Unmanned aerial vehicle cooperative delivery system for small items in cities

Abstract

The application relates to the technical field of low-altitude economy, and discloses a UAV cooperative delivery system for small and medium-sized object express in cities, which comprises a task publishing module, a task allocation module, a path planning module and a fault response module. The application lays a foundation for cooperative and efficient operation through the task publishing module; the task allocation module gives priority to single machine execution, adopts an auction mechanism and a hierarchical strategy for allocation after complex task splitting, and guarantees task executability; the path planning module optimizes a path with a cost function constraint, dynamically combines a dynamic complexity index of an environment, dynamically performs hybrid planning, and improves path optimality and dynamic environment adaptability; the fault response module determines a fault through weighted voting, relies on controlled retreat landing, a comprehensive cost function site selection and support UAV scheduling, and takes into account flight safety and task continuity; in summary, the application realizes cooperative improvement of delivery efficiency, environment adaptability and operation reliability, and perfectly adapts to complex scene requirements of small and medium-sized object express in cities.

Description

Technical Field

[0001] This application relates to the field of low-altitude economic technology, specifically a drone collaborative delivery system for small and medium-sized urban parcel delivery. Background Technology

[0002] In recent years, emerging service models such as e-commerce, food delivery, and pharmaceutical delivery have developed rapidly, significantly increasing the demands on urban logistics for delivery efficiency and reliability. Traditional same-city delivery methods mainly rely on manual dispatching and ground vehicle transfer of goods. When faced with large-scale, multi-batch orders and complex urban environments, they often suffer from problems such as low dispatching efficiency, inflexible route selection, and difficulty in guaranteeing delivery timeliness. The limitations of traditional systems are even more pronounced in scenarios with frequent road congestion or emergencies.

[0003] Drones, with their advantages of rapid response, direct aerial delivery, and flexible scheduling, have the potential to become an important supplement to urban delivery systems in the future. However, existing drone delivery systems still have shortcomings in task allocation, path planning, and emergency response. For example, task allocation lacks cross-regional dynamic optimization, path planning is inefficient, local obstacle avoidance capabilities are easily limited in complex environments, and the response of task transfer and support mechanisms is not efficient enough when drones malfunction. Summary of the Invention

[0004] The purpose of this application is to provide a drone-based collaborative delivery system for small and medium-sized urban parcel delivery, in order to solve the technical bottleneck problems existing in the current technology for urban logistics services using drones.

[0005] To achieve the above objectives, this application provides a drone-based collaborative delivery system for small and medium-sized urban parcels, the system comprising:

[0006] The task publishing module is configured to generate standardized task data packets based on express delivery information. These standardized task data packets are used for task allocation and path planning during drone delivery.

[0007] The task allocation module is configured to: parse standardized task data packets, analyze the matching degree between candidate drones and tasks, and obtain the final matching degree; determine the drones to execute tasks based on the final matching degree, and preferably allow single drones to execute tasks. When a single drone cannot execute a task, the task is split into multiple sub-tasks, and the sub-tasks are allocated using an auction mechanism and a hierarchical task allocation strategy to determine the drones to execute the sub-tasks.

[0008] The path planning module is configured to: parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built overall path cost function or return cost function; define an environmental dynamic complexity index for dynamic hybrid planning, the environmental dynamic complexity index being determined based on dynamic obstacle density, point cloud data, and scene information entropy; and perform dynamic hybrid planning on RRT. and / or VFH Adjustments were made; and local planning and multi-machine collaboration were optimized.

[0009] The fault response module is configured to: perform fault analysis on the UAV using a weighted voting-based fault determination mechanism; trigger a preset fault response strategy based on the fault analysis results; the fault response strategy includes at least a controlled retreat and landing process; determine the landing point based on a preset comprehensive cost function; and trigger the selection of a UAV to perform the support mission.

[0010] Preferably, the task allocation module includes a task parsing unit, which is used to parse standardized task data packets, analyze the matching degree between candidate UAVs and tasks, and obtain the final matching degree; the task parsing unit is configured as follows:

[0011] Construct a matching graph of task-drone-product point, and calculate the final matching degree for the edge of task-drone. The formula for calculating the final matching degree is:

[0012]

[0013] in: For the calculated task With drones The final match degree between them; For the task With drones Feasibility indicators between, the feasibility indicators The value is or It is defined as when the task The weight of the cargo shall not exceed that of the drone. The current available payload and drone Complete the task When the sum of the required energy and the reserved energy for return does not exceed its current remaining energy, and the estimated mission completion time is no later than the difference between the latest mission completion time and the current time. , indicating task For drones Feasible, otherwise , indicating task For drones Not feasible; For the first The normalized score components, The preset weight coefficients for the corresponding rating components satisfy... ,and , Corresponding to time, energy, load, distance, task priority, and management station load.

[0014] Preferably, the task allocation module includes a single-machine execution unit, which is used to select a single UAV to execute the task based on the final matching degree; the single-machine execution unit is configured as follows:

[0015] When it exists At that time, it was determined that there was a machine capable of completing the task independently. drones A local site priority strategy is adopted to prioritize tasks. The selection of drones is based on proximity and the specific mission requirements. All In the task-drone selection process, the drone output corresponding to the maximum edge value is used as the final task execution value. A single drone.

[0016] Preferably, the task allocation module includes a multi-machine execution unit. This multi-machine execution unit is used to split the task into multiple sub-tasks when a single drone cannot execute the task, and to allocate the sub-tasks using an auction mechanism and a hierarchical task allocation strategy to determine which drone will execute the sub-task. The multi-machine execution unit is configured as follows:

[0017] When a single drone cannot execute a task, the task is split into multiple subtasks, allowing multiple drones to collaboratively execute the subtasks. The minimum number of fragments is determined based on current drone resources. This minimum number of fragments is calculated by dividing the original task's cargo weight by the maximum remaining payload capacity of all available drones, then rounding up. The standard weight of each subtask is obtained by dividing the original task's cargo weight by the minimum number of fragments. For each subtask generated after splitting, a greedy one-way auction mechanism is used to select the optimal drone for execution, and the subtasks are allocated item by item, including:

[0018] The participating drones submit bids for the sub-tasks, and the mathematical expression for the bid function is:

[0019]

[0020] in, and The recalculated drones Execute subtasks Energy consumption and time estimation; and These are preset adjustable weighting coefficients;

[0021] The winning bid rule is for all subtasks. Among the recalculated feasibility conditions, select The drone with the lowest price was chosen as the final sub-task to be executed. drones;

[0022] When there are still unassigned subtasks after allocation through the greedy single-item auction mechanism, a hierarchical task allocation strategy is triggered or a request for cross-regional resource support is made to the central dispatch.

[0023] The execution flow of the hierarchical task allocation strategy is as follows:

[0024] The first level attempts to allocate tasks within the task initiation station using either a single-machine allocation or auction mechanism. If the task initiation station lacks sufficient resources or allocation fails, the process proceeds to the second level, broadcasting the task to adjacent stations and re-evaluating and allocating it based on a corrected matching degree. If the requirements still cannot be met, the process proceeds to the third level, submitting the task to the central scheduling node for global optimization. The formula for calculating the corrected matching degree is:

[0025]

[0026] in: The calculated corrected matching degree; This represents the original final match degree; The preset penalty weight coefficient is used to adjust the degree of impact of load balancing on task allocation; For drones The corresponding site load penalty item is determined based on the site's load level, which is calculated using the following formula:

[0027]

[0028] in, Belonging to the site A collection of drones Indicates drone The current load, A small constant used to avoid division by zero;

[0029] In response to the dynamic characteristics of task urgency changing over time, a dynamic priority mechanism is introduced to calculate the real-time priority score of the task and a preset reassignment trigger condition is set. When the reassignment trigger condition is met, the affected task will be returned to the queue of tasks to be assigned. Based on the updated real-time priority, the task-UAV edge is reconstructed and the latest matching degree is calculated. Then, through a hierarchical task allocation strategy and auction mechanism, the task is rescheduled according to the current UAV resource status.

[0030] The formula for calculating the real-time priority score of the task is as follows:

[0031]

[0032] in, For the task At the present moment Dynamic priority, , and The preset weighting coefficients, Indicates task Static priority, For the task Latest completion time For numerically stable terms, For characterization tasks The quantified value of economic benefits or service level;

[0033] The redistribution triggering conditions are as follows: the task redistribution process is automatically triggered when any of the following situations occur: if the remaining energy of the UAV is insufficient to support the energy consumption required to complete the task and the energy reserve for safe return; if the UAV malfunctions or communication is interrupted; if a new task arrives and causes a change in the task priority structure, and its dynamic priority difference exceeds a preset threshold; or if the expected task completion time has exceeded the latest deadline and the probability of default exceeds the allowed upper limit.

[0034] Preferably, the path planning module includes a total cost unit, which is used to parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built total path cost function or return cost function; the total cost unit is configured as follows:

[0035] Construct a total path cost function and constrain path planning based on the total path cost function; wherein, the mathematical expression of the total path cost function is:

[0036]

[0037] in: Energy consumption estimation for the path; For path time estimation; The integral of obstacle potential energy accumulated along the path is used to quantify obstacle avoidance risk. For position The obstacle potential energy function at the location; , and These are preset weighting coefficients for energy consumption, time, and obstacle penalties, respectively.

[0038] A return-to-origin cost function is constructed, and path planning is constrained based on the return-to-origin cost function to determine the optimal return-to-origin station; wherein, the mathematical expression of the return-to-origin cost function is:

[0039]

[0040] in, and These represent the distance from the drone's current location to the station. Estimated energy consumption and time; The preset site occupancy penalty item is used to reflect the current and short-term berth competition situation; This is a measure of airspace congestion, provided by the central dispatch center and normalized to [specific value]. interval, , , and The preset weighting coefficients are used; the station corresponding to the minimum value of the return cost function is selected as the optimal return station.

[0041] Preferably, the path planning module includes a complexity response unit, which defines a dynamic environmental complexity index for dynamic hybrid planning; the complexity response unit is configured as follows:

[0042] Define the environmental dynamic complexity index, and dynamically switch between global planning and local reactive control based on the environmental dynamic complexity index. The formula for calculating the environmental dynamic complexity index is:

[0043]

[0044] in, This is the calculated environmental dynamic complexity index; This represents the density of dynamic obstacles detected per unit volume; The statistical variance of LiDAR point cloud data is used to quantify the uncertainty of environmental perception. The scene information entropy is calculated based on the semantic segmentation results, reflecting the complexity of semantic categories in the environment; , and The preset weighting coefficients for each indicator; based on preset threshold parameters. and Dynamic hybrid planning is achieved by combining environmental dynamic complexity indicators:

[0045] when At that time, RRT was mainly used. Perform global path planning; when At that time, with VFH As the main force, local real-time obstacle avoidance planning is carried out; when When this happens, enter hybrid mode and use RRT in a weighted coordination manner based on the current complexity. and VFH .

[0046] Preferably, the path planning module includes a local planning optimization unit, which is used to optimize local planning and multi-machine collaboration; the local planning optimization unit is configured as follows:

[0047] A semantically enhanced vector field histogram method is used for real-time heading selection. For each candidate heading, a comprehensive evaluation function is constructed, and the mathematical expression of the comprehensive evaluation function is as follows:

[0048]

[0049] in, For the calculated heading A comprehensive evaluation; The direction of the current global path or local target; This indicates the cost of speed adjustment caused by a change in course; , and These are preset weighting coefficients; For heading The weighted obstacle quantity, which is based on the obstacle's heading. The perceived intensity and preset semantic weights are determined. The direction with the smallest overall evaluation value is selected as the optimal heading; This is a small constant used for numerical stability.

[0050] If the path is blocked by an obstacle, local replanning is immediately triggered. This local replanning obtains a sequence of path points by discretely sampling the globally optimal path and determines the local sharpness of each path point. When a continuous path interval is detected, and all path points in this interval satisfy the condition that the local sharpness is less than a safety threshold and the obstacle's duration is greater than a persistence threshold, the path interval is determined to be permanently blocked, and local RRT is initiated. The replanning process; the replanning area is limited to a scrolling window determined by a preset number of window expansion arcs, in order to ensure that the new planned path can be smoothly integrated into the original global path;

[0051] In multi-drone collaborative tasks, the speed obstacle method and collision risk measurement are used to assess and avoid potential conflicts. To address potential path conflicts and communication overhead among multiple drones, a lightweight dynamic priority negotiation mechanism based on game theory is adopted. This mechanism determines the initial priority weight of each drone based on its static task priority and remaining energy. When multiple drones collide in a local airspace, a round of local negotiation is initiated, and the priority weight of each drone is updated using an exponential recalibration method.

[0052] If a temporary airspace congestion problem is encountered during the mission, an ellipsoidal geometric model is used to set a temporary no-fly zone, thereby triggering local replanning. When local replanning is triggered, it responds to obstacle avoidance constraints, triggers a local rolling replanning mechanism, and incorporates the ellipsoidal constraints corresponding to the temporary no-fly zone as feasible region determination conditions into the planner's solution process.

[0053] During the multi-drone collaborative return process, each drone adopts a distributed consensus iteration strategy for flight altitude and speed scalars;

[0054] When local communication conditions are limited or neighborhood information is insufficient to achieve effective coordination, the central dispatcher will monitor the situation. The central node will only intervene and make the final arbitration when a potential conflict is detected that cannot be resolved through distributed negotiation.

[0055] Preferably, the fault response includes a fault analysis unit, which is used to perform fault analysis on the UAV using a weighted voting-based fault determination mechanism; the fault analysis unit is configured as follows:

[0056] The value of the constructed fault confirmation indicator is determined within a continuous time window; when the fault confirmation indicator is... If the fault is confirmed, continue normal operation and maintain monitoring; if the fault confirmation indicator is... If the fault is detected, the system will immediately switch to the local automatic recovery process to perform the corresponding fault handling and system recovery operations; wherein, the mathematical expression of the fault confirmation index is:

[0057]

[0058] in: The weighted voting value; The preset weighted vote threshold; This is the persistence time threshold; The specific logic for determining the obtained fault confirmation index is as follows: when the weighted voting value continuously exceeds the weighted vote threshold, a fault is confirmed. The determination of continuity is based on the persistence time threshold.

[0059] Preferably, the fault response includes a landing point analysis unit, which is used to determine the landing point based on a preset comprehensive cost function; the landing point analysis is configured as follows:

[0060] If local automatic recovery fails, the UAV will initiate the controlled retreat and landing process, selecting the optimal landing point from the set of nearest safe landing points. For each candidate landing point, an evaluation is performed using a preset comprehensive cost function. Under the premise of satisfying energy constraints, the landing point with the minimum comprehensive cost is selected as the optimal landing point. The mathematical expression of the comprehensive cost function is:

[0061]

[0062] in: This is the current location of the drone; This indicates the distance from the current location to the landing point or the flight cost; Risk assessment for terrain slope and flatness; Scoring the density and hazard level of surrounding obstacles; This reflects the likelihood or availability penalty of the landing site currently being occupied; , , and These are the preset corresponding weighting coefficients.

[0063] Preferably, the fault response includes a rescue drone determination unit, which is used to select the most suitable drone for performing the support mission; the rescue drone determination unit is configured as follows:

[0064] When a drone malfunctions and reports it, a fault report package is generated based on the fault information. Based on the fault report package, a support drone selection process is initiated. The support drone selection process selects the most suitable drone to perform the support mission from the candidate drone set through linear weighted scoring. The linear weighted scoring is normalized based on six dimensions: energy, payload, distance, time, reliability, and scheduling status, to obtain a comprehensive score for the candidate drones. After verifying that the drone meets the capability constraints, the drone with the maximum comprehensive score is selected as the rescue drone.

[0065] The mathematical expression for the comprehensive score is:

[0066]

[0067] in: For the corresponding preset weight coefficients, satisfying These correspond to six dimensions: energy, load capacity, distance, time, reliability, and scheduling status. These are the scores for each dimension after normalization.

[0068] Beneficial Effects: The drone collaborative delivery system for small and medium-sized urban parcels proposed in this application generates standardized task data packages through a task release module, providing unified data support for task allocation and path planning, laying the foundation for efficient collaborative operation. The task allocation module prioritizes single-drone execution, and after complex tasks are broken down, an auction mechanism and hierarchical strategy are used for allocation, which optimizes drone resource utilization, adapts to different task complexities, and ensures task executability. The path planning module uses a cost function to constrain path optimization, and combines dynamic obstacle density, point cloud data, and scene information entropy to construct an environmental dynamic complexity index, which is used to assess the RRT (Real-Time Tolerance). / VFH Dynamic hybrid planning is employed, simultaneously optimizing local planning and multi-drone collaboration to significantly improve path optimality and adaptability to dynamic environments. The fault response module accurately identifies faults through weighted voting, relying on controlled backoff landing, comprehensive cost function-based site selection, and support for drone scheduling, balancing flight safety and mission continuity. In summary, the system achieves a synergistic improvement in delivery efficiency, environmental adaptability, and operational reliability, perfectly meeting the complex needs of urban small-to-medium-sized parcel delivery scenarios. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a structural block diagram of a drone collaborative delivery system for urban small and medium-sized parcel delivery provided in an embodiment of this application;

[0071] Figure 2 This is a structural block diagram of the task allocation module provided in an embodiment of this application;

[0072] Figure 3 This is a structural block diagram of the path planning module provided in an embodiment of this application;

[0073] Figure 4 This is a structural block diagram of the fault response module provided in an embodiment of this application;

[0074] In the diagram: 100, Task Issuance Module; 200, Task Allocation Module; 300, Path Planning Module; 400, Fault Response Module; 201, Task Parsing Unit; 202, Single-Machine Execution Unit; 203, Multi-Machine Execution Unit; 301, Overall Cost Unit; 302, Complexity Response Unit; 303, Local Planning Optimization Unit; 401, Fault Analysis Unit; 402, Landing Point Analysis Unit; 403, Rescue Aircraft Determination Unit.

[0075] The implementation, functional features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0076] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0077] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0078] In response to the technological bottlenecks of urban logistics services utilizing drones, this embodiment discloses a drone collaborative delivery system for urban small and medium-sized parcel delivery. It proposes a task allocation method combining an auction mechanism and a hierarchical task allocation strategy, as well as a path planning method integrating a dynamic hybrid planning system. For ease of understanding, the drone collaborative delivery system for urban small and medium-sized parcel delivery of this embodiment is now summarized as follows:

[0079] The system begins with merchants uploading order information and their address information to the task management terminal. The task management terminal then publishes the task information to the drones. The drones calculate their matching degree based on their own status and provide feedback. The task management terminal uses a matching graph of task-drone-product point, combined with auction and hierarchical strategies, to complete task allocation and broadcast the results. During delivery, the drones employ a dynamic hybrid planning system to execute flight tasks. The system operates in low-dynamic environments using RRT (Real-Time Planning). (Rapidly-exploring Random Tree Star An improved version of the rapidly extended random tree algorithm (which achieves asymptotic optimality optimization of paths by introducing parent node reselection and cost update propagation mechanisms) generates global path optimization energy consumption and switches to VFH in highly dynamic environments. (Vector Field Histogram Star An improved version of the vector field histogram algorithm, which enhances adaptability to complex scenes by iteratively expanding path nodes. It is dominated by LiDAR (Light Laser Detection and Range) and cameras to perceive obstacles in real time and construct semantically enhanced vector field histograms to achieve local obstacle avoidance. RRT is triggered when the path is blocked. Local replanning is implemented via a rolling window. In multi-drone collaborative flight, collision avoidance and cooperative flight are achieved through speed obstacle avoidance and game-theoretic dynamic priority negotiation. Path planning considers power constraints and stable flight strategies throughout the entire process. After mission completion, the UAV enters the return-to-home phase, selecting the optimal return-to-home station through a semantic-assisted dynamic decision-making system, and performing dynamic planning based on visual sensors and airspace information. The entire return-to-home process employs Redirected Replanning (RRT). With VFH Hybrid planning supports rolling replanning. Multiple drones returning to base coordinate altitude and speed through lightweight communication and a consensus algorithm, with the central system only handling conflict arbitration. During a mission, if a drone malfunctions, the system monitors its status through multi-sensor redundancy and weighted voting, triggering a tiered response. The management system selects the appropriate drone for mission transfer and fault support based on a linear weighted score, ensuring delivery continuity and system stability.

[0080] The drone collaborative delivery system for small and medium-sized urban parcel delivery, as described in this embodiment, will now be described in detail.

[0081] Reference Figure 1 , Figure 1 This is a structural block diagram of a drone-based collaborative delivery system for small and medium-sized urban parcels, provided in an embodiment of this application.

[0082] like Figure 1 As shown, this embodiment discloses a drone collaborative delivery system for small and medium-sized urban parcel delivery, the system comprising:

[0083] The task publishing module is configured to generate standardized task data packets based on express delivery information. These standardized task data packets are used for task allocation and path planning during drone delivery.

[0084] In this specific application, the task publishing process is jointly completed by the merchant and the task management system. After a user generates an order on an e-commerce, food delivery, or pharmaceutical delivery platform, the merchant system automatically extracts key parameters such as the destination address, item weight, delivery time, and merchant location based on the order information, and generates a unique corresponding QR code. The merchant uploads this task information to the task management system, which, upon receiving it, formats the order and parses the task parameters to form a standardized task data package. Subsequently, the task management system forwards the task data to the drone scheduling system as input for task allocation and path planning, while simultaneously providing the QR code back to the merchant for verification upon pickup, laying the foundation for efficient drone scheduling and safe delivery.

[0085] The task allocation module is configured to: parse standardized task data packets, analyze the matching degree between candidate drones and tasks, and obtain the final matching degree; determine the drones to execute tasks based on the final matching degree, and preferably allow a single drone to execute the task. When a single drone cannot execute the task, the task is split into multiple sub-tasks, and the sub-tasks are allocated using an auction mechanism and a hierarchical task allocation strategy to determine the drones to execute the sub-tasks.

[0086] Reference Figure 2 , Figure 2 This is a structural block diagram of the task allocation module provided in an embodiment of this application.

[0087] like Figure 2 As shown, specifically, the task allocation module includes a task parsing unit. This unit parses standardized task data packets, analyzes the matching degree between candidate drones and tasks, and obtains the final matching degree. The task parsing unit is configured as follows:

[0088] Construct a matching graph of task-drone-product point, and calculate the final matching degree for the edge of task-drone. The formula for calculating the final matching degree is:

[0089]

[0090] in: For the calculated task With drones The final match degree between them; For the task With drones Feasibility indicators between, the feasibility indicators The value is or It is defined as when the task The weight of the cargo shall not exceed that of the drone. The current available payload and drone Complete the task When the sum of the required energy and the reserved energy for return does not exceed its current remaining energy, and the estimated mission completion time is no later than the difference between the latest mission completion time and the current time. , indicating task For drones Feasible, otherwise , indicating task For drones Not feasible; For the first The normalized score components, The preset weight coefficients for the corresponding rating components satisfy... ,and , Corresponding to time, energy, load, distance, task priority, and management station load.

[0091] In this specific application, the drone scheduling system first constructs a matching graph of task-drone-product point based on task data. For each task... With each drone Calculate the final matching degree And use feasibility indicators to filter out pairs that do not meet the basic constraints. Assume a drone... The current location is ,Task The coordinates of the pickup point are The coordinates of the delivery point are The coordinates of the drone's return-to-home base station or nearest repair point are: To assess the total path length of the drone from its current location, through the pickup point and drop-off point, to its return, the estimated distance is defined as the sum of the distances of the three path segments: the distance from the drone's current location to the pickup point, the distance from the pickup point to the drop-off point, and the distance from the drop-off point to the return base station. Based on this, the estimated flight distance between the drone and the mission can be obtained. The calculation formula is:

[0092]

[0093] If the drone's cruising speed is The loading and unloading time for the task is The estimated total time is defined as the ratio of flight distance to speed plus loading and unloading time for the drone. Complete the task Expected time for:

[0094]

[0095] in, This is a spatial distance function used to calculate the Euclidean distance between any two points. Based on the estimated flight distance. Expected time The unit flight energy consumption of the drone during flight, the additional energy loss caused by the cargo carried in the mission, and the hovering energy consumption during loading and unloading are expected to be significant for the drone. Execute the task energy consumption for:

[0096]

[0097] in, Indicates drone In the mission Total energy consumption; Energy consumption per unit distance of flight; Additional energy consumption per unit weight; Indicates the energy consumption per unit time of hovering; The weight of the cargo for the mission; The loading and unloading dwell time for the task. Define feasibility indicators. When the task cargo weight No more than drones Current available load capacity And the energy required for the drone to complete the mission Reserved energy for return The sum does not exceed its current remaining energy. Meanwhile, the estimated task completion time is... No later than the latest completion time of the task With the current moment When the difference is, This indicates that the task is feasible for the current drone; otherwise, it is considered... 0 is not feasible. The specific implementation is as follows:

[0098]

[0099] During task allocation, the task management system needs to normalize and score various indicators to quantitatively assess the matching degree between different tasks and drones. The scoring components and their calculation formulas are as follows:

[0100]

[0101]

[0102]

[0103]

[0104]

[0105]

[0106] in, The time score component reflects the estimated task time. With the system's maximum allowed time The relative margin; The energy score component represents the remaining energy of the drone. Energy consumption of the task Relationship, For the rated battery capacity of the drone, It is a numerically stable term; The load-bearing score is based on the task weight. Total payload capacity of drones The proportion is determined; For distance scoring components, an exponential decay form is used. For distance scale parameters; The task priority is scored based on its static priority weight. It is determined by the ratio of the highest priority weight of all tasks in the system; The management station is rated for load, which measures the current load on the management station. With its maximum carrying capacity The relative relationship.

[0107] Based on the above design, we have achieved a comprehensive calculation of the final matching degree, thus providing a data foundation for the selection of drones.

[0108] like Figure 2 As shown, specifically, the task allocation module is equipped with a single-machine execution unit, which is used to select the individual UAV to execute the task based on the final matching degree; the single-machine execution unit is configured as follows:

[0109] When it exists At that time, it was determined that there was a machine capable of completing the task independently. drones A local site priority strategy is adopted to prioritize tasks. The selection of drones is based on proximity and the specific mission requirements. All In the task-drone selection process, the drone output corresponding to the maximum edge value is used as the final task execution value. A single drone.

[0110] In the specific application of this embodiment, after the calculation is completed, the system will... The drone-task pairs are used as candidate assignment edges. For each task to be assigned, it is first checked whether there is a drone that meets the single-machine execution conditions. If there is a drone that meets the conditions, the single-machine assignment process is initiated; if there is no drone but the task supports splitting, the process proceeds to the task splitting stage.

[0111] When the task When a task can be completed by a single drone, the drone mission scheduling system adopts a local site priority strategy, allocating tasks accordingly. Assignment will be based on proximity. This will be determined based on feasibility. Among the drones, select the execution unit with the highest matching degree with the mission, that is:

[0112]

[0113] in, For the final execution of the task The drone. After the task is assigned, the drone status will be updated: its location will be updated. Adjusted to task drop-off point Or return to base station, remaining energy updated to Available for load update Simultaneously adjust the current load of the management station. .

[0114] If a task is successfully assigned to a single machine and completed, the process ends; if it cannot be completed by a single machine or the assignment fails, the task is marked as needing to be split and the process proceeds to the task splitting stage.

[0115] like Figure 2 As shown, specifically, the task allocation module is equipped with a multi-machine execution unit. This unit is used to split a task into multiple sub-tasks when a single drone cannot execute it. An auction mechanism and a hierarchical task allocation strategy are used to allocate these sub-tasks and determine which drone will execute them. The multi-machine execution unit is configured as follows:

[0116] When a single drone cannot perform a task, the task is split into multiple subtasks, so that the subtasks can be executed collaboratively by multiple drones. The minimum number of segments is determined based on the current drone resources. The minimum number of segments is calculated by dividing the weight of the cargo corresponding to the original task by the maximum remaining payload capacity of all available drones and then rounding up.

[0117] In the specific application of this embodiment, when transitioning to the task splitting stage, the system will activate the task splitting mechanism. The original task... Divided into several sub-task sets This allows each subtask to be executed collaboratively by multiple drones. The minimum number of segments is determined based on the current drone resources, and the calculation formula is as follows:

[0118]

[0119] in, This represents the total weight of the original task. This indicates the maximum remaining payload capacity among all currently available drones. This is the round-up operator. The minimum number of slices required to obtain the integer after rounding up.

[0120] Divide the weight of the goods corresponding to the original task by the minimum number of slices to obtain the standard weight of each subtask.

[0121] In this specific application, subtask weight allocation is performed, and the standard weight of each subtask is:

[0122]

[0123] In practice, the weight of the last piece can be slightly adjusted to match actual delivery needs. After splitting, the system will generate a set of subtasks. Each subtask is submitted to the auction process as an independent item to be assigned. Before entering the auction, each subtask must be processed individually. Recalculate its energy consumption Expected execution time Feasibility indicators and final matching degree This is to accurately reflect the scale and resource requirements of the sub-tasks.

[0124] For each subtask generated after splitting, a greedy one-way auction mechanism is used to select the optimal drone for execution, and the subtasks are allocated one by one, including:

[0125] The participating drones submit bids for the sub-tasks, and the mathematical expression for the bid function is:

[0126]

[0127] in, and The recalculated drones Execute subtasks Energy consumption and time estimation; and These are preset, adjustable weighting coefficients.

[0128] In the specific application of this embodiment, for each subtask generated after splitting... A greedy one-way auction mechanism is used to select the optimal drone for execution, allocating sub-tasks one by one. First, the drones participating in the bidding process assign tasks to specific sub-tasks. Submit a quote based on the aforementioned quote function.

[0129] The winning bid rule is for all subtasks. Among the recalculated feasibility conditions, select The drone with the lowest price was chosen as the final sub-task to be executed. The drone.

[0130] In the specific application of this embodiment, the mathematical expression for the winning bid rule is:

[0131]

[0132] After the auction and allocation are completed, the system will update the status of the winning drone, including its remaining energy. Available load capacity and the load of its management station At the same time, the assigned subtasks are removed from the pool of tasks to be assigned.

[0133] If unassigned subtasks remain after allocation through a greedy one-way auction mechanism, a hierarchical task allocation strategy is triggered or a request for cross-regional resource support is made to the central dispatcher. In one application, if a subtask fails to be successfully allocated, or if resource changes occur during execution (such as insufficient drone battery power), the system will re-place these unassigned or interrupted subtasks into the task pool to be allocated, and trigger a hierarchical task allocation mechanism or request cross-regional resource support from the central dispatcher to ensure the overall completion of the task.

[0134] When local resources cannot meet mission requirements, the UAV scheduling system will activate a hierarchical mission allocation and cross-regional coordination mechanism. This mechanism processes allocation requests through a layered diffusion approach, prioritizing resolution within the management station that initiated the mission. If resources are insufficient, the allocation is then diffused to adjacent stations or the central scheduling node, and a balanced utilization of system resources is achieved by introducing station load penalties. The station load level and its penalty coefficient are defined. load The calculation is as follows:

[0135]

[0136] in, Belonging to the site A collection of drones Indicates drone The current load, This is a small constant used to avoid division by zero.

[0137] Based on this load level, the site load penalty is further calculated:

[0138]

[0139] This is the mapping coefficient, used to control the upper limit of the penalty. In cross-site evaluation, site penalties are included in the matching degree calculation. The formula for calculating the corrected matching degree is:

[0140]

[0141] in: The calculated corrected matching degree; This represents the original final match degree; The preset penalty weight coefficient is used to adjust the degree of impact of load balancing on task allocation; For drones The corresponding site load penalty is determined based on the site's load level.

[0142] The execution flow of the hierarchical task allocation strategy is as follows:

[0143] The first level attempts to allocate tasks within the task initiation station using a single-machine allocation or auction mechanism. If the task initiation station lacks resources or allocation fails, the process proceeds to the second level, broadcasting the task to adjacent stations and re-evaluating and allocating it based on a corrected matching degree. If the requirements still cannot be met, the process proceeds to the third level, submitting the task to the central scheduling node for global optimization. In one application, the first level... Internally, the system attempts to allocate tasks via a single-machine allocation or auction mechanism; if resources are insufficient or allocation fails, it proceeds to the second level, broadcasting the task to adjacent sites. And based on the corrected matching degree The task is reassessed and reassigned; if the requirements are still not met, the process proceeds to the third level, where the task is submitted to the central scheduling node for global optimization.

[0144] Hierarchical allocation requires synchronous updates to the drone and site status after each level of processing. If a cross-site scheduling decision is made, the additional cost of cross-site scheduling must be factored into the allocation cost or matching degree.

[0145] In response to the dynamic nature of task urgency changing over time, a dynamic priority mechanism is introduced to calculate the real-time priority score of tasks and preset the reassignment trigger conditions. When the reassignment trigger conditions are met, the affected tasks will be rolled back to the task queue to be assigned. Based on the updated real-time priority, the task-drone edge is reconstructed and the latest matching degree is calculated. Then, through a hierarchical task allocation strategy and auction mechanism, the tasks are rescheduled according to the current drone resource status.

[0146] Formula for calculating task real-time priority score:

[0147]

[0148] in, For the task At the present moment Dynamic priority, , and The preset weighting coefficients, Indicates task Static priority, For the task Latest completion time For numerically stable terms, For characterization tasks The quantified value of economic benefits or service level;

[0149] The redistribution trigger condition is as follows: the task redistribution process is automatically triggered when any of the following conditions occur: the drone's remaining energy is insufficient to support the energy consumption required to complete the task and the reserve energy for safe return; the drone malfunctions or communication is interrupted; a new task arrives, causing a change in the task priority structure, and the dynamic priority difference exceeds a preset threshold; the expected task completion time has exceeded the latest deadline and the probability of default exceeds the allowable upper limit. In a specific application, the above redistribution trigger condition is specifically manifested as: if the drone's remaining energy... Insufficient energy to support the completion of its mission and reserve energy for safe return. That is, satisfying If the drone malfunctions or communication is interrupted; if there is a new mission The arrival of the system causes a change in the system priority structure, and its dynamic priority difference Exceeding the preset threshold Or the expected completion time has exceeded the latest deadline and the probability of default exceeds the allowable limit. Once any of the above conditions are met, the affected task will be rolled back to the task queue, and the system will update its real-time priority. Reconstruct the task-drone edges and calculate the latest matching degree. Then, through hierarchical task allocation and auction mechanisms, tasks are rescheduled based on the current system resource status.

[0150] The path planning module is configured to: parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built overall path cost function or return cost function; define an environmental dynamic complexity index for dynamic hybrid planning, the environmental dynamic complexity index being determined based on dynamic obstacle density, point cloud data, and scene information entropy; and perform dynamic hybrid planning on RRT. and / or VFH Adjustments were made; and local planning and multi-machine collaboration were optimized.

[0151] Reference Figure 3 , Figure 3 This is a structural block diagram of the path planning module provided in an embodiment of this application.

[0152] like Figure 3 As shown, specifically, the path planning module includes an overall cost unit. This unit is used to parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built overall path cost function or return cost function. The overall cost unit is configured as follows:

[0153] Construct the overall path cost function and constrain path planning based on the overall path cost function; the mathematical expression of the overall path cost function is:

[0154]

[0155] in: Energy consumption estimation for the path; For path time estimation; The integral of obstacle potential energy accumulated along the path is used to quantify obstacle avoidance risk. For position The obstacle potential energy function at the location; , and These are preset weighting coefficients for energy consumption, time, and obstacle penalties, respectively.

[0156] In the specific application of this embodiment, for the UAV path planning process, with energy consumption as the priority, and comprehensively considering flight time and obstacle avoidance safety, the above-mentioned overall path cost function is constructed. Its mathematical description in application is: Let the UAV... Along the continuous parameterized path Flight, among which Represents the three-dimensional coordinates of the path points. For arc length parameters, This represents the total path length.

[0157] In the specific application of this embodiment, the energy consumption and time in the overall path cost are modeled in detail. Using a linear approximation method consistent with the aforementioned energy consumption model, the path energy consumption and time are expressed as path integrals or discrete summations.

[0158] Path energy consumption It consists of two parts: flight energy consumption and hovering energy consumption, and its expression is:

[0159]

[0160] In the formula, This represents the baseline energy consumption per unit distance (energy / meter). Additional energy consumption per unit load (energy / (kg·m)) Current cargo weight (kg), the integral term describes the energy consumption during flight. Energy consumption per unit time of hovering (energy / second). This represents the total duration (in seconds) of all loading, unloading, or hovering phases in the path.

[0161] Path time It is then determined by combining the path integral and the hovering time:

[0162]

[0163] in, path point The effective cruising speed (m / s) at the location may be affected by external factors such as wind conditions or airspace speed limits. In practical calculations, the above integral can be numerically approximated by summing discrete samples.

[0164] A semantically weighted obstacle potential energy model is employed, assigning different weights to obstacles based on their dynamic characteristics and semantic categories to accurately characterize their impact on flight safety. For each obstacle detected by the sensors... (in (where is the current set of obstacles), its individual potential energy function is defined as an exponentially decaying form based on Euclidean distance:

[0165]

[0166] in, Indicates obstacles The estimated location, path point Euclidean distance to the obstacle Weighting coefficients are assigned based on the semantic category of obstacles; for example, pedestrians and vehicles are assigned higher weights, trees are assigned medium weights, and buildings are assigned lower weights. An intensity coefficient related to the type of obstacle and its motion state. The potential energy decay scale (meters) is used to control the effective radius of the obstacle's influence.

[0167] The total obstacle potential energy in the environment is obtained by superimposing the individual potential energies of all obstacles:

[0168]

[0169] This semantically weighted potential function enables drones to autonomously adjust their avoidance strategies based on the different types and threat levels of obstacles, optimizing their flight paths while ensuring safety.

[0170] Construct a return cost function and constrain path planning based on the return cost function to determine the optimal return station; the mathematical expression of the return cost function is:

[0171]

[0172] in, and These represent the distance from the drone's current location to the station. Estimated energy consumption and time; The preset site occupancy penalty item is used to reflect the current and short-term berth competition situation; This is a measure of airspace congestion, provided by the central dispatch center and normalized to [specific value]. interval, , , and The preset weighting coefficients are used; the station corresponding to the minimum value of the return cost function is selected as the optimal return station.

[0173] In this specific application, after the drone completes its delivery task, it needs to select the optimal return station, with minimizing the total return cost as the decision criterion. Let the candidate station set be... For any of these stations The return-to-home cost function is as shown above, where the site occupancy penalty term is determined based on the following formula:

[0174]

[0175] in, This represents the number of berths currently occupied at the site. This indicates the number of drones predicted by the central system to return in the near future. Given the total berth capacity of the station. Select the station that minimizes the return cost as the target return point:

[0176]

[0177] After selecting the return station, the system will generate the return path based on the aforementioned hybrid planner. If the berth status of the selected station changes during the return process, the system will trigger the station reselection mechanism and use the rolling window local replanning method to update the flight trajectory in real time, ensuring the dynamic adaptability of the return process.

[0178] like Figure 3 As shown, specifically, the path planning module includes a complexity response unit, which defines the dynamic complexity index of the environment for dynamic hybrid planning; the complexity response unit is configured as follows:

[0179] Define an environmental dynamic complexity index, and dynamically switch between global planning and local reactive control based on this index. The formula for calculating the environmental dynamic complexity index is as follows:

[0180]

[0181] in, This is the calculated environmental dynamic complexity index; This indicates the density of dynamic obstacles detected per unit volume (number of obstacles per cubic meter). The statistical variance of LiDAR point cloud data is used to quantify the uncertainty of environmental perception. The scene information entropy is calculated based on the semantic segmentation results, reflecting the complexity of semantic categories in the environment; , and The preset weighting coefficients for each indicator; based on preset threshold parameters. and Dynamic hybrid planning is achieved by combining environmental dynamic complexity indicators:

[0182] when At that time, RRT was mainly used. Perform global path planning; when At that time, with VFH As the main force, local real-time obstacle avoidance planning is carried out; when When this happens, enter hybrid mode and use RRT in a weighted coordination manner based on the current complexity. and VFH .

[0183] In the specific application of this embodiment, when At that time, the environment was relatively simple and stable, and RRT was mainly used. Perform global path planning; when At that time, the environment was highly dynamic and complex, so VFH was adopted. As the main force, local real-time obstacle avoidance planning is carried out; when When the system enters a hybrid mode, it uses two planning methods in a weighted and coordinated manner based on the current complexity. This mechanism can adaptively select the most suitable planning strategy according to environmental characteristics, taking into account both global optimality and local safety of the path.

[0184] like Figure 3 As shown, specifically, the path planning module is equipped with a local planning optimization unit, which is used to optimize local planning and multi-machine collaboration; the local planning optimization unit is configured as follows:

[0185] A semantically enhanced vector field histogram method is used for real-time heading selection. For each candidate heading, a comprehensive evaluation function is constructed, and the mathematical expression of the comprehensive evaluation function is as follows:

[0186]

[0187] in, For the calculated heading A comprehensive evaluation; The direction of the current global path or local target; This indicates the cost of speed adjustment caused by a change in course; , and These are preset weighting coefficients; For heading The weighted obstacle quantity, which is based on the obstacle's heading. The perceived intensity and preset semantic weights are determined. A small constant is used for numerical stability; finally, the direction with the smallest overall evaluation is selected as the optimal heading.

[0188] In this specific application, a semantically enhanced vector field histogram method is used for real-time heading selection, dividing the surrounding 360° space into several azimuth intervals. And calculate the weighted obstacle amount for each direction:

[0189]

[0190] in, Indicates obstacles In direction The perceived intensity on the surface is usually defined as ,in For the obstacle in this direction distance, Maximum sensing range; The semantic weights of obstacles reflect the degree to which their category affects flight safety. For each candidate heading... Construct the above comprehensive evaluation function, and finally select the direction with the minimum comprehensive evaluation as the optimal heading:

[0191]

[0192] If the path is blocked by an obstacle, local replanning is immediately triggered. This local replanning obtains a sequence of path points by discretely sampling the globally optimal path and determines the local sharpness of each path point. When a continuous path interval is detected, and all path points in this interval satisfy the condition that the local sharpness is less than a safety threshold and the obstacle's duration is greater than a persistence threshold, the path interval is determined to be permanently blocked, and local RRT is initiated. The replanning process; the replanning area is limited to a scrolling window determined by a preset number of window expansion arcs, in order to ensure that the new planned path can be smoothly integrated into the original global path;

[0193] In the specific application of this embodiment, for triggering local replanning, the globally optimal path is calculated. Discrete sampling is performed to obtain a path point sequence. And calculate the local sharpness of each path point:

[0194]

[0195] in, Representing path points To the nearest obstacle The Euclidean distance. When a continuous path interval is detected. All path points satisfy:

[0196]

[0197] and

[0198]

[0199] That is, the local sharpness is consistently below the safe threshold. And the duration of the obstacle Exceeding the set continuous threshold The system will determine that the path segment is "permanently blocked" and then initiate local RRT. The replanning process. The replanning area is limited to the included interval. Within the scrolling window, The number of arc segments in the window is increased to ensure that newly planned paths can be smoothly integrated into the existing global path. This mechanism effectively improves the path robustness and adaptability of UAVs in dynamic environments.

[0200] In multi-drone collaborative missions, the speed obstacle method and collision risk measurement are used to assess and avoid potential conflicts. To address potential path conflicts and communication overhead among multiple drones, a lightweight dynamic priority negotiation mechanism based on game theory is adopted. This mechanism determines the initial priority weight of each drone based on its static mission priority and remaining energy. When multiple drones collide in a local airspace, a round of local negotiation is initiated, and the priority weight of each drone is updated using an exponential recalibration method.

[0201] In the specific application of this embodiment, the speed obstacle method and collision risk measurement are used to assess and avoid potential conflicts for any two drones. and Define its prediction window The set of speed obstacles inside is:

[0202]

[0203] in, , The current position vector, , The current velocity vector, The critical safety interval is the sum of the safety radii of the two drones. (Drone) The set of feasible velocities is the kinematically permissible velocity. The difference between the set of all speed barriers and the set of all speed barriers:

[0204]

[0205] Choose the one closest to its preferred speed. The velocity vector is used as the optimal solution:

[0206]

[0207] To quantify collision risk, the estimated collision time between each pair of drones is calculated:

[0208]

[0209] in, A unit relative position vector, To ensure numerical stability and a small value, the semantic weights of the opposing drone are considered. The collision risk level is defined as:

[0210]

[0211] When the estimated collision time Less than the critical value or collision risk level Exceeding the risk threshold When this happens, the system marks the object as a high-risk target and prioritizes its avoidance.

[0212] Simultaneously, a lightweight dynamic priority negotiation mechanism based on game theory is employed to resolve potential path conflicts among multiple drones and reduce communication overhead. The initial priority weight of each drone is determined by its static task priority and remaining energy, calculated as follows:

[0213]

[0214] in, For drones The static priority score of the task being performed. For its current remaining energy, , The normalization coefficients ensure that the sum of all weights is 1. When multiple UAVs collide in a local airspace, the system initiates a round of local negotiation, updating the priority weights of each UAV using an exponential recalibration method.

[0215]

[0216] in, Indicates drone The projected generation in the current conflict includes additional energy consumption, etc. For negotiation sensitivity parameters, A group of nearby drones participating in the negotiation. The results obtained after the negotiation. This will be used to determine the avoidance responsibility of each drone, typically with a concession margin of... Proportional. Ultimately, each drone, considering its own updated priority weights and the expected behavior of other drones, solves a constrained local velocity selection problem, thereby achieving distributed cooperative obstacle avoidance with low communication costs while ensuring overall mission efficiency.

[0217] If temporary airspace congestion is encountered during the mission, a temporary no-fly zone is set up using an ellipsoidal geometric model, thereby triggering local replanning. When local replanning is triggered, it responds to obstacle avoidance constraints and triggers a local rolling replanning mechanism. The ellipsoidal constraints corresponding to the temporary no-fly zone are included as feasible region determination conditions in the planner's solution process.

[0218] In the specific application of this embodiment, if temporary airspace congestion is encountered during the mission, an ellipsoidal geometric model is used to establish a temporary no-fly zone. For a set of points in space that satisfies the following inequalities:

[0219]

[0220] in, Let be the coordinate vector of the center point of the ellipsoid. It is a symmetric positive definite matrix used to describe the axial lengths and directional characteristics of an ellipsoid. This is used when the principal axes of the ellipsoid are aligned with the coordinate axes and the lengths of the three axes are... , , hour, .

[0221] Path planning must meet strict obstacle avoidance constraints: for parametric paths Every point on It must meet the following conditions:

[0222]

[0223] in, This is a preset safety buffer margin. If the original planned route is detected to be outside the no-fly zone... If there is an intersection or the above inequality constraint is violated, the system will immediately trigger a local rolling replanning mechanism and incorporate the ellipsoid constraint as a feasible region determination condition into the planner's solution process, thereby generating an update path that meets safety requirements.

[0224] During the multi-drone collaborative return process, each drone adopts a distributed consensus iteration strategy for flight altitude and speed scalars.

[0225] In the specific application of this embodiment, for the multi-drone coordinated return process, each drone adjusts its flight altitude accordingly. With velocity scalar Using a distributed consensus iteration strategy, the update formula is:

[0226]

[0227]

[0228] in, The number of iterations. Indicates drone The collection of other drones within the communication neighborhood, and To control the convergence speed, a small positive step size is used. By repeatedly exchanging state information, the UAV swarm gradually converges to a consistent altitude and speed configuration, maintaining reasonable airspace layering and speed coordination.

[0229] When local communication conditions are limited or neighborhood information is insufficient to achieve effective coordination, the central dispatcher will monitor the situation. The central node will only intervene and make the final arbitration when a potential conflict is detected that cannot be resolved through distributed negotiation.

[0230] The fault response module is configured to: perform fault analysis on the UAV using a weighted voting-based fault determination mechanism; trigger a preset fault response strategy based on the fault analysis results; the fault response strategy includes at least a controlled retreat and landing process; determine the landing point based on a preset comprehensive cost function; and trigger the selection of a UAV to perform the support mission.

[0231] Reference Figure 4 , Figure 4 This is a structural block diagram of the fault response module provided in an embodiment of this application.

[0232] like Figure 4 As shown, specifically, the fault response system includes a fault analysis unit, which performs fault analysis on the UAV using a weighted voting-based fault determination mechanism. The fault analysis unit is configured as follows:

[0233] Determine the value of the constructed fault confirmation indicator within a continuous time window; when the fault confirmation indicator is... If the fault is confirmed, continue normal operation and maintain monitoring; if the fault confirmation indicator is... If the fault occurs, the system will immediately switch to the local automatic recovery process and perform the corresponding fault handling and system recovery operations; the mathematical expression for the fault confirmation indicator is:

[0234]

[0235] in: The weighted voting value; The preset weighted vote threshold; This is the persistence time threshold; The specific logic for determining the obtained fault confirmation index is as follows: when the weighted voting value continuously exceeds the weighted vote threshold, a fault is confirmed. The determination of continuity is based on the persistence time threshold.

[0236] In this specific application of the embodiment, in a multi-sensor system, to accurately identify anomalies in key state variables, the UAV employs a fault determination mechanism based on weighted voting. A sensor set is provided for key state variables such as IMU angular velocity, acceleration, GPS position, altitude, and power current. Each sensor At any moment The measured value is Its expected value or fusion estimate is Measurement uncertainty is determined by the standard deviation. express.

[0237] First, calculate the normalized residuals of each sensor:

[0238]

[0239] in, For numerical stability, then, based on a specific threshold of the sensor... Perform anomaly detection:

[0240]

[0241] The indicator function takes a value of 1 when the residual exceeds a threshold, and 0 otherwise. To synthesize the judgment results from all sensors, a weighted voting mechanism is introduced. Let the weight of each sensor be... And satisfy The weighted voting value is:

[0242]

[0243] Using the mathematical expression of the aforementioned fault confirmation indicator, when the weighted vote count continuously exceeds the threshold... At that time, the system confirmed the fault.

[0244] Continuously monitor fault confirmation indicators If its value is 0, the system will continue to operate normally and remain monitored; if its value is 1, the system will immediately switch to the local automatic recovery process and perform the corresponding fault handling and system recovery operations.

[0245] Once a fault is confirmed, a limited number of local automatic recovery cycles should be initiated. This applies to critical subsystems or subsets of sensors. Let its initial residual vector be The residual state update after each recovery operation can be modeled as follows:

[0246]

[0247] in, It is a diagonal matrix, each This represents the attenuation factor of the corresponding sensor residual in this recovery. This includes noise or recalibration errors that may still exist after recovery. Recovery is considered successful when all critical residuals fall below the recovery threshold.

[0248]

[0249] Number of attempts Reaching the preset limit If the above conditions are still not met, local recovery is deemed to have failed, and the system will immediately switch to the controlled backoff and landing process. If recovery is successful, the fault status is cleared and the latest status information is synchronized to the task management terminal, and then normal task execution resumes. Simultaneously, sensor weights can be dynamically updated based on the recovery effect. To improve the adaptability of subsequent monitoring.

[0250] like Figure 4 As shown, specifically, the fault response setup includes a landing point analysis unit, which is used to determine the landing point based on a preset comprehensive cost function; the landing point analysis is configured as follows:

[0251] If local automatic recovery fails, the drone will initiate a controlled retreat and landing process, selecting the optimal landing point from the set of nearest safe landing points. For each candidate landing point, it will be evaluated using a preset comprehensive cost function, and the optimal landing point will be selected based on the minimum comprehensive cost, provided that energy constraints are met. The mathematical expression for the comprehensive cost function is as follows:

[0252]

[0253] in: This is the current location of the drone; This indicates the distance from the current location to the landing point or the flight cost; Risk assessment for terrain slope and flatness; Scoring the density and hazard level of surrounding obstacles; This reflects the likelihood or availability penalty of the landing site currently being occupied; , , and These are the preset corresponding weighting coefficients.

[0254] In the specific application of this embodiment, after local automatic recovery fails, the drone will initiate a controlled retreat procedure and move from the nearest safe landing point. Choose the optimal landing point from the pool. For each candidate landing point... Based on the comprehensive cost function described above, and under the premise of satisfying energy constraints, the system selects the landing point with the minimum comprehensive cost as the target:

[0255]

[0256] in, To estimate the energy consumption to reach the landing site, The current remaining energy, To ensure a minimum energy reserve for safety. If no candidate landing site meets the energy constraints, the system will initiate an emergency retreat strategy, performing operations such as load shedding or deceleration landing based on the platform's capabilities, and reselecting a landing site based on the minimum flight distance as the primary criterion.

[0257] Determine the target landing point Afterwards, the drone will execute a controlled retreat trajectory, call the aforementioned hybrid planner to generate a return path, and enter a state of waiting for ground support after landing. At the same time, the entire event will be reported to the mission management terminal for recording and subsequent processing.

[0258] like Figure 4 As shown, specifically, the fault response system includes a rescue drone determination unit, which is used to select the most suitable drone for the support mission; the rescue drone determination unit is configured as follows:

[0259] When a drone malfunctions and reports it, a fault report package is generated based on the fault information. Based on the fault report package, the support drone selection process is initiated. The support drone selection process selects the most suitable drone to perform the support mission from the candidate drone set through linear weighted scoring. The linear weighted scoring is normalized for six dimensions: energy, payload, distance, time, reliability, and scheduling status, to obtain a comprehensive score for the candidate drones. After verifying that the drone meets the capability constraints, the drone with the maximum comprehensive score is selected as the rescue drone.

[0260] The mathematical expression for the overall score is:

[0261]

[0262] in: For the corresponding preset weight coefficients, satisfying These correspond to six dimensions: energy, load capacity, distance, time, reliability, and scheduling status. These are the scores for each dimension after normalization.

[0263] When a drone malfunctions and reports it, the task management terminal will analyze the fault report package. Initiate the drone selection process. This process selects from a set of candidate drones. In this study, the most suitable drone for performing support missions is selected through linear weighted scoring.

[0264] For each candidate drone Calculate the following six normalized features:

[0265]

[0266]

[0267]

[0268]

[0269]

[0270]

[0271] in, and These are the remaining energy and rated battery capacity of the candidate unit, respectively. and The remaining and rated load capacity, This represents the distance from the candidate machine to the faulty machine. For distance scale parameters, For the estimated arrival time, Score historical reliability. This indicates that the aircraft is in a schedulable state. The overall score of the candidate drones is a weighted sum of each feature, calculated based on the aforementioned mathematical expression for the overall score. The drone with the highest score is selected as the support aircraft.

[0272]

[0273] Before making a final selection, it is necessary to verify whether it meets the capability constraints:

[0274]

[0275] in, This refers to the weight of the goods that need to be transferred. and These are the estimated energy consumptions for traveling to the fault location and for transferring materials. Set a safe energy reserve threshold. Select a suitable support unit. Subsequently, the task management terminal will issue a support command to the machine, which will then plan the shortest or lowest energy consumption path to the fault point to perform material transfer. If no candidate machine meets the constraints, the system will trigger a wider-area scheduling mechanism or notify manual intervention.

[0276] During the transfer of supplies, strict constraints on both load capacity and energy must be met. This applies when supporting drones. Arrival at the faulty machine After the location is confirmed and the material transfer operation is completed, its status will be updated as follows:

[0277]

[0278] After the materials are transferred, the system will proceed with the following steps based on the actual status of the faulty machine: if the faulty machine can resume autonomous operation through material replenishment or status reset, it can continue to execute the original tasks; if the faulty machine needs to be recovered and repaired, the task management terminal will reallocate its remaining unfinished tasks and re-enter the task scheduling process.

[0279] Based on the above, the drone collaborative delivery system for small and medium-sized urban parcels in this embodiment generates standardized task data packages through the task release module, providing unified data support for task allocation and path planning, and laying the foundation for efficient collaborative operation. The task allocation module prioritizes single-machine execution, and after complex tasks are broken down, an auction mechanism and hierarchical strategy are used for allocation, which optimizes drone resource utilization, adapts to different task complexities, and ensures task executability. The path planning module uses a cost function to constrain path optimization, and combines dynamic obstacle density, point cloud data, and scene information entropy to construct an environmental dynamic complexity index, which is used to optimize the RRT (Real-Time Tolerance). / VFH Dynamic hybrid planning is employed, simultaneously optimizing local planning and multi-drone collaboration to significantly improve path optimality and adaptability to dynamic environments. The fault response module accurately identifies faults through weighted voting, relying on controlled backoff landing, comprehensive cost function-based site selection, and support for drone scheduling, balancing flight safety and mission continuity. In summary, the system achieves a synergistic improvement in delivery efficiency, environmental adaptability, and operational reliability, perfectly meeting the complex needs of urban small-to-medium-sized parcel delivery scenarios.

[0280] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0281] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A drone-based collaborative delivery system for small and medium-sized urban parcels, characterized in that, The system includes: The task publishing module is configured to generate standardized task data packets based on express delivery information. These standardized task data packets are used for task allocation and path planning during drone delivery. The task allocation module is configured to: parse standardized task data packets, analyze the matching degree between candidate drones and tasks, and obtain the final matching degree; determine the drones to execute tasks based on the final matching degree, and prioritize single-drone execution of tasks. When a single drone cannot execute a task, the task is split into multiple sub-tasks, and the sub-tasks are allocated using an auction mechanism and a hierarchical task allocation strategy to determine the drones to execute the sub-tasks. The path planning module is configured to: parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built overall path cost function or return cost function; define an environmental dynamic complexity index for dynamic hybrid planning, the environmental dynamic complexity index being determined based on dynamic obstacle density, point cloud data, and scene information entropy; and perform dynamic hybrid planning on RRT. and / or VFH Adjustments were made; and local planning and multi-machine collaboration were optimized. The fault response module is configured to: perform fault analysis on the UAV using a weighted voting-based fault determination mechanism; trigger a preset fault response strategy based on the fault analysis results; the fault response strategy includes at least a controlled retreat and landing process; determine the landing point based on a preset comprehensive cost function; and trigger the selection of a UAV to perform the support mission. The task allocation module includes a task parsing unit, which is used to parse standardized task data packets, analyze the matching degree between candidate drones and tasks, and obtain the final matching degree. The task parsing unit is configured as follows: Construct a matching graph of task-drone-product point, and calculate the final matching degree for the edge of task-drone. The formula for calculating the final matching degree is: in: For the calculated task With drones The final match degree between them; For the task With drones Feasibility indicators between, the feasibility indicators The value is or It is defined as when the task The weight of the cargo shall not exceed that of the drone. The current available payload and drone Complete the task When the sum of the required energy and the reserved energy for return does not exceed its current remaining energy, and the estimated mission completion time is no later than the difference between the latest mission completion time and the current time. , indicating task For drones Feasible, otherwise , indicating task For drones Not feasible; For the first The normalized score components, The preset weight coefficients for the corresponding rating components satisfy... ,and , Corresponding to time, energy, load capacity, distance, task priority, and management station load; The task allocation module includes a single-machine execution unit, which is used to execute tasks for a single UAV based on the final matching degree; the single-machine execution unit is configured as follows: When it exists At that time, it was determined that there was a machine capable of completing the task independently. drones A local site priority strategy is adopted to prioritize tasks. The selection of drones is based on proximity and the specific mission requirements. All In the task-drone selection process, the drone output corresponding to the maximum edge value is used as the final task execution value. A single drone; The task allocation module is equipped with a multi-machine execution unit. This multi-machine execution unit is used to split a task into multiple sub-tasks when a single drone cannot execute the task. It then uses an auction mechanism and a hierarchical task allocation strategy to allocate the sub-tasks and determine which drone will execute them. The multi-machine execution unit is configured as follows: When a single drone cannot execute a task, the task is split into multiple subtasks, allowing multiple drones to collaboratively execute the subtasks. The minimum number of fragments is determined based on current drone resources. This minimum number of fragments is calculated by dividing the original task's cargo weight by the maximum remaining payload capacity of all available drones, then rounding up. The standard weight of each subtask is obtained by dividing the original task's cargo weight by the minimum number of fragments. For each subtask generated after splitting, a greedy one-way auction mechanism is used to select the optimal drone for execution, and the subtasks are allocated item by item, including: The participating drones submit bids for the sub-tasks, and the mathematical expression for the bid function is: in, and The recalculated drones Execute subtasks Energy consumption and time estimation; and These are preset adjustable weighting coefficients; The winning bid rule is for all subtasks. Among the recalculated feasibility conditions, select The drone with the lowest price was chosen as the final sub-task to be executed. drones; When there are still unassigned subtasks after allocation through the greedy single-item auction mechanism, a hierarchical task allocation strategy is triggered or a request for cross-regional resource support is made to the central dispatch. The execution flow of the hierarchical task allocation strategy is as follows: The first level attempts to allocate tasks within the task initiation station using either a single-machine allocation or auction mechanism. If the task initiation station lacks sufficient resources or allocation fails, the process proceeds to the second level, broadcasting the task to adjacent stations and re-evaluating and allocating it based on a corrected matching degree. If the requirements still cannot be met, the process proceeds to the third level, submitting the task to the central scheduling node for global optimization. The formula for calculating the corrected matching degree is: in: The calculated corrected matching degree; This represents the original final match degree; The preset penalty weight coefficient is used to adjust the degree of impact of load balancing on task allocation; For drones The corresponding site load penalty item is determined based on the site's load level, which is calculated using the following formula: in, Belonging to the site A collection of drones Indicates drone The current load, For small constants used to avoid division by zero, For drones The current available load capacity; In response to the dynamic characteristics of task urgency changing over time, a dynamic priority mechanism is introduced to calculate the real-time priority score of the task and a preset reassignment trigger condition is set. When the reassignment trigger condition is met, the affected task will be returned to the queue of tasks to be assigned. Based on the updated real-time priority, the task-UAV edge is reconstructed and the latest matching degree is calculated. Then, through a hierarchical task allocation strategy and auction mechanism, the task is rescheduled according to the current UAV resource status. The formula for calculating the real-time priority score of the task is as follows: in, For the task At the present moment Dynamic priority, , and The preset weighting coefficients, Indicates task Static priority, For the task Latest completion time For numerically stable terms, For characterization tasks The quantified value of economic benefits or service level; The redistribution triggering conditions are as follows: the task redistribution process is automatically triggered when any of the following situations occur: if the remaining energy of the UAV is insufficient to support the energy consumption required to complete the task and the energy reserve for safe return; if the UAV malfunctions or communication is interrupted; if a new task arrives and causes a change in the task priority structure, and its dynamic priority difference exceeds a preset threshold; or if the expected task completion time has exceeded the latest deadline and the probability of default exceeds the allowed upper limit.

2. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 1, characterized in that, The path planning module includes an overall cost unit, which is used to parse standardized task data packets, analyze the paths corresponding to tasks or subtasks, and constrain path planning based on a pre-built overall path cost function or return cost function. The overall cost unit is configured as follows: Construct a total path cost function and constrain path planning based on the total path cost function; wherein, the mathematical expression of the total path cost function is: in: Energy consumption estimation for the path; For path time estimation; The integral of obstacle potential energy accumulated along the path is used to quantify obstacle avoidance risk. For position The obstacle potential energy function at the location; , and These are preset weighting coefficients for energy consumption, time, and obstacle penalties, respectively. A return-to-origin cost function is constructed, and path planning is constrained based on the return-to-origin cost function to determine the optimal return-to-origin station; wherein, the mathematical expression of the return-to-origin cost function is: in, and These represent the distance from the drone's current location to the station. Estimated energy consumption and time; The preset site occupancy penalty item is used to reflect the current and short-term berth competition situation; This is a measure of airspace congestion, provided by the central dispatch center and normalized to [specific value]. interval, , , and The preset weighting coefficients are used; the station corresponding to the minimum value of the return cost function is selected as the optimal return station.

3. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 2, characterized in that, The path planning module includes a complexity response unit, which defines a dynamic environmental complexity index for dynamic hybrid planning. The complexity response unit is configured as follows: Define the environmental dynamic complexity index, and dynamically switch between global planning and local reactive control based on the environmental dynamic complexity index. The formula for calculating the environmental dynamic complexity index is: in, This is the calculated environmental dynamic complexity index; This represents the density of dynamic obstacles detected per unit volume; The statistical variance of LiDAR point cloud data is used to quantify the uncertainty of environmental perception. The scene information entropy is calculated based on the semantic segmentation results, reflecting the complexity of semantic categories in the environment; , and The preset weighting coefficients for each indicator; based on preset threshold parameters. and Dynamic hybrid planning is achieved by combining environmental dynamic complexity indicators: when At that time, RRT was mainly used. Perform global path planning; when At that time, with VFH As the main force, local real-time obstacle avoidance planning is carried out; when When this happens, enter hybrid mode and use RRT in a weighted coordination manner based on the current complexity. and VFH .

4. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 3, characterized in that, The path planning module includes a local planning optimization unit, which is used to optimize local planning and multi-machine collaboration; the local planning optimization unit is configured as follows: A semantically enhanced vector field histogram method is used for real-time heading selection. For each candidate heading, a comprehensive evaluation function is constructed, and the mathematical expression of the comprehensive evaluation function is as follows: in, For the calculated heading A comprehensive evaluation; The direction of the current global path or local target; This indicates the cost of speed adjustment caused by a change in course; , and These are preset weighting coefficients; For heading The weighted obstacle quantity, which is based on the obstacle's heading. The perceived intensity and preset semantic weights are determined. The direction with the smallest overall evaluation value is selected as the optimal heading; This is a small constant used for numerical stability. If the path is blocked by an obstacle, local replanning is immediately triggered. This local replanning obtains a sequence of path points by discretely sampling the globally optimal path and determines the local sharpness of each path point. When a continuous path interval is detected, and all path points in this interval satisfy the condition that the local sharpness is less than a safety threshold and the obstacle's duration is greater than a persistence threshold, the path interval is determined to be permanently blocked, and local RRT is initiated. The replanning process; the replanning area is limited to a scrolling window determined by a preset number of window expansion arcs, in order to ensure that the new planned path can be smoothly integrated into the original global path; In multi-drone collaborative tasks, the speed obstacle method and collision risk measurement are used to assess and avoid potential conflicts. To address potential path conflicts and communication overhead among multiple drones, a lightweight dynamic priority negotiation mechanism based on game theory is adopted. This mechanism determines the initial priority weight of each drone based on its static task priority and remaining energy. When multiple drones collide in a local airspace, a round of local negotiation is initiated, and the priority weight of each drone is updated using an exponential recalibration method. If a temporary airspace congestion problem is encountered during the mission, an ellipsoidal geometric model is used to set a temporary no-fly zone, thereby triggering local replanning. When local replanning is triggered, it responds to obstacle avoidance constraints, triggers a local rolling replanning mechanism, and incorporates the ellipsoidal constraints corresponding to the temporary no-fly zone as feasible region determination conditions into the planner's solution process. During the multi-drone collaborative return process, each drone adopts a distributed consensus iteration strategy for flight altitude and speed scalars; When local communication conditions are limited or neighborhood information is insufficient to achieve effective coordination, the central dispatcher will monitor the situation. The central node will only intervene and make the final arbitration when a potential conflict is detected that cannot be resolved through distributed negotiation.

5. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 1, characterized in that, The fault response includes a fault analysis unit, which performs fault analysis on the UAV using a weighted voting-based fault determination mechanism; the fault analysis unit is configured as follows: The value of the constructed fault confirmation indicator is determined within a continuous time window; when the fault confirmation indicator is... If the fault is confirmed, continue normal operation and maintain monitoring; if the fault confirmation indicator is... If the fault is detected, the system will immediately switch to the local automatic recovery process to perform the corresponding fault handling and system recovery operations; wherein, the mathematical expression of the fault confirmation index is: in: The weighted voting value; The preset weighted vote threshold; This is the persistence time threshold; The specific logic for determining the obtained fault confirmation index is as follows: when the weighted voting value continuously exceeds the weighted vote threshold, a fault is confirmed. The determination of continuity is based on the persistence time threshold.

6. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 5, characterized in that, The fault response setup includes a landing point analysis unit, which is used to determine the landing point based on a preset comprehensive cost function; the landing point analysis is configured as follows: If local automatic recovery fails, the UAV will initiate the controlled retreat and landing process, selecting the optimal landing point from the set of nearest safe landing points. For each candidate landing point, an evaluation is performed using a preset comprehensive cost function. Under the premise of satisfying energy constraints, the landing point with the minimum comprehensive cost is selected as the optimal landing point. The mathematical expression of the comprehensive cost function is: in: This is the current location of the drone; This indicates the distance from the current location to the landing point or the flight cost; Risk assessment for terrain slope and flatness; Scoring the density and hazard level of surrounding obstacles; This reflects the likelihood or availability penalty of the landing site currently being occupied; , , and These are the preset corresponding weighting coefficients.

7. The drone-based collaborative delivery system for small and medium-sized urban parcels according to claim 6, characterized in that, The fault response system includes a rescue drone determination unit, which is used to select the most suitable drone for performing the support mission; the rescue drone determination unit is configured as follows: When a drone malfunctions and reports it, a fault report package is generated based on the fault information, and the drone selection process is initiated based on the fault report package. The support drone selection process selects the most suitable drone for performing the support mission from the candidate drone set through linear weighted scoring; The linear weighted scoring is normalized for six dimensions: energy, payload, distance, time, reliability, and scheduling status, to obtain a comprehensive score for the candidate UAVs. After verifying that the UAVs meet the capability constraints, the UAV with the maximum comprehensive score is selected as the rescue UAV. The mathematical expression for the comprehensive score is: in: For the corresponding preset weight coefficients, satisfying These correspond to six dimensions: energy, load capacity, distance, time, reliability, and scheduling status. These are the scores for each dimension after normalization.

Images