A heterogeneous unmanned aerial vehicle cluster decentralized cooperative task system and method

Abstract

The application discloses a kind of heterogeneous unmanned aerial vehicle cluster decentralized cooperative task systems and methods, system is applied to heterogeneous unmanned aerial vehicle cluster, heterogeneous unmanned aerial vehicle cluster uses decentralized communication architecture, system includes task allocation subsystem, path planning subsystem and cooperative search subsystem;Task allocation subsystem carries out dynamic task allocation to heterogeneous unmanned aerial vehicle cluster according to unmanned aerial vehicle real-time state vector, and output task allocation scheme;Path planning subsystem generates and real-time re-plans feasible flight path according to task allocation scheme, obstacle map and unmanned aerial vehicle dynamics constraint;Cooperative search subsystem dynamically divides search area according to target probability thermodynamic diagram and generates hierarchical cooperative search instruction.The application solves the problems of traditional cluster task mismatch, search blind area, path infeasibility and single-point failure, and improves the search efficiency and task continuity in complex low-altitude scenarios.

Description

Technical Field

[0001] This invention relates to the field of collaborative control technology for unmanned aerial vehicles (UAVs), and in particular to a decentralized collaborative task system and method for heterogeneous UAV swarms. Background Technology

[0002] In the current context of the booming low-altitude economy, existing drone swarm technology has the following limitations in terms of mission coordination: 1. Traditional task allocation methods do not fully consider the heterogeneous characteristics of fixed-wing UAVs (long endurance), rotary-wing UAVs (high maneuverability), and loitering UAVs (rapid pinpoint reconnaissance), leading to task load imbalance. For example, fixed-wing UAVs are inefficient when performing low-altitude precision search missions, while rotary-wing UAVs have insufficient endurance when performing long-distance transport missions.

[0003] 2. Existing search methods mostly use static area division, which lacks dynamic environmental adaptability. In complex low-altitude environments (such as forests and urban building clusters), search blind spots are easily formed. Moreover, each UAV only relies on its own sensor data and fails to achieve multi-aircraft information sharing and collaboration, resulting in a low probability of target detection.

[0004] 3. Traditional algorithms (such as A) Dijkstra's replanning capability is weak in dynamic obstacle environments (such as sudden weather or moving interference sources), and it does not incorporate UAV dynamic constraints (such as minimum flight speed of fixed wings and maximum climb angle of rotors), which can easily generate infeasible paths, affecting operational safety and efficiency.

[0005] 4. The existing cluster relies on the central base station for communication, which poses a single point of failure risk. Communication interruptions are likely to occur in scenarios such as electromagnetic interference, terrain obstruction, or remote areas, leading to cluster loss of control and affecting the continuity and reliability of tasks. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a decentralized collaborative task system and method for heterogeneous drone swarms. Through heterogeneous task allocation adaptation, dynamic collaborative regional search, multi-constraint optimization of path planning, and the robustness and resilience of decentralized network communication modules, the system improves the operational efficiency and feasibility of heterogeneous drone swarms in low-altitude scenarios.

[0007] This invention provides a decentralized collaborative task system for heterogeneous drone swarms, applied to heterogeneous drone swarms. The heterogeneous drone swarm adopts a decentralized communication architecture, and the specific technical solution is as follows: The system includes a task allocation subsystem, a path planning subsystem, and a collaborative search subsystem; The task allocation subsystem dynamically allocates tasks to the heterogeneous UAV cluster based on the UAV's real-time state vector and outputs the task allocation scheme. The path planning subsystem and the task allocation subsystem interact based on the decentralized communication architecture, and generate and replan feasible flight paths in real time according to the task allocation scheme, obstacle map and UAV dynamics constraints; The collaborative search subsystem interacts with the path planning subsystem and the task allocation subsystem based on the decentralized communication architecture, dynamically divides the search area according to the target probability heatmap, and generates hierarchical collaborative search instructions.

[0008] Furthermore, the task allocation subsystem includes a particle swarm task allocation unit, which employs an improved particle swarm algorithm and embeds a dynamic weight adjustment strategy in the fitness function to dynamically allocate long-endurance tasks, high-maneuverability tasks, or rapid fixed-point reconnaissance tasks based on the UAV's remaining battery power, payload capacity, and heterogeneous characteristics of the UAV model.

[0009] The dynamic adjustment strategy is to prioritize the allocation of fixed-wing UAVs to long-endurance missions, rotary-wing UAVs to high-mobility missions, and loitering UAVs to rapid fixed-point reconnaissance missions, thereby achieving global load balancing.

[0010] Furthermore, the task allocation subsystem also includes a distributed alliance game negotiation unit, which dynamically divides the heterogeneous drone cluster into several task alliances according to task type, and completes several rounds of bidding within the task alliances to generate a task allocation scheme under the condition that the communication overhead is reduced by ≥70% compared with centralized allocation.

[0011] Furthermore, the task allocation subsystem also includes a multi-objective collaborative optimization unit, which constructs the fitness function based on three objectives: range cost, time cost, and operational efficiency.

[0012] The multi-objective collaborative optimization unit is used to construct a fitness function in the particle swarm task allocation unit by weighting three objectives: range cost, time cost, and operational efficiency, and introduces a deadlock detection and repair mechanism to reduce the task conflict rate from 35% to ≤8%, thereby achieving global load balancing.

[0013] Furthermore, the path planning subsystem includes a chaos initialization unit, a dynamic constraint fusion unit, and a path iteration generation unit; The chaotic initialization unit obtains the key points of the route according to the task allocation scheme and uses Kent chaotic mapping to generate an initial path population; The dynamic constraint fusion unit transforms the UAV's minimum flight speed, maximum climb angle, and turning radius into hard constraint penalty functions to filter paths; The path iteration generation unit uses the goat algorithm, which introduces an α-stable distribution for dynamic perturbation and iteratively outputs the optimal path.

[0014] Furthermore, the path planning subsystem also includes a multimodal environment perception and fusion unit, which receives global obstacle data in real time, updates the obstacle map in real time, and provides it to the chaos initialization unit.

[0015] Furthermore, the collaborative search subsystem includes a multi-machine information sharing protocol unit, a dynamic region division unit, and a hierarchical collaborative search strategy unit; The multi-machine information sharing protocol unit exchanges sensor data of each drone in the heterogeneous drone cluster in real time through the decentralized communication architecture, and generates the target probability heatmap using kernel density estimation. The dynamic region division unit uses the centroid of the target probability heatmap as the seed point and updates the centroid Voronoi diagram boundary as needed to obtain the updated data. The hierarchical collaborative search strategy unit controls the drone swarm to perform searches based on the updated data and according to the set collaborative strategy logic.

[0016] Furthermore, the collaborative search subsystem also includes a dynamic task reassignment mechanism unit. When the UAV communication is interrupted or a target is detected, the task handover is completed within a set time through a distributed alliance game negotiation unit, and the task is reorganized through the task allocation subsystem.

[0017] This invention also provides a decentralized collaborative method for heterogeneous drone swarms, characterized in that, based on the aforementioned decentralized collaborative task system for heterogeneous drone swarms, the method includes: S1: The task allocation subsystem outputs a task allocation scheme based on the real-time state vector of the heterogeneous UAVs. The task allocation scheme specifies the long-endurance, high-maneuverability, or rapid fixed-point reconnaissance tasks to be undertaken by each UAV. S2: The path planning subsystem, based on the task allocation scheme, obstacle map and UAV dynamic constraints, uses the chaotic augmented goat algorithm to generate and replan feasible flight paths in real time, so that each UAV obtains a sequence of waypoints with timestamps. S3: The cooperative search subsystem performs a hierarchical cooperative search along the waypoint sequence, as follows: S301: Each UAV broadcasts sensor data in real time through a decentralized communication architecture, and the kernel density estimation is fused to obtain a global target probability heatmap; S302: Using the centroid of the heatmap as the seed point, and combining the real-time location and remaining power of each machine, update the boundary of the centroid Voronoi diagram as needed to obtain the search area allocation result that matches the current cluster capability; S303: High-altitude UAVs cruise along the Voronoi boundary and maintain wide-angle imaging, while low-altitude UAVs perform detailed searches along a spiral descent path centered on their respective Voronoi centroids. If the obstacle map is updated or a target is discovered during flight, the path planning subsystem returns to S2 with the new key points to re-execute the chaotic augmented goat algorithm, outputs the corrected route, and continues the search.

[0018] Furthermore, it also includes: when the drone communication is interrupted, the battery power is less than the set threshold, or the target is found in the search in S303, the task handover is completed through the distributed alliance game negotiation unit, and the new task package is reconfirmed, tracked or located through the task allocation subsystem, and the process returns to step S1 to be re-executed.

[0019] The beneficial effects of this invention are as follows: 1. The system of this invention is applied to heterogeneous drone clusters to realize a decentralized communication architecture, turning each drone into a routing node at the same time. This solves the risk of cluster loss of control due to single point failure of the central base station and electromagnetic shielding, ensuring the continuous and uninterrupted task chain. Through the task allocation subsystem, dynamic task allocation is performed based on the characteristics of heterogeneous drone models using a hybrid allocation mechanism of particle swarm optimization and alliance game theory. This solves the problem of load imbalance and energy waste caused by task mismatch due to insufficient consideration of drone model characteristics in traditional task allocation methods.

[0020] 2. This invention uses Kent chaotic mapping + α-stable perturbation goat algorithm for path planning, embedding dynamic hard constraints into the iterative cost, thus solving the problems of traditional A... Algorithms like this are prone to getting stuck in local conditions, slow replanning, and generating paths that cannot be flown.

[0021] 3. This invention eliminates the search blind spots of static area division by using a probabilistic heatmap-driven centroid Voronoi dynamic partitioning and a dual-layer collaborative mode of "high-altitude wide-area imaging - low-altitude spiral fine search", thereby improving the target detection speed and area coverage. At the same time, through a dynamic task reassignment mechanism, sub-tasks such as confirmation, tracking, and positioning are re-matched within 30 seconds, solving the problem of target loss due to lack of fine sensors or insufficient battery life after detection. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall system architecture and data interaction logic of the present invention.

[0023] Figure 2 This is a schematic diagram of the task input to path execution process of the present invention.

[0024] Figure 3 This is a schematic diagram of the collaborative search process based on the centroid Voronoi diagram of the present invention. Detailed Implementation

[0025] The technical solutions in the embodiments of the present invention are clearly and completely described in the following description. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of the embodiments of the present invention, it should be noted that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is conventionally placed during use, or the orientation or positional relationship in which those skilled in the art conventionally understand it during use. This is only for the convenience of describing the present invention and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of the present invention. Furthermore, the terms "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0028] Example 1 Embodiment 1 of the present invention discloses a decentralized collaborative task system for heterogeneous drone swarms, applied to heterogeneous drone swarms, wherein the heterogeneous drone swarms adopt a decentralized communication architecture, such as... Figure 1 As shown, the system includes a task allocation subsystem, a path planning subsystem, and a collaborative search subsystem.

[0029] The task allocation subsystem dynamically allocates tasks to the heterogeneous UAV cluster based on the UAV's real-time state vector and outputs the task allocation scheme. In a preferred embodiment, the task allocation subsystem includes a particle swarm task allocation unit. The particle swarm task allocation unit adopts an improved particle swarm algorithm and embeds a dynamic weight adjustment strategy in the fitness function to dynamically allocate long-endurance tasks, high-maneuverability tasks, or rapid fixed-point reconnaissance tasks according to the UAV's remaining battery power, payload capacity, and heterogeneous characteristics of the UAV model.

[0030] Specifically, the dynamic adjustment strategy is to prioritize the allocation of fixed-wing UAVs to long-endurance missions, rotary-wing UAVs to high-mobility missions, and loitering UAVs to rapid fixed-point reconnaissance missions, thereby achieving global load balancing.

[0031] In a preferred embodiment, the task allocation subsystem further includes a distributed alliance game negotiation unit and a multi-objective collaborative optimization unit; The distributed alliance game negotiation unit works in collaboration with the particle swarm task allocation unit to dynamically divide the heterogeneous UAV cluster into several task alliances according to task type, and complete several rounds of bidding within the task alliances through the contract network protocol, so as to generate a task allocation scheme with a communication overhead that is ≥70% lower than that of centralized allocation. The specific process is as follows: S101: Based on the type of task to be performed (such as search, relay, transportation, etc.), the entire drone swarm is dynamically divided into 2-5 task alliances; the division is based on the matching degree between task attributes and drone basic capabilities, ensuring that drones in each alliance have the basic conditions to complete the corresponding type of task. S102: Each task alliance shares real-time status data of all drones in the alliance (including remaining battery power, payload type, payload matching degree, current location, etc.) through a decentralized communication module, and selects drones with the best task adaptability as candidate execution nodes based on preset screening conditions (such as payload matching degree > 80% and remaining battery power > 60%). S103: The Contract Network protocol is used to initiate local negotiations within the alliance. The candidate UAV sends a mission commitment to other nodes in the alliance, clarifying its own ability to perform the target mission, the expected cost (range, time) and the expected operational efficiency, and initiates an auction application. S104: Advance consensus through a two-round negotiation process of quotation → counter-quote: Quotation stage: Candidate drones submit initial mission execution plans (including mission execution resource allocation plans, completion time limits, etc.); Counter-quote phase: Other nodes in the alliance evaluate the initial plan and propose adjustments based on issues such as cost reasonableness and resource conflicts. The candidate drones optimize the plan based on the suggestions and resubmit it. S105: After two rounds of negotiations, all nodes within the alliance quickly reached a consensus on task allocation, determined the final task allocation plan, and clarified the specific task responsibilities and cooperation requirements of each drone. S106: Through a decentralized communication module, the final allocation scheme is synchronized to the cluster's global state database, while reserving a dynamic adjustment interface; when there are changes in tasks or abnormal drone status (such as insufficient power or loss of connection), the corresponding task alliance can quickly restart the local negotiation process to complete task reorganization and redistribution, ensuring the continuity of task execution.

[0032] The multi-objective collaborative optimization unit is used in the particle swarm task allocation unit to construct a fitness function with three weighted objectives: range cost, time cost, and operational efficiency, and introduces a deadlock detection and repair mechanism to reduce the task conflict rate from 35% to ≤8%, thereby achieving global load balancing.

[0033] The fitness function is expressed as follows:

[0034] in, Represents particles The final fitness value, The larger the size, the better the solution. The plan is invalid; , , Weights and weight constraints: ; Represents the deadlock penalty coefficient ( (It can be dynamically adjusted).

[0035] The optimization terms for each sub-objective are calculated (all normalized to [0,1]), as follows: Optimization of travel cost , means as follows:

[0036] Representation scheme Total flight range of all drones in the region; This represents the maximum total distance traveled across all possible routes. This represents the theoretical minimum total range.

[0037] Time cost optimization items , means as follows:

[0038] Representation scheme Total task completion time; Indicates the maximum possible completion time; This represents the theoretical minimum completion time (calculated based on the maximum flight speed and shortest path of the UAV).

[0039] Work efficiency optimization items , means as follows:

[0040] Indicates load matching weight ( ); The load matching degree is represented and calculated as follows:

[0041] in, The number of drones participating in the mission. (No. (Payment matching degree of the drone) At 0.8, the scheme invalidation determination is triggered directly; : Expected quality of task completion (value range [0,1], calculated based on data such as historical operation accuracy and payload stability of UAV).

[0042] The deadlock condition is:

[0043] in, This represents the number of resource conflicts (the number of drone pairs competing for the same resource). Indicates the task waiting time (the cumulative waiting time during which the task cannot be started); This indicates the preset waiting threshold.

[0044] Deadlock penalty (Normalized to [0,1]), expressed as follows:

[0045] The number of remaining conflicts after the repair is: ( (Number of resource conflicts after repair) The percentage of repair time spent is: Special case: If deadlock repair fails, .

[0046] plan For a statement to be valid, both of the following conditions must be met simultaneously: Heterogeneous resource fundamental constraints: (Lower limit of load matching) (Minimum remaining battery power for the drone); and fitness value constraints: .

[0047] If any condition is not met, the solution is deemed invalid and excluded from subsequent iterations.

[0048] The path planning subsystem and the task allocation subsystem interact based on the decentralized communication architecture, and generate and replan feasible flight paths in real time according to the task allocation scheme, obstacle map and UAV dynamics constraints; In a preferred embodiment, the path planning subsystem includes a chaos initialization unit, a dynamic constraint fusion unit, and a path iteration generation unit; The chaotic initialization unit obtains the key points of the route according to the task allocation scheme and uses Kent chaotic mapping to generate an initial path population; The dynamic constraint fusion unit transforms the UAV's minimum flight speed, maximum climb angle, and turning radius into a hard constraint penalty function (the path cost increases by 10 times when the minimum speed / maximum climb angle is violated), filters the path, and ensures that the generated path meets the UAV's physical limitations (such as turning radius ≥ 50m and climb angle ≤ 30°), thus ensuring the safety and stability of the UAV during flight. The path iteration generation unit adopts the goat algorithm, which introduces α-stable distributed random numbers (feature index α=1.5, scale parameter γ=0.8), dynamically perturbs the path inflection points every 500ms, and iteratively outputs the optimal path.

[0049] In a preferred embodiment, the path planning subsystem further includes a multimodal environment perception fusion unit, which receives global obstacle data in real time, including obstacle data collected by airborne lidar and visual camera sensors, as well as obstacle data shared by heterogeneous UAV clusters through a decentralized communication architecture, updates the obstacle map in real time, and provides it to the chaos initialization unit.

[0050] The collaborative search subsystem interacts with the path planning subsystem and the task allocation subsystem based on the decentralized communication architecture, dynamically divides the search area according to the target probability heatmap, and generates hierarchical collaborative search instructions.

[0051] In a preferred embodiment, the collaborative search subsystem includes a multi-machine information sharing protocol unit, a dynamic region division unit, and a hierarchical collaborative search strategy unit; The multi-machine information sharing protocol unit exchanges LiDAR and visual camera sensor data of each UAV in the heterogeneous UAV cluster in real time through the decentralized communication architecture, and generates the target probability heatmap using kernel density estimation. The dynamic region division unit uses the centroid of the target probability heatmap as the seed point and updates the centroid Voronoi diagram boundary as needed to obtain the updated data. The hierarchical collaborative search strategy unit controls the drone swarm to perform searches based on the updated data and according to the set collaborative strategy logic.

[0052] The collaborative strategy adopts a two-layer architecture of "high-altitude mapping - low-altitude fine search": the fixed-wing UAV cruises along the Voronoi boundary at a speed of 30 m / s and broadcasts the centroid coordinates of the heat map to the rotary-wing UAV every 20 seconds; the rotary-wing UAV performs a spiral search for the centroid point at a speed of 20 m / s (radius 500 m, layer spacing 100 m), and when the target is found, the location is marked through a decentralized network to trigger collaborative verification by 3 neighboring UAVs.

[0053] In a preferred embodiment, the collaborative search subsystem further includes a dynamic task reassignment mechanism unit. When the drone's battery level is less than 20%, communication is interrupted, or a target is detected, the task handover is completed within 30 seconds through a distributed alliance game negotiation unit (the nearest drone takes over first), and the task is reorganized through the task allocation subsystem.

[0054] Example 2 Embodiment 2 of the present invention discloses a decentralized collaborative method for heterogeneous drone swarms, based on the decentralized collaborative task system for heterogeneous drone swarms described in Embodiment 1 above, such as... Figure 2 As shown, the specific method is as follows: S1: The task allocation subsystem outputs a task allocation scheme based on the real-time state vector of the heterogeneous UAVs. The task allocation scheme specifies the long-endurance, high-maneuverability, or rapid fixed-point reconnaissance tasks to be undertaken by each UAV. Specifically, each drone collects differentiated capability parameters (fixed-wing endurance, rotor maneuver radius, and loitering drone speed) in real time through onboard sensors, and broadcasts them to the cluster at 500ms intervals through a decentralized communication module, thus constructing a dynamic resource pool containing 12-dimensional state vectors (position, battery level, payload type, maximum range, etc.). An improved particle swarm optimization algorithm is used to output a task allocation scheme. By embedding a dynamic weight adjustment formula into the fitness function (the inertial weight decreases linearly with the remaining battery power of the UAV, and long-distance tasks (such as area scanning with a radius > 10km) are given priority to fixed-wing UAVs, while low-altitude high-maneuverability tasks (such as fine reconnaissance within 300m) are given to rotary-wing UAVs).

[0055] In step S1, in this embodiment, based on the distributed alliance game negotiation unit, 2-5 task alliances are dynamically generated according to task type (search / relay, etc.), and distributed bidding is realized through the contract network protocol: the drone with the best adaptability (such as payload matching degree > 80% and remaining power > 60%) sends the task commitment, and after 2 rounds of negotiation (bid → counter-bid), the final allocation scheme is formed, which reduces the communication overhead by 70% compared with the traditional centralized allocation.

[0056] The process is as described in steps S101-S106 of Example 1, and will not be repeated here.

[0057] S2: The path planning subsystem, based on the task allocation scheme, obstacle map and UAV dynamic constraints, uses the chaotic augmented goat algorithm to generate and replan feasible flight paths in real time, so that each UAV obtains a sequence of waypoints with timestamps. The specific process is as follows: S201: First, an initial path population is generated using the Kent chaotic mapping, with the mapping equation being:

[0058] in, The value of the nth generation of the chaotic sequence. r These are the parameters for controlling the chaotic state.

[0059] Based on this mapping, 100 sets of continuous path node sequences are generated (after normalization, covering more than 90% of the search space), forming an initial path population (satisfying the basic connectivity requirements of adjacent node spacing ≤ the maximum single flight distance of the UAV and not crossing known fixed obstacles).

[0060] The fitness value of each initial path is then calculated. F (Comprehensive path cost of fusion penalty terms), filter out F The minimum path is used as the initial global optimal solution. Each path itself is the optimal solution. .

[0061] S202: The fitness function of the fusion dynamic constraint penalty term is used to evaluate the path quality, and the expression is:

[0062] in, , , The weights are: L (summing up to 1, dynamically adjusted according to task priority), T (total path length, sum of Euclidean distances between nodes), P (flight time, calculated from L and the average speed of the UAV), and P (penalty term).

[0063] in, The basic path cost is used as a penalty when constraints such as minimum speed and maximum climb angle are violated, forcibly eliminating infeasible paths.

[0064] S203: Simulates the large-scale foraging behavior of goats, updates path nodes based on the exploration equation to expand the search range, and introduces a symmetric α-stable distribution dynamic perturbation to break through local optima. The formula for fusing exploration and perturbation is as follows:

[0065] in, For the first i Path number t+1 Substitute position, α To reduce the exploration step size coefficient, () represents an α-stable random number; The timing of the disturbance is as follows: it only applies to the turning points of the path, the amplitude is dynamically adjusted according to the obstacle density, and the disturbance is checked by a penalty function after it is performed. If it violates the rules, the amplitude is halved and the disturbance is performed again.

[0066] S204: Simulating goat gathering behavior towards high-quality food sources, guiding pathways towards... and To refine and optimize accuracy, the formula is updated as follows:

[0067] in, β To develop learning factors, γ For local disturbance coefficients, () We still use α-stable distribution random numbers to ensure that the development process is both accurate and robust.

[0068] S205: For continuous L For stagnant paths whose fitness value changes less than a threshold, a jump strategy is implemented (randomly replacing some nodes to generate local path segments); at the same time, the 10% of paths with the worst fitness are removed, and the same number of new paths are regenerated through Kent chaotic mapping to replenish the population and maintain diversity.

[0069] S206: Calculate the fitness values ​​(including penalty terms) of all paths after iteration, and update... and each The iteration terminates if any of the following conditions are met: Reaching the preset maximum number of iterations or The F-value changes by less than a threshold over L consecutive generations, or the total path cost drops below a preset threshold. Output after termination The corresponding path serves as the basic planning path.

[0070] S3: The cooperative search subsystem performs a hierarchical cooperative search along the waypoint sequence, such as... Figure 3 As shown, the details are as follows: S301: Each UAV broadcasts sensor data in real time through a decentralized communication architecture, and the kernel density estimation is fused to obtain a global target probability heatmap; Specifically, each drone collects environmental data in real time through onboard sensors (LiDAR, visual cameras, etc.), including terrain features, information on suspected target points, and the accuracy parameters of the sensors themselves; at the same time, it collects auxiliary data on its own status (such as sensor working status and current position coordinates). The data acquisition frequency is matched with the type of sensor recorded by the UAV, and no specific limitation is made here. In this embodiment, the high-altitude UAV acquires global images at a frequency of 10Hz. Each drone broadcasts the collected sensor data to neighboring nodes via a decentralized communication architecture. Nodes then forward the data through multi-hop relay links, monitoring link quality in real time during transmission. All data is ultimately shared at the cluster level through distributed node interaction. The data receiving end preprocesses the sensor data shared by the entire cluster, filters out abnormal data (such as invalid values ​​caused by sensor failure), and constructs a fusion model based on the kernel density estimation (KDE) algorithm. It uses the suspected target points collected by each UAV as the core samples, sets the kernel function bandwidth in combination with the sensor accuracy parameters, and calculates the probability of target existence at each spatial point in the search area. At the same time, a Bayesian model is introduced to fuse historical search data, correct and optimize the probability values, and generate a target probability heatmap with a resolution of 10m×10m.

[0071] S302: Using the centroid of the heatmap as the seed point, and combining the real-time location and remaining power of each machine, update the boundary of the centroid Voronoi diagram as needed to obtain the search area allocation result that matches the current cluster capability; Specifically as follows: Based on the obtained target probability heatmap, the target existence probability of each pixel in the region is used as the weight, and combined with the spatial coordinates of the pixel, the weighted average coordinates are calculated as the centroid of the heatmap to ensure that the seed points focus on the region where the target has a high probability of existence. Each drone broadcasts its real-time status data, including current location coordinates, remaining battery power, flight endurance, and payload working status, at 500ms intervals via a decentralized communication module. By analyzing the data in real time, drone nodes with search capabilities are selected. Initialize the centroid Voronoi diagram using the centroid of the heatmap as the seed point; calculate the spatial distance between the drone node and the seed point, and between adjacent drone nodes, based on the real-time position coordinates of each drone, and set the node weights in conjunction with the remaining battery power; adopt a dynamic boundary adjustment algorithm to recalculate the Voronoi diagram boundary according to the heatmap update results (changes in target probability distribution) and drone position changes, ensuring that the boundary always adapts to the target distribution and cluster position status; if a drone is disconnected or has insufficient battery power, invalid nodes are automatically removed and the boundary is redefined. Based on the updated centroid Voronoi diagram, the search area is divided into several sub-regions, and each sub-region is assigned to a corresponding UAV node. During the allocation process, it is ensured that the area of ​​the sub-region matches the UAV's endurance and payload coverage (e.g., fixed-wing UAVs are assigned a larger area, and rotary-wing UAVs are assigned a smaller, more refined search area). Finally, the allocation result, which includes the boundary coordinates of each UAV sub-region and the search priority, is output and broadcast to each execution node through a decentralized communication network to achieve dynamic allocation of the search area.

[0072] S303: Based on the regional allocation results and UAV dynamic constraints, initialize the flight path parameters of each UAV and perform high-low altitude cooperative search; Specifically, the high-altitude UAV cruises along the Voronoi boundary along the initial route while maintaining wide-angle imaging, and collects global image data in real time. The updated local heat map centroid coordinates are broadcast to the corresponding low-altitude UAV nodes through a decentralized communication network. The low-altitude UAVs receive the centroid coordinates broadcast by the high-altitude UAVs, and perform a fine search with their assigned Voronoi region centroid as the search center, following a spiral descent route (with set search radius, interlayer spacing, and flight speed). They collect detailed target data in the region in real time, and simultaneously receive status and environmental data from adjacent UAVs to avoid search overlap and collisions. If the obstacle map is updated or a target is discovered during flight, the path planning subsystem returns to S2 with the new key points to re-execute the chaotic augmented goat algorithm, outputs the corrected route, and continues the search.

[0073] During the above process, when the drone communication is interrupted, the battery power is less than the set threshold, or the target is found in the search in S303, the task handover is completed through the distributed alliance game negotiation unit, and the new task package is reconstructed, confirmed, tracked or located through the task allocation subsystem, and the process returns to step S1 to be re-executed.

[0074] Specifically, search roles are assigned based on the heterogeneous characteristics of UAVs. For example, fixed-wing UAVs (with long endurance) are assigned to high-altitude patrol missions, while rotary-wing UAVs (with high maneuverability) are assigned to low-altitude fine-grained search missions.

[0075] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A decentralized collaborative task system for heterogeneous unmanned aerial vehicle (UAV) swarms, characterized in that, This system is applied to heterogeneous drone swarms, which employ a decentralized communication architecture. The system includes a task allocation subsystem, a path planning subsystem, and a collaborative search subsystem. The task allocation subsystem dynamically allocates tasks to the heterogeneous UAV cluster based on the UAV's real-time state vector and outputs the task allocation scheme. The path planning subsystem and the task allocation subsystem interact based on the decentralized communication architecture, and generate and replan feasible flight paths in real time according to the task allocation scheme, obstacle map and UAV dynamics constraints; The collaborative search subsystem interacts with the path planning subsystem and the task allocation subsystem based on the decentralized communication architecture, dynamically divides the search area according to the target probability heatmap, and generates hierarchical collaborative search instructions.

2. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 1, characterized in that, The task allocation subsystem includes a particle swarm task allocation unit. The particle swarm task allocation unit adopts an improved particle swarm algorithm and embeds a dynamic weight adjustment strategy in the fitness function to dynamically allocate long-endurance tasks, high-maneuverability tasks, or rapid fixed-point reconnaissance tasks according to the UAV's remaining battery power, payload capacity, and heterogeneous characteristics of the UAV model.

3. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 2, characterized in that, The task allocation subsystem also includes a distributed alliance game negotiation unit, which dynamically divides the heterogeneous drone cluster into several task alliances according to task type, and completes several rounds of bidding within the task alliances to generate a task allocation scheme.

4. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 2 or 3, characterized in that, The task allocation subsystem also includes a multi-objective collaborative optimization unit, which constructs the fitness function based on three objectives: range cost, time cost, and operational efficiency.

5. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 1, characterized in that, The path planning subsystem includes a chaos initialization unit, a dynamic constraint fusion unit, and a path iteration generation unit; The chaotic initialization unit obtains the key points of the route according to the task allocation scheme and uses Kent chaotic mapping to generate an initial path population; The dynamic constraint fusion unit transforms the UAV's minimum flight speed, maximum climb angle, and turning radius into hard constraint penalty functions to filter paths; The path iteration generation unit uses the goat algorithm, which introduces an α-stable distribution for dynamic perturbation and iteratively outputs the optimal path.

6. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 5, characterized in that, The path planning subsystem also includes a multimodal environment perception and fusion unit, which receives global obstacle data in real time, updates the obstacle map in real time, and provides it to the chaos initialization unit.

7. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 1, characterized in that, The collaborative search subsystem includes a multi-machine information sharing protocol unit, a dynamic region division unit, and a hierarchical collaborative search strategy unit. The multi-machine information sharing protocol unit exchanges sensor data of each drone in the heterogeneous drone cluster in real time through the decentralized communication architecture, and generates the target probability heatmap using kernel density estimation. The dynamic region division unit uses the centroid of the target probability heatmap as the seed point and updates the centroid Voronoi diagram boundary as needed to obtain the updated data. The hierarchical collaborative search strategy unit controls the drone swarm to perform searches based on the updated data and according to the set collaborative strategy logic.

8. The heterogeneous unmanned aerial vehicle (UAV) swarm decentralized collaborative task system according to claim 7, characterized in that, The collaborative search subsystem also includes a dynamic task reassignment mechanism unit. When the UAV communication is interrupted or a target is detected, the task handover is completed within a set time through a distributed alliance game negotiation unit, and the task is reorganized through the task allocation subsystem.

9. A decentralized collaborative method for heterogeneous unmanned aerial vehicle (UAV) swarms, characterized in that, Based on the heterogeneous drone swarm decentralized collaborative task system according to any one of claims 1-8, the method includes: S1: The task allocation subsystem outputs a task allocation scheme based on the real-time state vector of the heterogeneous UAVs; S2: The path planning subsystem, based on the task allocation scheme, obstacle map and UAV dynamic constraints, uses the chaotic augmented goat algorithm to generate and replan feasible flight paths in real time, so that each UAV obtains a sequence of waypoints with timestamps. S3: The cooperative search subsystem performs a hierarchical cooperative search along the waypoint sequence, as follows: S301: Each UAV broadcasts sensor data in real time through a decentralized communication architecture, and the kernel density estimation is fused to obtain a global target probability heatmap; S302: Using the centroid of the heatmap as the seed point, and combining the real-time location and remaining power of each machine, update the boundary of the centroid Voronoi diagram as needed to obtain the search area allocation result that matches the current cluster capability; S303: High-altitude UAVs cruise along the Voronoi boundary and maintain wide-angle imaging, while low-altitude UAVs perform detailed searches along a spiral descent path centered on their respective Voronoi centroids. If the obstacle map is updated or a target is discovered during flight, the path planning subsystem returns to S2 with the new key points to re-execute the chaotic augmented goat algorithm, outputs the corrected route, and continues the search.

10. The decentralized collaborative method for heterogeneous drone swarms according to claim 9, characterized in that, Also includes: When the drone communication is interrupted, the battery level is less than the set threshold, or a target is found in the search in S303, the task handover is completed through the distributed alliance game negotiation unit, and the new task package is reconfirmed, tracked or located through the task allocation subsystem, and the process returns to step S1 to be re-executed.

Images