usually consider dozens, hundreds, or even thousands of drones. The difficulty of coordination increases sharply with the number.

[0008] 2. Heterogeneity: Swarm members typically share the same basic platform, but may have different sensors, reconnaissance equipment, or payloads, resulting in varying capabilities for completing specific missions. During mission execution, swarm members must be appropriately assigned based on their capabilities.

[0009] 3. Multiple tasks: Clusters usually need to complete different tasks synchronously and in parallel. Different types of tasks have different requirements, and there may be constraints between tasks.

[0010] 4. Multiple constraints: In addition to the performance constraints of the UAV system, there are also tactical requirement constraints, battlefield environment constraints, communication constraints, platform space constraints, time constraints, task coupling constraints, track collision avoidance constraints, etc., as well as a large number of complex coupling and cross-linking relationships between constraints.

[0011] 5. Dynamic Change: The situations faced by swarms when performing missions are usually dynamic, and the targets, threats, missions, and the status of the drones themselves are constantly changing; in particular, in a confrontational environment, decisions and actions may be affected by the enemy's decisions and actions. Summary of the Invention

[0012] In order to solve the technical problems in the prior art that when there are a large number of heterogeneous drones, the drone task allocation calculation is huge and the calculation efficiency is low, and it is difficult to obtain answers within a reasonable time, the present invention provides a drone task execution method and system based on a two-level task allocation algorithm.

[0013] The technical solution of the present invention to solve the above technical problems is as follows:

[0014] A method for executing UAV tasks based on a two-level task allocation algorithm comprises the following steps:

[0015] Obtain payload information of multiple drones;

[0016] Selecting a corresponding task allocation method based on the load information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method;

[0017] When the task allocation method is selected as the first-level task allocation method, the nearest neighbor search method and the Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation position and path planning scheme of each UAV;

[0018] When the second-level task allocation method is selected as the task allocation method, a graph neural network algorithm is used to match the formation position and path planning scheme of each drone;

[0019] Generate drone missions based on the matched formation positions and path planning schemes of each drone.

[0020] The beneficial effects of the present invention are: since heterogeneous UAVs have different loads, a two-level task allocation algorithm is used to allocate UAV formation tasks to situations with small and large load differences respectively; a suitable algorithm is used to allocate UAV formation tasks to improve the efficiency of UAV task allocation and ensure that the task is answered within a reasonable time.

[0021] On the basis of the above technical solution, the present invention can also be improved as follows.

[0022] The load information includes load type and load weight;

[0023] Selecting a task allocation method according to the load information specifically includes the following steps:

[0024] Comparing the payload types of the plurality of drones, and if the payload types of two or more of the plurality of drones are inconsistent, generating payload difference information; if the payload types of the plurality of drones are consistent, comparing the payload weights of the plurality of drones, and generating payload consistency information when the payload weight difference between any two of the plurality of drones is less than a preset weight threshold; generating the payload difference information when the payload weight difference between two or more of the plurality of drones is greater than or equal to the preset weight threshold;

[0025] If the load consistency information is generated, the task allocation method is selected as the first-level task allocation method; if the load difference information is generated, the task allocation method is selected as the second-level task allocation method.

[0026] Furthermore, the nearest neighbor search method is used to calculate and match the formation position and path planning scheme of each drone, including the following steps:

[0027] Collecting preset position data and real-time position data of each of the drones;

[0028] The preset position data and the real-time position data of each drone are stored in a tree data structure;

[0029] Configuring multiple initial path planning schemes for each of the drones;

[0030] Using a nearest neighbor allocation algorithm to query the preset position data that is the nearest neighbor to the real-time position data, so as to match and obtain the formation position of each of the UAVs;

[0031] According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

[0032] Furthermore, the Hungarian matching algorithm is used to calculate the formation position and path planning scheme of each drone. The specific steps are as follows:

[0033] Collecting preset position data and real-time position data of each of the drones;

[0034] Configuring the preset position data and the real-time position data of each of the drones into two independent point sets of a bipartite graph;

[0035] Configuring multiple initial path planning schemes for each of the drones;

[0036] A Hungarian matching algorithm is used to select the preset position data that matches the maximum weight of the real-time position data from two independent point sets to match and obtain the formation position of each of the UAVs;

[0037] According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

[0038] Furthermore, collecting the preset position data of each of the drones includes the following steps:

[0039] Generate a plurality of dynamic hash 3D map blocks according to the preset motion range of each of the drones;

[0040] Storing map block indexes of the plurality of dynamically hashed 3D map blocks in an octree map;

[0041] Obtaining the map block index from the octree map to obtain a plurality of the dynamic hash 3D map blocks;

[0042] Constructing a three-dimensional space-time trajectory map based on the multiple dynamic hash 3D map blocks obtained by indexing;

[0043] The position coordinates of each drone are preset on the three-dimensional space-time trajectory map to obtain the preset position data of each drone. 6. The drone task execution method based on the two-level task allocation algorithm according to claim 5 is characterized in that collecting the real-time position data of each drone specifically includes the following steps:

[0044] Collecting images of each of the drones to obtain a multi-visual image dataset of each of the drones;

[0045] Build a target detection model based on computer vision recognition methods;

[0046] Training the target detection model using the multi-visual image dataset of each drone to obtain a drone target detection model;

[0047] Using the drone target detection model to identify each of the drones, and using a multi-target tracking algorithm to track each of the identified drones;

[0048] Calculating the relative speed of each of the identified and tracked drones using an optical flow method;

[0049] Calculate the distance change between each of the drones based on a monocular ranging method;

[0050] The real-time position data of each of the UAVs is generated according to the relative speed of each of the UAVs and the distance change between each of the UAVs.

[0051] Furthermore, when the plurality of drones perform the drone cluster mission according to the drone formation mission, a graph neural network algorithm is used to match the formation position and path planning scheme of each drone. The specific steps are as follows:

[0052] Constructing a preset node topology graph based on the preset position data, preset load types, and preset load weights of the plurality of drones; wherein the node information of the preset node topology graph includes the preset position data, the preset load type, and the preset load weight of each drone, and a line connecting two nodes of the preset node topology graph represents an edge;

[0053] Build a graph neural network model;

[0054] Training the graph neural network model using the preset position data of the plurality of drones, the preset load type, the preset load weight, the edges of the preset node topology graph, and the edge weights of the preset node topology graph to obtain a node graph generation model;

[0055] Collect real-time location data, real-time payload type, and real-time payload weight of multiple drones to build a real-time node topology map;

[0056] After inputting the real-time position data of the plurality of drones, the real-time load types, the real-time load weights, the edges of the real-time node topology graph, and the edge weights of the real-time node topology graph into the node graph generation model, the node graph generation model outputs a predicted node topology graph;

[0057] Matching the position data of each node in the predicted node topology map with the real-time position data of each drone, matching the load type of each node in the predicted node topology map with the real-time load type of each drone, and matching the load weight of each node in the predicted node topology map with the real-time load weight of each drone to match the formation position of each drone;

[0058] Configuring multiple initial path planning schemes for each of the UAVs according to the formation position;

[0059] According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

[0060] Furthermore, generating a drone mission based on the matched formation position and the path planning scheme specifically includes the following steps:

[0061] Setting a safe flight distance according to the body size and positioning error of each of the drones;

[0062] Generate a path sequence smooth route for each of the UAVs according to the formation position, the path planning scheme, and the safe flight distance;

[0063] Determining whether the smoothed paths of the path sequences of the respective UAVs have intersections at the same time step, and if so, performing local path optimization on the smoothed paths of the path sequences of the respective UAVs having intersections to obtain final path planning curves for the respective UAVs; if not, selecting the smoothed paths of the path sequences as the final path planning curves;

[0064] The UAV mission is generated according to the final path planning curve.

[0065] Furthermore, local path optimization is performed on the smoothed path of the path sequence of each UAV having an intersection, and the specific steps are as follows:

[0066] A path planning A* algorithm or a path planning D* algorithm based on an artificial potential field method is used to perform local path optimization on the path sequence smooth route of each of the UAVs having intersections.

[0067] In order to solve the above technical problems, the present invention also provides a UAV task execution system based on a two-level task allocation algorithm, the specific technical contents of which are as follows:

[0068] A UAV task execution system based on a two-level task allocation algorithm includes:

[0069] Information acquisition module: used to obtain payload information of multiple drones;

[0070] A task allocation module is configured to select a corresponding task allocation method based on the payload information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method; when the first-level task allocation method is selected, a nearest neighbor search method and a Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation position and path planning scheme of each UAV; when the second-level task allocation method is selected, a graph neural network algorithm is used to match the formation position and path planning scheme of each UAV;

[0071] Mission execution module: used to generate drone missions based on the matched formation positions and path planning schemes of each drone. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of a method for executing UAV tasks based on a two-level task allocation algorithm in an embodiment of the present invention;

[0073] Figure 2 Schematic diagram of an octree structure in an embodiment of the present invention;

[0074] Figure 3 A spline curve diagram for path planning in an embodiment of the present invention;

[0075] Figure 4 Schematic diagram of the structure of a UAV task execution system based on a two-level task allocation algorithm in an embodiment of the present invention;

[0076] Figure 5 The figure is a schematic diagram of the task structure of a UAV task execution system based on a two-level task allocation algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0078] like Figure 1 As shown, a method for executing UAV tasks based on a two-level task allocation algorithm includes the following steps:

[0079] S1. Obtaining payload information of multiple drones; the payload information includes payload type and payload weight.

[0080] S2. Select a corresponding task allocation method based on the payload information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method.

[0081] Selecting a task allocation method according to the load information specifically includes the following steps:

[0082] S201: Comparing the payload types of the plurality of drones; if the payload types of two or more of the plurality of drones are inconsistent, generating payload difference information; if the payload types of the plurality of drones are consistent, comparing the payload weights of the plurality of drones; if the payload weight difference between any two of the plurality of drones is less than a preset weight threshold, generating payload consistency information; if the payload weight difference between any two of the plurality of drones is greater than or equal to the preset weight threshold, generating payload difference information;

[0083] S202: If the load consistency information is generated, the task allocation method is selected as the first-level task allocation method; if the load difference information is generated, the task allocation method is selected as the second-level task allocation method.

[0084] Among them, the method for calculating the load weight difference between multiple drones is: calculating the load weight difference between any two drones among the load weights between the multiple drones to obtain multiple weight difference values; selecting the weight difference with the largest value among the multiple weight difference values ​​as the selected comparison difference value; comparing the comparison difference value with the preset weight threshold, when the comparison difference value is less than the preset weight threshold, generating load consistency information; when the comparison difference value is greater than or equal to the preset weight threshold, generating the load difference information.

[0085] S3. When the task allocation method is selected as the first-level task allocation method, a nearest neighbor search method and a Hungarian matching algorithm are used to perform multi-threaded parallel calculations to match the formation positions and path planning schemes of each UAV.

[0086] The Hungarian matching method and the nearest neighbor search method based on the tree data structure are used for multi-threaded concurrent calculation. The UAV formation task with no collision on the initial path is selected first, and then the UAV formation task with the shortest total distance is selected.

[0087] Adopt the nearest neighbor search method based on tree data structure to optimize storage and search speed, with a time complexity of O(N log N);

[0088] The Hungarian matching method is a bipartite graph maximum weight matching algorithm. Its time complexity is O(N 3 );

[0089] The hybrid task matching scheme MIN automatically matches two algorithms: the nearest neighbor search method based on a tree data structure and the Hungarian matching method. It prioritizes the initial path collision-free scheme, followed by the scheme with the shortest total distance. The shortest total distance scheme is the one that minimizes the sum of all UAV flight distances. An initial path collision-free scheme is one that contains collision-free trajectory points at the same time step in the initial flight trajectory of each UAV in the initial path planning scheme. The shortest total distance path planning scheme is one that minimizes the sum of all UAV flight path paths among all initial path planning schemes.

[0090] The nearest neighbor search method and the Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation positions and path planning schemes of each UAV, including the following steps:

[0091] S301, collecting preset position data and real-time position data of each of the drones;

[0092] Collecting the preset position data of each of the drones includes the following steps:

[0093] S3011, collecting the preset position data of each drone, including the following steps:

[0094] S3012, generating a plurality of dynamic hash 3D map blocks according to the preset motion range of each of the drones;

[0095] S3013, storing the map block indexes of the multiple dynamically hashed 3D map blocks in an octree map;

[0096] S3014, obtaining the map block index from the octree map to obtain a plurality of the dynamic hash 3D map blocks;

[0097] S3015, constructing a three-dimensional space-time trajectory map according to the multiple dynamic hash 3D map blocks obtained by indexing;

[0098] S3016: Preset the position coordinates of each of the UAVs on the three-dimensional space-time trajectory map to obtain the preset position data of each of the UAVs.

[0099] Collecting the real-time location data of each drone specifically includes the following steps:

[0100] S3017. Collect images of each of the drones to obtain a multi-visual image dataset of each of the drones;

[0101] S3018. Build a target detection model based on computer vision recognition methods;

[0102] S3019: Train the target detection model using the multi-visual image dataset of each drone to obtain a drone target detection model;

[0103] S3020: using the drone target detection model to identify each of the drones, and using a multi-target tracking algorithm to track each of the identified drones;

[0104] S3021. Calculate the relative speed of each of the identified and tracked drones using an optical flow method;

[0105] S3022, calculating the distance change between each of the drones based on a monocular ranging method;

[0106] The real-time position data of each of the drones is generated according to the relative speed of each of the drones and the distance change between each of the drones.

[0107] S302, storing the preset position data and the real-time position data of each drone in a tree data structure;

[0108] S303, configuring multiple initial path planning schemes for each of the drones; configuring the multiple initial path planning schemes according to the preset position data of each of the drones and the real-time position data;

[0109] S304, using a nearest neighbor allocation algorithm to query the preset position data that is the nearest neighbor to the real-time position data, so as to obtain the formation position of each of the UAVs;

[0110] S305. First, a collision-free path plan is selected from a plurality of initial path planning plans according to the formation position of each of the UAVs, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

[0111] The Hungarian matching algorithm is used to calculate the formation position and path planning scheme of each drone. The specific steps are as follows:

[0112] S310, collecting preset position data and real-time position data of each of the drones;

[0113] S311, configuring the preset position data and the real-time position data of each drone into two independent point sets of a bipartite graph;

[0114] S312, configuring multiple initial path planning schemes for each of the drones; configuring the multiple initial path planning schemes according to the preset position data of each of the drones and the real-time position data;

[0115] S313, using the Hungarian matching algorithm to select the preset position data that matches the maximum weight of the real-time position data from the two independent point sets, so as to obtain the formation position of each of the UAVs;

[0116] S314. First, a collision-free path plan is selected from a plurality of initial path planning plans according to the formation position of each of the UAVs, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

[0117] S4. When the task allocation method is selected as the second-level task allocation method, a graph neural network algorithm is used to match the formation position and path planning scheme of each drone.

[0118] When multiple drones perform the drone swarm mission according to the drone formation mission, a graph neural network algorithm is used to match the formation position and path planning scheme of each drone. The specific steps are as follows:

[0119] S401: Constructing a preset node topology graph based on the preset position data, preset load types, and preset load weights of the plurality of drones; wherein the node information of the preset node topology graph includes the preset position data, the preset load type, and the preset load weight of each drone, and a line connecting two nodes of the preset node topology graph represents an edge;

[0120] S402, constructing a graph neural network model;

[0121] S403: training the graph neural network model using the preset position data of the plurality of drones, the preset payload type, the preset payload weight, the edges of the preset node topology graph, and the edge weights of the preset node topology graph to obtain a node graph generation model;

[0122] S404, collecting real-time location data, real-time payload types, and real-time payload weights of multiple UAVs to build a real-time node topology map;

[0123] S405: After inputting the real-time position data of the plurality of drones, the real-time load types, the real-time load weights, the edges of the real-time node topology graph, and the edge weights of the real-time node topology graph into the node graph generation model, the node graph generation model outputs a predicted node topology graph;

[0124] The task assignment algorithm based on graph convolutional neural network is specifically G-TaskAssignment:

[0125] The UAVs and payloads are abstracted as weighted graph nodes, and the normalized inter-drone distances are abstracted as edge weights (normalized inter-drone distances are used to represent formations). The multi-payload UAV task allocation problem is abstracted as a graph generation problem. The current cluster state is mathematically represented as: G = (V, E, W), where V is the set of nodes and the attributes of each node; E is the set of edges, representing the connections between edges; and W is the edge weight, representing the strength (distance) of the relationship between nodes.

[0126] Mathematical form:

[0127] Node set V: V = {v1,v2,...,v n}, the total number of nodes is n. Node feature matrix: X∈R n×d , node v i By the d-dimensional feature vector x i Description x i ∈R d Here, the node features include: real-time location data, the real-time load type, and the real-time load weight; E represents the set of edges in the real-time node topology graph; and W represents the edge weight input of the real-time node topology graph.

[0128] Edge set E: E={(v i ,v j )|v i ,v j ∈V}, the total number of edges is m;

[0129] Edge weight W: W = {w ij |(v i ,v j )∈E}, and express it as the weighted adjacency matrix A∈R n×n , if (v i ,v j )∈E, then A ij =w ij , otherwise A ij =0.

[0130] Normalization method of edge weight: Min-Max normalization,

[0131] Model definition:

[0132] Input: G input =(V,E,W);

[0133] Output: G output =(V′,E,W′);

[0134] Node feature update: V→V′;

[0135] Edge weight update: W→W′.

[0136] S406: Matching the position data, load type, and load weight corresponding to each node in the predicted node topology graph with the real-time position data, the real-time load type, and the real-time load weight of each UAV to determine the formation position of each UAV;

[0137] Match the position data of each node in the predicted node topology map with the real-time position data of each UAV, match the load type of each node in the predicted node topology map with the real-time load type of each UAV, and match the load weight of each node in the predicted node topology map with the real-time load weight of each UAV to match the formation position of each UAV.

[0138] S407. Configure multiple initial path planning schemes for each of the UAVs according to the formation position; specifically, configure multiple initial path planning schemes for each UAV according to the formation position of each UAV; the formation position of the UAV is the position coordinates and real-time coordinates of the UAV in the target formation; the position coordinates in the target formation are the above-mentioned preset position data, and the real-time coordinates are the above-mentioned real-time position data.

[0139] S408. First, select a collision-free path plan from a plurality of initial path planning plans according to the formation position of each of the UAVs, and then select a path planning plan with the shortest total distance from the collision-free path plans to obtain the path planning plan.

[0140] S5. Generate a drone mission based on the matched formation positions and path planning schemes of each drone.

[0141] Generate a drone mission based on the matched formation positions and path planning schemes of each drone, specifically including the following steps:

[0142] S501, setting a safe flight distance according to the body size and positioning error of each UAV;

[0143] S502: Generate a path sequence smooth route for each of the UAVs based on the formation position, the path planning scheme, and the safe flight distance;

[0144] S503: Determine whether the smoothed paths of the path sequences of the respective UAVs have any intersections at the same time step. If so, locally optimize the paths of the smoothed paths of the path sequences of the respective UAVs having any intersections to obtain a final path planning curve for each UAV. If not, select the smoothed paths of the path sequences as the final path planning curve.

[0145] The path sequence smoothing route of each UAV having an intersection is locally optimized, and the specific steps are as follows:

[0146] A path planning A* algorithm or a path planning D* algorithm based on an artificial potential field method is used to perform local path optimization on the path sequence smooth route of each of the UAVs having intersections.

[0147] S504: Generate the UAV mission according to the final path planning curve.

[0148] When the plurality of drones perform the drone cluster mission according to the drone formation mission, a method of identifying the drones by computer vision is used to locate the drones.

[0149] In the process of multiple drones performing the drone cluster task according to the drone formation task, real-time cluster path planning for the multiple drones is completed through the construction, storage and query method of the three-dimensional space-time trajectory map.

[0150] The method for constructing a three-dimensional space-time trajectory map is to generate a dynamic hash 3D map according to a preset motion range of the UAV; wherein the dynamic hash 3D map is the three-dimensional space-time trajectory map.

[0151] like Figure 2 As shown, the map block index of the three-dimensional space-time trajectory map is stored using an octree spatial data structure.

[0152] Map storage: Generates a dynamic hash 3D map based on the motion range and uses an octree to optimize local map response.

[0153] Conflict Detection: Based on the planning granularity and current speed, the path control points are dynamically adjusted according to the Δt time. Detect whether there are conflicting control points at the same time step. If so, perform local path planning for them.

[0154] Safety corridor: In the process of space-time path conflict detection, the size and positioning error of the drone are taken into consideration and an air safety corridor is set to supplement the safety hazards caused by the basic conflict detection assumption that the drone is a point mass model.

[0155] Numerical optimization:

[0156] Map query optimization: To adapt to large-scale local path solving and path query, an octree map is designed. The octree example is as follows:

[0157] By optimizing the storage structure, we avoid long search times and the problem of re-creating forgotten maps. Map block indexes are stored in an octree, and the octree is searched to obtain local map responses. This balances the paradox between search time complexity and search granularity: higher search granularity leads to higher path planning accuracy but higher computational complexity, while larger planning spaces lead to higher computational complexity.

[0158] like Figure 3 As shown, the real-time cluster path planning of the UAV adopts the path planning A* algorithm or the path planning D* algorithm based on the artificial potential field method. Local planning of conflicting paths:

[0159] In a three-dimensional space map, path planning algorithms such as A* and D* based on potential energy fields are used to perform local path planning at two control points before and after the continuous conflict path. D* represents the Dynamic A Star algorithm; A* represents the A-Star algorithm.

[0160] Path point sequence smoothing calculation: B-spline or polynomial fitting;

[0161] Control point time reallocation: Reconstruct the same Δt path points based on the current UAV speed.

[0162] like Figure 4 As shown, in some other embodiments, a UAV task execution system based on a two-level task allocation algorithm is also provided, and its specific technical content is as follows:

[0163] A UAV task execution system based on a two-level task allocation algorithm includes:

[0164] Information acquisition module: used to obtain payload information of multiple drones;

[0165] A task allocation module is configured to select a corresponding task allocation method based on the payload information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method; when the first-level task allocation method is selected, a nearest neighbor search method and a Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation position and path planning scheme of each UAV; when the second-level task allocation method is selected, a graph neural network algorithm is used to match the formation position and path planning scheme of each UAV;

[0166] Mission execution module: used to generate drone missions based on the matched formation positions and path planning schemes of each drone.

[0167] like Figure 5 As shown, an initial task allocation is constructed according to the target formation; the two-level task allocation algorithm provided by the present invention is used for task pairing, so that the drone swarm selects the corresponding drone formation task according to the paired task; the drone swarm executes the corresponding drone swarm flight formation according to the selected drone formation task; thereafter, when the drone swarm executes the drone task, the swarm visual positioning is used to realize positioning of the drone swarm; path conflict detection is used to prevent path conflicts during drone flight; conflict path local planning: in the process of spatiotemporal path conflict detection, the drone body size and positioning error are taken into consideration, and an air safety corridor is set to supplement the safety hazards caused by the basic conflict detection assumption that the drone is a particle model.

[0168] Because heterogeneous drones have varying payloads, a two-level task allocation algorithm is employed to assign drone formation tasks to scenarios with small and large payload differences, respectively. Using a suitable algorithm to allocate drone formation tasks improves efficiency and ensures that tasks are answered within a reasonable timeframe. Prioritizing task allocation in swarm tasks addresses both cluster task allocation and formation reorganization planning in dynamic and complex environments. All individuals form a stable cluster structure. If any individual leaves the group or the group structure changes for any reason, a new cluster structure quickly forms and remains stable, ensuring cluster autonomy. Optimized map construction, storage, and query logic enable real-time cluster path planning in any scenario.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the concept and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for executing UAV tasks based on a two-level task allocation algorithm, characterized in that: The steps include: Obtain payload information of multiple drones; Selecting a corresponding task allocation method based on the load information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method; When the task allocation method is selected as the first-level task allocation method, the nearest neighbor search method and the Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation position and path planning scheme of each UAV; When the second-level task allocation method is selected as the task allocation method, a graph neural network algorithm is used to match the formation position and path planning scheme of each drone; Generate drone missions based on the matched formation positions and path planning schemes of each drone.

2. The method for executing UAV tasks based on the two-level task allocation algorithm according to claim 1 is characterized in that: The load information includes load type and load weight; Selecting a task allocation method according to the load information specifically includes the following steps: Comparing the payload types of the plurality of drones, and if the payload types of two or more of the plurality of drones are inconsistent, generating payload difference information; if the payload types of the plurality of drones are consistent, comparing the payload weights of the plurality of drones, and generating payload consistency information when the payload weight difference between any two of the plurality of drones is less than a preset weight threshold; generating the payload difference information when the payload weight difference between two or more of the plurality of drones is greater than or equal to the preset weight threshold; If the load consistency information is generated, the task allocation method is selected as the first-level task allocation method; If the load difference information is generated, the task allocation method is selected as the second-level task allocation method.

3. The method for executing UAV tasks based on the two-level task allocation algorithm according to claim 2, characterized in that: The nearest neighbor search method is used to calculate and match the formation positions and path planning schemes of each drone, including the following steps: Collecting preset position data and real-time position data of each of the drones; The preset position data and the real-time position data of each drone are stored in a tree data structure; Configuring multiple initial path planning schemes for each of the drones; Using a nearest neighbor allocation algorithm to query the preset position data that is the nearest neighbor to the real-time position data, so as to match and obtain the formation position of each of the UAVs; According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

4. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 2, characterized in that: The Hungarian matching algorithm is used to calculate the formation position and path planning scheme of each drone. The specific steps are as follows: Collecting preset position data and real-time position data of each of the drones; Configuring the preset position data and the real-time position data of each of the drones into two independent point sets of a bipartite graph; Configuring multiple initial path planning schemes for each of the drones; A Hungarian matching algorithm is used to select the preset position data that matches the maximum weight of the real-time position data from two independent point sets to match and obtain the formation position of each of the UAVs; According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

5. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 3 or 4, characterized in that: Collecting the preset position data of each of the drones includes the following steps: Generate a plurality of dynamic hash 3D map blocks according to the preset motion range of each of the drones; Storing map block indexes of the plurality of dynamically hashed 3D map blocks in an octree map; Obtaining the map block index from the octree map to obtain a plurality of the dynamic hash 3D map blocks; Constructing a three-dimensional space-time trajectory map based on the multiple dynamic hash 3D map blocks obtained by indexing; The position coordinates of each of the UAVs are preset on the three-dimensional space-time trajectory map to obtain the preset position data of each of the UAVs.

6. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 5, characterized in that: Collecting the real-time location data of each drone specifically includes the following steps: Collecting images of each of the drones to obtain a multi-visual image dataset of each of the drones; Build a target detection model based on computer vision recognition methods; Training the target detection model using the multi-visual image dataset of each drone to obtain a drone target detection model; Using the drone target detection model to identify each of the drones, and using a multi-target tracking algorithm to track each of the identified drones; Calculating the relative speed of each of the identified and tracked drones using an optical flow method; Calculate the distance change between each of the drones based on a monocular ranging method; The real-time position data of each of the UAVs is generated according to the relative speed of each of the UAVs and the distance change between each of the UAVs.

7. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 2, characterized in that: When multiple drones perform a drone swarm mission according to the drone formation mission, a graph neural network algorithm is used to match the formation positions and path planning schemes of each drone. The specific steps are as follows: Constructing a preset node topology graph based on the preset position data, preset load types, and preset load weights of the plurality of drones; wherein the node information of the preset node topology graph includes the preset position data, the preset load type, and the preset load weight of each drone, and a line connecting two nodes of the preset node topology graph represents an edge; Build a graph neural network model; Training the graph neural network model using the preset position data of the plurality of drones, the preset load type, the preset load weight, the edges of the preset node topology graph, and the edge weights of the preset node topology graph to obtain a node graph generation model; Collect real-time location data, real-time payload type, and real-time payload weight of multiple drones to build a real-time node topology map; After inputting the real-time position data of the plurality of drones, the real-time load types, the real-time load weights, the edges of the real-time node topology graph, and the edge weights of the real-time node topology graph into the node graph generation model, the node graph generation model outputs a predicted node topology graph; Matching the position data of each node in the predicted node topology map with the real-time position data of each drone, matching the load type of each node in the predicted node topology map with the real-time load type of each drone, and matching the load weight of each node in the predicted node topology map with the real-time load weight of each drone to match the formation position of each drone; Configuring multiple initial path planning schemes for each of the UAVs according to the formation position; According to the formation position of each of the UAVs, a collision-free path plan is first selected from a plurality of initial path planning plans, and then a path planning plan with the shortest total distance is selected from the collision-free path plans to obtain the path planning plan.

8. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 1, characterized in that: Generating a drone mission based on the matched formation position and the path planning scheme specifically includes the following steps: Setting a safe flight distance according to the body size and positioning error of each of the drones; Generate a path sequence smooth route for each of the UAVs according to the formation position, the path planning scheme, and the safe flight distance; Determining whether the smoothed paths of the path sequences of the respective UAVs have intersections at the same time step, and if so, performing local path optimization on the smoothed paths of the path sequences of the respective UAVs having intersections to obtain final path planning curves for the respective UAVs; if not, selecting the smoothed paths of the path sequences as the final path planning curves; The UAV mission is generated according to the final path planning curve.

9. The method for executing UAV tasks based on a two-level task allocation algorithm according to claim 8, characterized in that: The path sequence smoothing route of each UAV having an intersection is locally optimized, and the specific steps are as follows: A path planning A* algorithm or a path planning D* algorithm based on an artificial potential field method is used to perform local path optimization on the path sequence smooth route of each of the UAVs having intersections.

10. A UAV task execution system based on a two-level task allocation algorithm, characterized in that: include: Information acquisition module: used to obtain payload information of multiple drones; A task allocation module is configured to select a corresponding task allocation method based on the payload information to match the formation position and path planning scheme of each UAV; wherein the task allocation method includes a first-level task allocation method and a second-level task allocation method; when the first-level task allocation method is selected, a nearest neighbor search method and a Hungarian matching algorithm are used for multi-threaded parallel computing to match the formation position and path planning scheme of each UAV; when the second-level task allocation method is selected, a graph neural network algorithm is used to match the formation position and path planning scheme of each UAV; Mission execution module: used to generate drone missions based on the matched formation positions and path planning schemes of each drone.

Images