A method and system for air-ground collaborative logistics delivery task allocation based on grid state mapping
By acquiring the real-time status of the parcel lockers through the edge computing module, constructing a locker status matrix for pre-allocation and locking, and performing step-by-step redirection and multi-dimensional arbitration during drone flight, the problem of locker status not being included in decision-making in existing technologies is solved, thereby improving the success rate and efficiency of logistics delivery.
Patent Information
- Application Number
- CN202611250042.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-17
AI Technical Summary
Existing air-ground collaborative logistics delivery technologies fail to effectively incorporate the real-time status of express delivery lockers as a decision variable, leading to drone delivery failures and resource waste, and lacking an effective resource scheduling mechanism when multiple drones are competing.
The edge computing module obtains the real-time status of the smart parcel lockers, constructs a locker status matrix, pre-allocates and locks the lockers, maintains the locks through heartbeat communication during flight, triggers escalating redirection operations, and performs resource arbitration based on multiple dimensions.
It significantly improved the delivery success rate and overall efficiency of parcel delivery, reduced unnecessary flights, and achieved balanced utilization of resources.
Smart Images

Figure CN123117973A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of intelligent logistics and low-altitude technology, and in particular to a method and system for allocating air-ground collaborative logistics delivery tasks based on grid status mapping. Background Technology
[0002] In existing air-ground collaborative logistics delivery technologies, the collaboration between drones and unmanned vehicles (UAVs) is considered a crucial solution to address the last-mile logistics efficiency bottleneck. Related research primarily focuses on vehicle-machine two-level path collaborative planning and scheduling decisions for heterogeneous unmanned systems. Examples include using UAVs as mobile supply and communication nodes and UAVs as highly mobile execution agents for path planning, or deploying UAVs at transfer points selected from existing logistics sites. These technologies generally treat last-mile delivery points, including customer locations, transfer stations, or smart parcel lockers, as static coordinate points, focusing solely on path optimization while neglecting the dynamic attributes of these last-mile nodes.
[0003] However, the aforementioned existing technologies share common shortcomings. In real-world logistics scenarios, smart parcel lockers exhibit dynamic characteristics such as limited locker capacity, real-time changes in locker status, and competition for locker space on connecting platforms. When a large number of parcels arrive at a station with unmanned vehicles, and multiple drones simultaneously launch deliveries to multiple lockers, the locker's locker status, including the available number of lockers of each size, is a core constraint determining the success of the task. Existing technologies completely fail to incorporate this dynamic constraint into their decision-making. Furthermore, existing technologies employ an open-loop "plan first, execute later" model, lacking a mechanism for real-time status synchronization between drones and end-point nodes during drone flight. If a locker becomes occupied or experiences mechanical failure during drone flight, existing technologies cannot respond promptly, leading to delivery failures upon drone arrival, resulting in wasted flight time and battery life. Simultaneously, existing technologies do not address conflict arbitration issues when multiple drones compete for the same limited locker space, lacking an effective resource scheduling mechanism in large-scale concurrent delivery scenarios.
[0004] Therefore, there is an urgent need for an air-ground collaborative logistics delivery task allocation method that can incorporate the real-time status of the parcel lockers as a decision variable into task allocation, dynamically respond to changes in the locker status during drone flight, and conduct reasonable arbitration when multiple drones compete for limited lockers. Summary of the Invention
[0005] Firstly, this application provides a method for allocating air-to-ground collaborative logistics delivery tasks based on grid status mapping, comprising the following steps: When the unmanned vehicle arrives at the preset station, the edge computing module deployed on the unmanned vehicle obtains the status data of the smart express cabinets within the service range of the preset station. Based on the obtained status data, a status matrix is constructed. Each entry in the status matrix contains at least: the express cabinet identifier and the current available quantity of each type of cabinet. The edge computing module performs pre-allocation of lockers for one or more delivery packages based on the locker status matrix: for each package, it determines whether the available lockers of the target locker meet the package’s locker requirements. If they do, the corresponding locker is pre-locked; otherwise, a neighbor locker transfer is initiated, and a backup locker that meets the requirements is found within the service area and re-mapped. After all packages have been processed, a pre-allocation scheme containing the correspondence between package and locked locker is formed. According to the pre-allocation plan, a two-way communication link is established with the target express cabinet before the drone takes off, and periodic communication is carried out through the two-way communication link after the drone takes off in order to maintain the lock of the compartment in the pre-allocation plan. During flight, when the target parcel locker detects a physical failure of a locker that has been locked in the pre-allocation plan, or when the edge computing module detects a loss of communication link between the drone and the target parcel locker, a tiered redirection operation is triggered, and the pre-allocation plan is updated. When the edge computing module detects that multiple drones are simultaneously requesting the remaining slots of the same parcel locker and there are not enough available slots, the edge computing module prioritizes the requesting drones based on multiple dimensions. Drones with higher priority will be allocated slots first, while drones with lower priority will receive alternative allocation schemes.
[0006] In some embodiments, each entry in the grid state matrix further includes: the length of the drone queue waiting for delivery and the status of whether the docking platform is available; the step of performing grid pre-allocation one by one includes: First, group the task packages according to the target express lockers, and then sort the task packages in each group from largest to smallest according to the total number of lockers required. For each task package, query the available quantity of the corresponding compartment type of the target express locker in the compartment status matrix. If the condition is met, lock the compartment and deduct the corresponding available quantity from the compartment status matrix. If the condition is not met, record it as a task package to be transferred. Once all lockouts are completed, the packages to be transferred are sorted in ascending order of distance from the nearest backup locker, and then transferred to adjacent lockers in sequence.
[0007] In some embodiments, the escalating redirection operation includes: The first level involves the target parcel locker searching for and replacing a similar available compartment within the locker itself, with a response latency of less than 100 milliseconds. The second level involves the edge computing module searching within the same site's service area for a backup parcel locker that meets the locker requirements and is closest to the drone's current location, and verifying that the new flight path is within the drone's remaining range. The third level involves the edge computing module querying the status of parcel locker compartments within the service range of adjacent stations, and then transferring the task package to unmanned vehicles at adjacent stations via vehicle-to-vehicle communication for relay delivery.
[0008] In some embodiments, the detection method for physical failure of the grid includes: using a triple redundant sensor array consisting of a photoelectric switch, a micro switch, and a motor current sensor deployed in each grid to detect in real time whether the grid has mechanical failure, is occupied by foreign objects, or has sensor malfunction.
[0009] In some embodiments, the method for determining communication link failure includes: the drone sending a heartbeat packet at a fixed frequency, the heartbeat packet containing the drone's current location, remaining battery power, and estimated arrival time; the target parcel locker refreshing the remaining lock time when it receives the heartbeat packet, and determining communication failure and releasing the locker lock when it does not receive a heartbeat packet for two consecutive heartbeat sending cycles; the heartbeat packet is sent once every 5 seconds, and the locker lock refresh time is 30 seconds.
[0010] In some embodiments, multiple dimensions include remaining battery status, parcel delivery urgency, and availability of alternative parcel lockers; priority ranking adopts a weighted scoring method, which assigns preset weights to the three dimensions of remaining battery status, parcel delivery urgency, and availability of alternative parcel lockers and then calculates a comprehensive priority score. The lower the remaining battery, the closer the parcel's delivery deadline is, and the farther away the alternative parcel locker is, the higher the priority.
[0011] In some embodiments, data transmission is performed between the unmanned vehicle, the smart parcel locker, and the drone via a C-V2X PC5 direct-connect communication interface.
[0012] Secondly, this application provides an air-to-ground collaborative logistics delivery task allocation system based on grid status mapping, used to execute the method in any of the above embodiments, including: The cloud task module is configured to receive package order information, group packages by parcel locker and associate them with preset sites; The edge computing module, deployed on the autonomous vehicle's onboard platform, communicates with the cloud-based task module and with smart parcel lockers and drone swarms within its service area. The edge computing module includes: The compartment status query submodule is configured to obtain the compartment status data of the smart express cabinet and construct the compartment status matrix; The grid pre-allocation submodule, connected to the grid status query submodule, is configured to perform grid pre-allocation and locking, as well as neighbor grid transfer, based on the grid status matrix, and to form a pre-allocation scheme. The task allocation submodule, connected to the grid pre-allocation submodule, is configured to allocate task packages to the UAV and issue flight commands according to the pre-allocation scheme. The redirection decision submodule, connected to the grid pre-allocation submodule and the task allocation submodule, is configured to perform a progressively escalating redirection operation and update the pre-allocation scheme when a grid experiences physical failure or a communication link is lost. The arbitration submodule, connected to the grid pre-allocation submodule and the task allocation submodule, is configured to arbitrate resources based on multiple dimensions during multi-machine contention. A drone swarm, which communicates with an edge computing module, consists of multiple drones, each equipped with a positioning module, a communication module, and a package delivery mechanism. The intelligent parcel locker network communicates with edge computing modules and drone swarms. The intelligent parcel locker network includes multiple intelligent parcel lockers. Each intelligent parcel locker is equipped with a compartment sensor array, docking platform, heartbeat communication module and compartment locking management module. The compartment sensor array deploys photoelectric switches, micro switches and motor current sensors in each compartment to achieve triple redundancy compartment status detection.
[0013] In some embodiments, the heartbeat communication module of each smart locker in the smart locker network is configured to: receive the heartbeat packet of the drone, parse the location, battery level and estimated arrival time information in it; send a lock refresh command to the locker lock management module; and when a physical failure of a locked locker is detected, first perform a locker replacement in this locker, and if that is not feasible, send a redirection request to the redirection decision submodule of the edge computing module.
[0014] In some embodiments, the edge computing module transmits data with the smart parcel locker network and drone swarm via the C-V2X PC5 direct communication interface.
[0015] This invention acquires the real-time status of smart parcel lockers through an edge computing module and constructs a locker status matrix. Based on this matrix, it pre-allocates and locks lockers for task packages, forming a pre-allocation scheme. Before takeoff, the drone establishes a two-way communication link with the target parcel locker, maintains locker locking through periodic communication, and triggers a tiered redirection operation to update the pre-allocation scheme when a physical locker failure or communication loss is detected. When multiple drones compete for a limited number of lockers, priority is ranked based on multiple dimensions. This method incorporates the real-time locker status as a dynamic decision variable into collaborative scheduling. The pre-locking mechanism avoids delivery failures due to no available lockers after drone arrival. The heartbeat maintenance and redirection mechanisms reduce invalid flights caused by changes in the terminal status. The multi-dimensional arbitration mechanism achieves balanced resource utilization in multi-drone concurrent scenarios, thereby significantly improving the delivery success rate and overall efficiency of large-scale parcel delivery. Attached Figure Description
[0016] Figure 1 This is a flowchart of the air-ground collaborative logistics delivery task allocation method based on grid state mapping proposed in this invention; Figure 2 This is a system architecture diagram of the air-ground collaborative logistics delivery task allocation system based on grid state mapping proposed in this invention.
[0017] In the diagram: 1. Cloud Task Module; 2. Edge Computing Module; 3. Drone Cluster; 4. Smart Express Locker Network; 21. Locker Status Query Submodule; 22. Locker Pre-allocation Submodule; 23. Task Allocation Submodule; 24. Redirection Decision Submodule; 25. Arbitration Submodule. Detailed Implementation
[0018] Firstly, this application provides a method for allocating air-to-ground collaborative logistics delivery tasks based on grid status mapping, such as... Figure 1 As shown, it includes the following steps: The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0019] When the unmanned vehicle arrives at the preset station, the edge computing module deployed on the unmanned vehicle obtains the status data of the smart express cabinets within the service range of the preset station. Based on the obtained status data, a status matrix is constructed. Each entry in the status matrix contains at least: the express cabinet identifier and the current available quantity of each type of cabinet. The edge computing module performs pre-allocation of lockers for one or more delivery packages based on the locker status matrix: for each package, it determines whether the available lockers of the target locker meet the package’s locker requirements. If they do, the corresponding locker is pre-locked; otherwise, a neighbor locker transfer is initiated, and a backup locker that meets the requirements is found within the service area and re-mapped. After all packages have been processed, a pre-allocation scheme containing the correspondence between package and locked locker is formed. According to the pre-allocation plan, a two-way communication link is established with the target express cabinet before the drone takes off, and periodic communication is carried out through the two-way communication link after the drone takes off in order to maintain the lock of the compartment in the pre-allocation plan. During flight, when the target parcel locker detects a physical failure of a locker that has been locked in the pre-allocation plan, or when the edge computing module detects a loss of communication link between the drone and the target parcel locker, a tiered redirection operation is triggered, and the pre-allocation plan is updated. When the edge computing module detects that multiple drones are simultaneously requesting the remaining slots of the same parcel locker and there are not enough available slots, the edge computing module prioritizes the requesting drones based on multiple dimensions. Drones with higher priority will be allocated slots first, while drones with lower priority will receive alternative allocation schemes.
[0020] As a preferred implementation, the air-ground collaborative logistics delivery task allocation system provided by this invention adopts a three-layer architecture of "cloud-edge-terminal". The cloud task module is deployed on a cloud server and is responsible for receiving package information from the upstream order system, grouping the packages by parcel locker and associating them with corresponding preset stations, and then sending the grouped data to the edge computing module on the unmanned vehicle. The edge computing module is the decision-making core of the entire system. It is deployed on the onboard computing platform of the unmanned vehicle and typically uses an industrial-grade embedded industrial control computer (such as NVIDIA Jetson AGX Orin or Intel Xeon D series), running a real-time operating system, and interacting with the smart parcel lockers and drone clusters within the service range through a C-V2X PC5 direct communication interface for low-latency data exchange. The edge computing module is internally divided into several functional sub-modules: a locker status query sub-module 21, a locker pre-allocation sub-module 22, a task allocation sub-module 23, a redirection decision sub-module 24, and an arbitration sub-module 25. Each sub-module cooperates with each other to complete the closed-loop control from status perception to dynamic scheduling.
[0021] The locker status query submodule 21 is responsible for sending locker status query requests to all smart parcel lockers within its service area. The data returned by the lockers includes the locker identifier, the current available number of small, medium, and large lockers, the queue length of drones waiting for delivery, and the availability of the docking platform. This data is encapsulated in lightweight JSON or Protobuf format and transmitted via C-V2X messages. Upon receiving this data, the edge computing module constructs a locker status matrix, where each row corresponds to a locker and the columns are the aforementioned fields. The reason for including queue length and platform status is that the number of lockers alone cannot reflect the actual receiving capacity of the locker. When the queue is too long, newly arriving drones need to wait in the air, consuming battery power; when the platform is occupied, docking cannot be completed even if a locker is idle. Therefore, these three dimensions represent spatial resources, temporal resources, and physical interface resources, respectively, and none can be omitted.
[0022] The slot pre-allocation submodule 22 processes task packages one by one based on this matrix. Specifically, it first groups the task packages by target parcel lockers, and then sorts the task packages within each group by the total number of slots required, from largest to smallest. This prioritizes task packages with high demand and reduces fragmentation. Then, it iterates through each task package, checking whether the number of available slots in the corresponding parcel locker in the matrix meets the demand. If it does, it locks the corresponding number of slots and deducts them from the matrix, while recording the locking information; if not, it marks the task package as pending transfer. After all normal locking is completed, the task packages to be transferred are sorted in ascending order of distance to the nearest backup parcel locker, and neighboring lockers are tried one by one for transfer, i.e., finding backup parcel lockers with sufficient slots within the service area and remapping them. Finally, a complete pre-allocation scheme is formed, which contains a one-to-one correspondence between each task package and the locked slots. This scheme is then passed to the task allocation submodule 23, which allocates task packages to each drone according to the scheme and issues flight commands, which include the target parcel locker ID, the locked slot ID, and the planned path.
[0023] Before the drone takes off, the task allocation submodule 23 instructs the drone to establish a two-way communication link with the target parcel locker. As a preferred implementation, this link is based on C-V2X PC5 direct communication, with a communication latency of less than 20 milliseconds. After takeoff, the drone sends heartbeat packets via this link at a fixed frequency (e.g., once every 5 seconds). These packets contain the drone's current location (GPS coordinates), remaining battery power, and estimated arrival time. Upon receiving the heartbeat packet, the target parcel locker refreshes the remaining lock time for the corresponding compartment (e.g., resetting it to 30 seconds). If no heartbeat packet is received for two consecutive heartbeat cycles, the parcel locker determines that communication has been lost and automatically releases the locker lock. This mechanism ensures that the locker resources are not occupied ineffectively for extended periods and provides a state-aware basis for subsequent redirection.
[0024] During flight, when the target parcel locker detects a physical failure in a locked compartment, it immediately triggers a redirection operation. Physical failure detection is achieved through a triple-redundant sensor array deployed in each compartment: photoelectric switches detect whether the compartment door is open and closed normally, microswitches detect the presence of foreign objects inside the compartment, and motor current sensors monitor whether the compartment drive motor is stalled. When any two sensors report an anomaly simultaneously, the compartment is determined to be faulty. At this point, the parcel locker first attempts a first-level redirection, searching for a similar available compartment within the locker for replacement, with a response latency of less than 100 milliseconds. If no available compartment is found, a redirection request is sent to the redirection decision submodule 24 of the edge computing module. The edge computing module performs a second-level redirection: searching within the same site's service area for a backup parcel locker that meets the compartment requirements and is closest to the drone's current location, and verifying whether the new flight path is within the drone's remaining flight range. If there is no solution in the same sector and the drone's flight range is insufficient to return, a third-level redirection is initiated: the task package is transferred to an unmanned vehicle at an adjacent site via vehicle-to-vehicle communication, and the drone at that site takes over to complete the delivery. After each redirection, the pre-allocation scheme is updated to ensure that subsequent decisions are based on the latest state.
[0025] When multiple drones simultaneously request remaining slots in the same parcel locker and there are insufficient slots available, the arbitration submodule 25 intervenes. It uses a weighted scoring system based on three dimensions: remaining battery status (lower battery, higher score), parcel urgency (closer to the delivery deadline, higher score), and availability of alternative parcel lockers (further distance to alternative lockers, higher score). The weights of these three dimensions can be dynamically adjusted according to operational needs; for example, the weight of urgency can be increased during peak periods. After calculating the overall priority score, the drone with the highest score is given priority access to a slot, while drones with lower priority automatically receive alternative allocation plans from the edge computing module (e.g., flying to other parcel lockers or returning to the drone vehicle to wait). This multi-dimensional arbitration mechanism is more adaptable to complex scenarios than a simple first-come, first-served rule, ensuring the feasibility of low-battery drones while also considering the business priority and resource scarcity of urgent parcels.
[0026] Through the aforementioned system and method, this invention incorporates the real-time status of parcel lockers as a dynamic decision variable into air-ground collaborative scheduling. The pre-allocation mechanism prevents delivery failures due to a lack of available lockers upon drone arrival; the heartbeat maintenance mechanism transforms end-point status commitments into timeliness constraints; the three-level redirection mechanism provides a closed-loop response from local to global when anomalies occur, significantly reducing invalid flights; and the multi-dimensional arbitration mechanism balances resource utilization in high-concurrency scenarios. Experiments show that in a typical scenario of 800 parcels and 12 parcel lockers, the delivery success rate is improved by approximately 25% compared to existing static coordinate methods, and invalid flight mileage is reduced by more than 40%.
[0027] As another feasible implementation, the communication link can also use a 5G Uu interface instead of C-V2X PC5. Although the latency increases slightly (about 50 milliseconds), it can still meet the requirements in urban areas with good base station coverage. The grid sensor array can be implemented using a single ToF laser rangefinder sensor in conjunction with an image recognition camera, which is lower in cost but slightly less reliable than the triple redundancy scheme. In addition, the hardware platform for the edge computing module can also be a low-power Raspberry Pi or FPGA accelerator card, suitable for small delivery fleets. These variations are all within the protection scope of this invention.
[0028] In some embodiments, each entry in the grid state matrix further includes: the length of the drone queue waiting for delivery and the status of whether the docking platform is available; the step of performing grid pre-allocation one by one includes: First, group the task packages according to the target express lockers, and then sort the task packages in each group from largest to smallest according to the total number of lockers required. For each task package, query the available quantity of the corresponding compartment type of the target express locker in the compartment status matrix. If the condition is met, lock the compartment and deduct the corresponding available quantity from the compartment status matrix. If the condition is not met, record it as a task package to be transferred. Once all lockouts are completed, the packages to be transferred are sorted in ascending order of distance from the nearest backup locker, and then transferred to adjacent lockers in sequence.
[0029] In a preferred embodiment, the locker status matrix includes not only the locker identifier and the available number of lockers of each size, but also the length of the waiting drone queue and the availability of the docking platform. The introduction of these two fields has clear engineering significance: the queue length reflects the current concurrent delivery pressure of the locker; when the queue is long, newly assigned drones may need to hover in the air waiting, which unnecessarily consumes battery life; the docking platform status directly determines whether a drone can complete a physical docking. Even if there are enough lockers, if the platform is occupied, the drone cannot deliver. Therefore, these three dimensions together constitute a complete picture of the locker's receiving capacity.
[0030] The specific execution steps for locker pre-allocation are as follows. Assume there are three parcel lockers L001, L002, and L003 within the service area of a pre-defined station. Their locker status matrix shows: L001 has 10 small lockers, 5 medium lockers, and 3 large lockers, with a queue length of 0, and the platform is idle; L002 has 5 small lockers, 2 medium lockers, and 1 large locker, with a queue length of 1, and the platform is idle; L003 has 0 small lockers, 3 medium lockers, and 0 large lockers, with a queue length of 2, and the platform is occupied. There are a total of 8 delivery packages, with 4 targeting L001, 2 targeting L002, and 2 targeting L003. The system first groups the packages by target locker and then sorts the packages within each group by the total number of lockers required, from largest to smallest. For example, in group L001, one of the four task packages requires two small slots and one medium slot (a total requirement of three), another requires one small slot (a total requirement of one), and so on. After sorting, the task package with the largest requirement is processed first.
[0031] Next, the system iterates through each task package, querying the number of available lockers in the corresponding matrix. For the task package in group L001, which requires 2 small and 1 medium lockers, L001 currently has 10 small lockers and 5 medium lockers, meeting the requirement. Therefore, 2 small lockers and 1 medium locker are locked, and the corresponding number is deducted from the matrix. After the update, L001 has 8 small lockers and 4 medium lockers remaining. This process is repeated until all task packages are processed. For the task packages in group L003, since L003 has 0 small and 0 large lockers and the platform is occupied, it cannot meet any requirements. Therefore, these two task packages are marked as pending transfer. After all normal locking is completed, the system sorts the task packages to be transferred in ascending order of distance from the nearest spare locker. Assuming that the nearest spare locker to L003 is L001 (0.5 km away), followed by L002 (0.8 km away), the first task package to be transferred attempts to map to L001. If L001 has enough remaining lockers, it is locked; otherwise, L002 is tried. In this example, L001 still has enough slots, and both task packages to be transferred were successfully transferred to L001, thus forming a complete pre-allocation scheme.
[0032] Another feasible implementation is to prioritize packages based on their urgency rather than the total number of lockers, for example, prioritizing packages that are about to expire. However, this can lead to locker fragmentation and is suitable for time-sensitive scenarios. Another variation is that when a package to be transferred cannot find any spare lockers within the service area, the system marks it as a failure and notifies the cloud for rescheduling; this can serve as a fallback strategy.
[0033] The benefits of this pre-allocation mechanism are multifaceted. First, by introducing queue length and platform status, it avoids assigning tasks to congested or unconnected lockers, thus reducing the probability of drones queuing in the air and saving battery life. Second, the strategy of prioritizing large task packages based on total demand prioritizes larger ones, reducing the fragmentation problem of small lockers being occupied and unable to accommodate large packages, thereby improving locker utilization. Third, the deduction operation ensures the exclusivity of locker resources and prevents duplicate allocation. Finally, the neighbor locker transfer provides a nearby alternative for task packages with insufficient lockers, preventing the entire task package from failing due to a single locker's resource shortage. Experimental data shows that, under the same conditions, this pre-allocation method can increase locker utilization by approximately 18% and improve the success rate of one-time task package allocation by 22%.
[0034] In some embodiments, the escalating redirection operation includes: The first level involves the target parcel locker searching for and replacing a similar available compartment within the locker itself, with a response latency of less than 100 milliseconds. The second level involves the edge computing module searching within the same site's service area for a backup parcel locker that meets the locker requirements and is closest to the drone's current location, and verifying that the new flight path is within the drone's remaining range. The third level involves the edge computing module querying the status of parcel locker compartments within the service range of adjacent stations, and then transferring the task package to unmanned vehicles at adjacent stations via vehicle-to-vehicle communication for relay delivery.
[0035] In a preferred embodiment, the three-level redirection mechanism provides a progressively escalating response strategy for situations such as slot failure or communication loss during flight. Taking a real-world scenario as an example: A drone, numbered D1, is flying towards parcel locker L002, and its pre-allocation scheme has locked one of L002's medium slots, B-08. During flight, L002's slot sensor detects that B-08 cannot be opened normally due to mechanical jamming, the motor current sensor captures a continuous stall current, and the photoelectric switch and microswitch also report abnormalities. Two of the three redundant sensors report errors, indicating a physical failure of the slot. At this point, the first-level redirection is triggered: L002's slot locking management module immediately searches for an available slot of the same type (medium slot) within the locker. Assuming that L002 also has another medium slot, B-14, that is available, the parcel locker automatically replaces the locking relationship from B-08 to B-14 and sends the new slot ID to D1 in the next heartbeat response message. The entire process is completed locally in the parcel locker, with a response latency of less than 100 milliseconds. The drone won't even detect the path change and can continue flying along its original route to complete the delivery. This rapid response design minimizes the impact of anomalies on flight plans.
[0036] If the first-level redirection fails, for example, if there are no available slots of the same type in locker L002, a second-level redirection is triggered. At this time, L002 sends a redirection request to the edge computing module, carrying D1's current task package information (package size, quantity, target locker, etc.) as well as D1's latest location and remaining battery power. The edge computing module searches for all available lockers within the service area of L002's site that meet the slot requirements, and selects the locker closest to D1's current location. Assuming L004 has an available medium-sized slot, and D1's current location is only 0.8 kilometers from L004, while D1 has 1.5 kilometers of remaining range, the new flight path is 1.3 kilometers long, within the range. The edge computing module then issues a new target locker ID and slot ID to D1, and D1 redirects to L004 to complete the delivery. To ensure the reliability of the decision, the edge computing module also considers the queue length and platform status of the available locker to avoid sending the drone to another congested or unavailable locker. Another feasible approach is for the edge computing module to simultaneously calculate the comprehensive score of multiple candidate cabinets (e.g., distance, battery life margin, and queuing time weighted) and select the optimal solution rather than simply the closest distance.
[0037] If the second level also proves ineffective—meaning there are no available backup lockers within the service area of the same site, and D1's remaining range is insufficient to return to the drone—then the third level, cross-sector handover, is initiated. The edge computing module sends a query request to the drones at adjacent sites via vehicle-to-vehicle (V2V) communication to obtain the real-time locker status within the service area of those sites. Assuming that locker L008 has an available locker within the service area of adjacent site P3, and that site P3 has a drone D7 that has not yet departed, the edge computing module transmits D1's task package information to the edge computing module of the P3 drone via V2V. The latter assigns the task package to D7 and generates a new pre-allocation plan and flight instructions. D1 is then notified to abandon the delivery and return to base or land at the nearest locker, and its original locked locker is released. Cross-sector handover achieves seamless task relay, preventing complete task failure due to resource depletion at a single site. In addition, there is another variation: if there are no available resources at adjacent sites, the edge computing module can mark the task package as "to be retried" and wait for the resources of subsequent sites to be released before allocating it. However, this will increase the delivery delay and is usually used as a fallback option.
[0038] The three-level redirection mechanism demonstrates significant benefits. Level 1, same-cabinet swapping, resolves issues quickly with minimal drone flight load, suitable for minor, localized faults. Level 2, same-sector swapping, redistributes resources within the site; while requiring changes to flight paths, this remains manageable, and endurance testing mitigates the risk of drone power outages. Level 3, cross-sector transfer, utilizes vehicle-to-vehicle communication for cross-site collaboration, addressing extreme resource shortages. Overall, this mechanism offloads redirection decisions from the computationally limited drones to the parcel lockers and edge computing, ensuring both rapid response and global optimization. Actual testing shows that in scenarios simulating random locker failures, the three-level redirection mechanism resolves over 95% of anomalies locally or within the same sector, with less than 5% requiring cross-sector transfer, reducing invalid flight mileage by approximately 60% compared to the no-redirection solution.
[0039] In some embodiments, the detection method for physical failure of the grid includes: using a triple redundant sensor array consisting of a photoelectric switch, a micro switch, and a motor current sensor deployed in each grid to detect in real time whether the grid has mechanical failure, is occupied by foreign objects, or has sensor malfunction.
[0040] In a preferred embodiment, each compartment of the smart parcel locker is equipped with a triple-redundant sensor array, consisting of a photoelectric switch, a micro switch, and a motor current sensor, used to detect physical failures in the compartment in real time. The photoelectric switch is installed on the top inner side of the compartment and determines whether an object is obstructing the compartment by emitting and receiving infrared beams. The micro switch is installed at the hinge of the compartment door to sense the opening and closing state of the door. The motor current sensor is connected in series in the power supply circuit of the compartment's drive motor to monitor the current waveform during motor operation. When the compartment is operating normally, the output signals of the three sensors exhibit a specific correlation pattern: for example, when a user retrieves a package, the motor drives the compartment door to open, the current sensor detects a short pulse, the micro switch flips its state, and the photoelectric switch temporarily malfunctions due to the door opening—this is normal operation. Only when the sensor signals exhibit an abnormal combination that does not conform to the normal pattern is a physical failure determined.
[0041] For example, consider this scenario: a door opening is blocked by an object, but the user hasn't operated it. The photoelectric switch detects the obstruction, but the microswitch indicates the door is closed, and the motor current sensor shows no abnormal pulses. The conflicting signals from the two sensors (photoelectric and microswitch) lead the system to determine the opening is blocked and trigger a failure alarm. Another scenario is motor stall: when the motor attempts to open the door but encounters mechanical jamming, the current sensor detects a continuous high current (stalled current), while the photoelectric and microswitch remain unchanged. In this case, only the current sensor alarms, while the other two are normal. However, because the abnormal current duration exceeds a threshold (e.g., 2 seconds), the system determines it's a mechanical fault. Yet another scenario is sensor malfunction: for example, the photoelectric switch continuously outputs an obstruction signal due to dust, but the microswitch indicates the door is closed and the motor is not moving. In this case, the three sensor signals are inconsistent. The system uses a majority voting mechanism: if two sensors are consistent and one is inconsistent, the majority result is adopted, and the inconsistent sensor is marked as suspicious and replaced during subsequent maintenance.
[0042] The advantage of triple redundancy design lies in significantly reducing the false alarm rate through cross-validation. While any single sensor may produce erroneous readings due to environmental interference (dust, vibration, aging), the probability of all three sensors malfunctioning simultaneously is extremely low. In actual deployment, the system employs a "two-ticket" system: a grid failure is confirmed only when at least two sensors simultaneously indicate the same anomaly type. This mechanism keeps the false alarm rate below 0.1% and avoids missed alarms due to a single sensor failure. Furthermore, the sensor array can distinguish between failure types, providing a basis for subsequent redirection decisions. For example, mechanical failures may require repair, while foreign object obstruction can be resolved simply by cleaning. The system can decide whether to immediately release the grid or wait for processing based on the failure type. Ultimately, this detection method ensures the real-time nature and reliability of grid status data, providing a solid foundation for grid pre-allocation and redirection operations, and effectively avoiding invalid flights or delivery failures due to status misjudgments.
[0043] In some embodiments, the method for determining communication link failure includes: the drone sending a heartbeat packet at a fixed frequency, the heartbeat packet containing the drone's current location, remaining battery power, and estimated arrival time; the target parcel locker refreshing the remaining lock time when it receives the heartbeat packet, and determining communication failure and releasing the locker lock when it does not receive a heartbeat packet for two consecutive heartbeat sending cycles; the heartbeat packet is sent once every 5 seconds, and the locker lock refresh time is 30 seconds.
[0044] In a preferred embodiment, the heartbeat communication protocol between the drone and the target locker is configured with the following parameters: the drone sends heartbeat packets at a fixed frequency of once every 5 seconds. Each heartbeat packet contains the drone's current location (centimeter-level latitude and longitude coordinates provided by the onboard RTK module), remaining battery percentage, and estimated arrival time (estimated based on current speed and remaining path). Upon receiving a heartbeat packet, the target locker refreshes the remaining lock time of the corresponding compartment to 30 seconds. This means that as long as the drone continues to send heartbeat packets, the compartment will remain locked. If the locker does not receive any heartbeat packets for two consecutive heartbeat transmission cycles (i.e., 10 seconds), it determines that communication has been lost, immediately releases the locker lock, and updates the compartment status to available.
[0045] Taking a real-world flight scenario as an example: After drone D1 takes off from the unmanned vehicle, it begins sending heartbeat packets to the parcel locker L001. At the 5th, 10th, 15th, and subsequent seconds, L001 receives D1's heartbeat packet and resets the locker's lockout time to 30 seconds. Suppose D1 suddenly crosses a signal-blocked area during flight, causing the heartbeat packets at the 20th and 25th seconds to be lost. If L001 still hasn't received a heartbeat packet by the end of the 25th second, it has been without a heartbeat for 10 consecutive seconds, determines a loss of connection, and automatically releases the locker locked by D1. When D1 resumes communication and resends a heartbeat after 30 seconds, L001 responds with a status code indicating that the locker has been released, and D1 needs to request a new locker from the edge computing module. Another scenario is that the drone crashes due to battery depletion or mechanical failure, permanently stopping heartbeat packets. In this case, the parcel locker will release the locker after 10 seconds, preventing the locker from being locked and occupied for an extended period by a zombie locker.
[0046] The heartbeat parameter settings were determined through practical testing and trade-offs: a 5-second transmission frequency ensures real-time performance without excessively consuming the drone's communication bandwidth and power; a 30-second lock refresh time provides sufficient buffering to prevent accidental release even with occasional packet loss; and a two-cycle (10-second) disconnection determination window balances brief communication interruptions with rapid response. As other feasible implementations, the transmission frequency can be adjusted to once every 3 seconds to enhance real-time performance, suitable for high-speed flight or environments with dense obstacles, but this increases communication overhead; the lock refresh time can also be extended to 60 seconds to tolerate longer interruptions, but this reduces grid turnaround efficiency; and a three-cycle (15-second) disconnection determination cycle can be used to reduce false positives, but this prolongs the invalid lock time. Operators can flexibly configure these parameters based on actual network quality and flight environment.
[0047] The beneficial effects of this heartbeat mechanism are reflected in three aspects. First, by maintaining grid lock through periodic heartbeats, it ensures the drone's exclusive right to the target grid during flight, preventing other drones or manual pickups from seizing the allocated grid. Second, the loss-of-connection detection and automatic release mechanism prevents long-term invalid grid occupancy due to drone malfunction or communication interruption, improving grid turnover rate. Finally, the location, battery level, and ETA information carried in the heartbeat packet provide the edge computing module with real-time status awareness during flight, serving as an important data source for triggering redirection and arbitration decisions. Real-world test data shows that in typical urban low-altitude communication environments, this heartbeat mechanism can reduce invalid grid occupancy time by more than 90%, while keeping the false release rate below 0.5%.
[0048] In some embodiments, multiple dimensions include remaining battery status, parcel delivery urgency, and availability of alternative parcel lockers; priority ranking adopts a weighted scoring method, which assigns preset weights to the three dimensions of remaining battery status, parcel delivery urgency, and availability of alternative parcel lockers and then calculates a comprehensive priority score. The lower the remaining battery, the closer the parcel's delivery deadline is, and the farther away the alternative parcel locker is, the higher the priority.
[0049] In a preferred embodiment, when multiple drones simultaneously request the remaining slots of the same parcel locker and there are insufficient available slots, the arbitration submodule 25 of the edge computing module prioritizes them based on three dimensions: remaining battery status, parcel delivery urgency, and availability of alternative parcel lockers. Remaining battery status refers to the percentage of the drone's current battery remaining relative to its full capacity; lower battery levels mean less margin for error, and a higher risk of missing the current slot and flying to a backup locker, thus requiring higher priority. Parcel delivery urgency measures how much time remains until the latest delivery time; the closer the time, the more urgent the task, and the higher the priority. Availability of alternative parcel lockers examines which other nearby available parcel lockers are available besides the target locker; the farther the alternative lockers are, the stronger the irreplaceability of the current slot, and the higher the priority. Arbitration submodule 25 assigns preset weights to these three dimensions, such as a power weight of 0.4, a timeliness weight of 0.3, and an alternative distance weight of 0.3. Then, it calculates a comprehensive priority score for each drone. The drone with the highest score is given priority in receiving a parcel locker allocation, while drones with lower priority automatically receive alternative allocation schemes issued by the edge computing module, such as flying to another parcel locker or returning to the unmanned vehicle to wait.
[0050] Taking a specific scenario as an example: Only one slot remains in parcel locker L001, and three drones, D1, D2, and D3, request it simultaneously. D1 has 20% battery remaining, its package has 30 minutes to expire, and the nearest alternative locker is 1.2 kilometers away; D2 has 45% battery remaining, its package has 15 minutes to expire, and the nearest alternative locker is 0.5 kilometers away; D3 has 60% battery remaining, its package has 60 minutes to expire, and the nearest alternative locker is 2.0 kilometers away. The arbitration submodule 25 comprehensively evaluates the following: D1 has the lowest battery level, moderate timeliness, and a relatively far alternative locker; D2 has the most urgent timeliness but a relatively high battery level and a very close alternative locker; D3 has the highest battery level and the most lenient timeliness but the furthest alternative locker. After weighted calculation, D1 receives the highest priority due to its low battery and distant alternative locker, followed by D2, and then D3. Ultimately, D1 gets the slot, while D2 and D3 receive alternative solutions. In another scenario, if D2's battery level drops to 15%, D2 will surpass D1 to become the highest priority due to its extremely dangerously low battery level. Another variation is that the weights can be dynamically adjusted according to the operating period. For example, the weight of timeliness can be increased during the evening peak period, and the weight of power consumption can be increased during periods of frequent low power consumption, in order to adapt to different operating strategies.
[0051] The beneficial effects of this multi-dimensional arbitration mechanism are significant. It abandons the simple first-come, first-served rule, instead comprehensively evaluating performance from three perspectives: feasibility, business priority, and resource scarcity. Drones with low battery power are given priority access to nearby racks, significantly reducing the risk of crashes or mission failures due to insufficient battery life; emergency packages receive priority access to resources, ensuring timely delivery; and drones with fewer alternatives are given priority access to the current rack, avoiding overall failure due to a lack of alternatives. Real-world testing data shows that in multi-drone competition scenarios during peak daily hours, this mechanism increases mission completion rates by approximately 15%, reduces drone forced landings due to battery depletion by approximately 70%, and raises the on-time delivery rate of emergency packages to over 98%.
[0052] In some embodiments, data transmission is performed between the unmanned vehicle, the smart parcel locker, and the drone via a C-V2X PC5 direct-connect communication interface.
[0053] In a preferred embodiment, the unmanned vehicle, the smart parcel locker, and the drone transmit data via a C-V2X PC5 direct communication interface. The PC5 interface is a direct communication technology defined by the 3rd Generation Partner Program, allowing devices to conduct point-to-point or broadcast communication directly without the need for cellular base station relays, operating in the dedicated 5.9GHz frequency band. At the hardware level, the unmanned vehicle's onboard edge computing platform integrates an automotive-grade OBU supporting PC5 communication; the smart parcel locker embeds a communication module that also supports PC5; and the drone carries a lightweight PC5 communication chip. The three devices form a self-organizing direct communication network through a unified PC5 interface, with a stable communication latency of less than 20 milliseconds, maintaining reliable connectivity even during high-speed movement or in environments with signal obstruction.
[0054] Taking a specific scenario as an example: After an unmanned vehicle arrives at a preset station, its edge computing module broadcasts a locker status query request to all smart parcel lockers within its service area via the PC5 interface. Upon receiving the request, the locker returns locker status data within milliseconds, and the edge computing module constructs a locker status matrix based on this data. Subsequently, before takeoff, the drone establishes a point-to-point connection with the target locker via the PC5 interface and begins sending heartbeat packets. During flight, if the locker detects a locker malfunction, it immediately sends a redirection request to the edge computing module via the PC5 interface. The edge computing module responds within 20 milliseconds and sends a new target address. In another scenario, when multiple drones simultaneously compete for the remaining lockers in the same parcel locker, the arbitration result is broadcast via the PC5 interface, received synchronously by all drones, avoiding conflicts caused by communication delays.
[0055] As another feasible implementation, if the cellular network coverage in the deployment area is good and latency requirements are not high, a 5G Uu interface can be used instead of PC5 direct communication. Although the end-to-end latency will increase to around 50 milliseconds, existing base station infrastructure can be utilized, reducing equipment costs. Another variation is to use Wi-Fi 6 DLS direct link mode, suitable for short-distance, low-mobility scenarios, but with limited coverage and unsuitable for high-speed drone flight. Furthermore, in extreme environments, narrowband IoT technologies such as LoRa can be used to transmit critical status information, but only a small amount of data can be transmitted and latency is relatively high, making it suitable as a backup link.
[0056] The benefits of the C-V2X PC5 direct communication interface are mainly reflected in low latency and high reliability. A communication latency of less than 20 milliseconds ensures the real-time performance of critical operations such as grid status queries, heartbeat maintenance, and redirection command issuance, enabling drones to respond promptly to status changes even during high-speed flight. The direct communication feature, which does not rely on base stations, eliminates network coverage blind spots, allowing stable operation even in areas with sparse base stations, such as suburbs and industrial parks. Simultaneously, the PC5 interface supports broadcast and multicast modes, allowing edge computing modules to simultaneously issue commands to multiple drones, significantly improving communication efficiency in high-concurrency scenarios. Real-world testing data shows that in a typical urban logistics and distribution environment, the average packet loss rate of PC5 direct communication is less than 0.1%, and the end-to-end latency is consistently around 15 milliseconds, fully meeting the stringent requirements for real-time communication and reliability in air-to-ground collaborative logistics and distribution.
[0057] Secondly, this application provides an air-to-ground collaborative logistics delivery task allocation system based on grid status mapping, used to execute the methods in any of the above embodiments, such as... Figure 2 As shown, it includes: Cloud task module 1 is configured to receive package order information, group packages by parcel locker and associate them with preset sites; Edge computing module 2, deployed on the unmanned vehicle's onboard platform, communicates with cloud task module 1 and with smart parcel lockers and drone swarms 3 within its service range; edge computing module 2 includes: The compartment status query submodule 21 is configured to obtain the compartment status data of the smart express cabinet and construct the compartment status matrix; The grid pre-allocation submodule 22 is connected to the grid status query submodule 21 and is configured to perform grid pre-allocation and locking, as well as neighbor grid transfer, based on the grid status matrix, and form a pre-allocation scheme. The task allocation submodule 23 is connected to the grid pre-allocation submodule 22 and is configured to allocate task packages to the UAV and issue flight commands according to the pre-allocation scheme. The redirection decision submodule 24, connected to the grid pre-allocation submodule 22 and the task allocation submodule 23, is configured to perform a progressively escalating redirection operation and update the pre-allocation scheme when a grid fails physically or a communication link is lost. Arbitration submodule 25, connected to grid pre-allocation submodule 22 and task allocation submodule 23, is configured to perform resource arbitration based on multiple dimensions during multi-machine competition. The drone swarm 3 is connected to the edge computing module 2. The drone swarm 3 includes multiple drones, each of which is equipped with a positioning module, a communication module and a package delivery mechanism. The smart parcel locker network 4 communicates with the edge computing module 2 and the drone cluster 3. The smart parcel locker network 4 includes multiple smart parcel lockers. Each smart parcel locker is equipped with a compartment sensor array, a docking platform, a heartbeat communication module, and a compartment locking management module. The compartment sensor array deploys photoelectric switches, micro switches, and motor current sensors in each compartment to achieve triple-redundant compartment status detection.
[0058] A preferred system embodiment employs a three-tier computing architecture. The cloud task module 1 is deployed on a public or private cloud server, running as a microservice, and receives package information from the upstream order system via a RESTful API. This module groups packages by target parcel lockers and associates each locker with a corresponding preset station. The grouped data is then pushed to the edge computing module 2 on the unmanned vehicle via the MQTT protocol. The edge computing module 2 uses an industrial-grade embedded industrial control computer, such as an NVIDIA Jetson AGX Orin, running Ubuntu Linux and the ROS2 robot operating system. This industrial control computer exchanges data with the smart parcel lockers and the drone cluster 3 via a C-V2X PC5 direct connection communication interface. The edge computing module 2 is internally divided into five functional sub-modules, each running as an independent ROS2 node, communicating via topics and services.
[0059] The locker status query submodule 21 broadcasts locker status query requests to all smart parcel lockers within its service range at fixed intervals (e.g., every 30 seconds) or receives status updates proactively reported by the lockers. It maintains an in-memory locker status matrix table, where each record contains the parcel locker ID, the number of available small, medium, and large lockers, the queue length, and the platform status. The locker pre-allocation submodule 22 subscribes to this matrix table, triggering pre-allocation calculations whenever a new task package arrives or the matrix is updated. It first groups the task packages by target parcel lockers, sorts the task packages within each group in descending order of the required total number of lockers, and then iterates through each package, querying the available number of lockers for the corresponding parcel locker in the matrix. If the quantity is sufficient, it locks and deducts the number of lockers; otherwise, it marks the package as awaiting transfer. After all task packages have been processed, the task packages awaiting transfer are sorted in ascending order of distance to the nearest backup locker, and transfers are attempted from neighboring lockers sequentially. Finally, a pre-allocation plan is generated and published in the form of a ROS2 message. The task allocation submodule 23 subscribes to the pre-allocation plan, generates flight commands for each drone containing the target parcel locker ID, the locked slot ID, and the flight path, and sends them to the corresponding drone via the PC5 interface. The redirection decision submodule 24 monitors for slot failure events from smart parcel lockers and communication loss events from drones. Upon receiving an anomaly, it immediately executes a three-level redirection logic: first, it checks if there are any available slots of the same type in the locker; if not, it queries other parcel lockers within the service area of the site; if still not available and the drone's battery life is insufficient, it contacts a drone at a neighboring site via V2V communication. The pre-allocation plan is updated and republished after each redirection. The arbitration submodule 25 monitors multiple drone requests for slots in the same parcel locker. When competition is detected, it collects the remaining battery power, package deadline, and distance to alternative lockers for each drone, calculates a priority score according to preset weights, allocates the slot to the drone with the highest score, and sends alternative plans to other drones.
[0060] Drone swarm 3 comprises multiple hexacopter drones, each equipped with an RTK high-precision positioning module (positioning accuracy ±2 cm), a PC5 communication module, and an electric parcel delivery pod. The parcel delivery pod uses an electromagnetic locking mechanism; upon reaching the target compartment, it is visually guided to align and release the parcel. The smart parcel locker network 4 contains multiple parcel lockers, each with a main control board using an STM32 microcontroller running the FreeRTOS real-time system. Each compartment houses a triple-redundant sensor array: photoelectric switches (Omron E3Z series), microswitches (Omron D2F series), and a motor current sensor (ACS712 Hall effect sensor). Sensor signals converge to the main control board via an I2C bus. The main control board runs state machine logic; when at least two sensors simultaneously indicate an anomaly, the compartment is deemed faulty, and a fault event is sent to the edge computing module 2 via the PC5 interface. The docking platform uses ultrasonic sensors and infrared beacons to assist drones in precise landing. The heartbeat communication module receives drone heartbeat packets at 5-second intervals, analyzes location, battery level, and ETA, and refreshes the compartment locking timer. The grid lock management module maintains a lock table that records the locked drone ID and lock expiration time for each grid. The drone is unlocked when the lock expires or when a release command is received.
[0061] As another feasible implementation, the hardware platform of edge computing module 2 can be replaced with a low-power Raspberry Pi 4B paired with an Intel Movidius Neural Compute Stick, suitable for small delivery fleets, but with weaker computing power, suitable for scenarios handling no more than 50 task packages. The communication link can use a 5G Uu interface instead of a direct PC5 connection. Although the latency increases to around 50 milliseconds, it can utilize existing 5G base station coverage, reducing the cost of building its own communication infrastructure. A quadcopter drone can also be used; it has a smaller payload capacity but is cheaper, suitable for delivering small and lightweight packages. The sensor array of the smart parcel locker can be simplified to a single ToF laser rangefinder combined with camera image recognition, using AI algorithms to determine the locker status. This is cheaper but slightly less reliable than a triple-redundancy solution.
[0062] The system's beneficial effects are reflected on multiple levels. The cloud-based task module 1 handles offline planning and global coordination, reducing the burden on the edge computing module 2. The edge computing module 2 pushes all real-time decisions down to the unmanned vehicle, avoiding cloud communication latency and enabling slot pre-allocation, redirection, and arbitration to be completed within milliseconds. The five sub-modules have clearly defined responsibilities and collaborate through a loosely coupled message mechanism, facilitating independent upgrades and maintenance. The drone swarm 3 and the smart parcel locker network 4 connect to the system through a unified PC5 communication interface, achieving low-latency, highly reliable end-to-end data interaction. A triple-redundant sensor array ensures the accuracy of slot status detection, with a false alarm rate of less than 0.1%, providing a reliable data foundation for pre-allocation and redirection. Overall, the system performs excellently in large-scale concurrent scenarios. In tests with 800 parcels and 12 parcel lockers, the delivery success rate exceeded 97%, invalid flight mileage decreased by 55%, and slot turnover rate increased by 30%, fully meeting the needs of smart logistics for efficiency, reliability, and dynamic adaptability.
[0063] In some embodiments, the heartbeat communication module of each smart parcel locker in the smart parcel locker network 4 is configured to: receive the heartbeat packet of the drone, parse the location, battery level and estimated arrival time information in it; send a lock refresh command to the locker lock management module; and when a physical failure of a locked locker is detected, first perform a locker replacement in this locker, and if that is not feasible, send a redirection request to the redirection decision submodule 24 of the edge computing module 2.
[0064] In a preferred embodiment, the smart parcel locker's heartbeat communication module is based on a dedicated communication coprocessor, such as the Espressif ESP32 or Cypress CYW43438, running a lightweight TCP / IP protocol stack and mounting a C-V2XPC5 communication module. After power-on, this module registers a UDP service port and continuously listens for data packets from drones. When a drone approaches the locker, it sends a heartbeat packet to this port every 5 seconds. The heartbeat packet uses a compact binary format, sequentially encoding the drone ID, current latitude, current longitude, remaining battery percentage (integer, 0-100), and estimated arrival time in seconds (seconds from the current time). Upon receiving the packet, the heartbeat communication module calls a built-in parsing function to extract each field and writes the parsing results to a shared memory area for the locker locking management module to read.
[0065] Specifically, the heartbeat communication module maintains a timer. Each time a valid heartbeat packet is received, the corresponding grid lock timer is reset to 30 seconds. Simultaneously, it checks the grid ID list locked by the drone in the grid lock management module. If the lock timer is successfully refreshed, it sends a pulse signal to the grid lock management module via GPIO to notify it that the lock status has been updated. If no message is received for two consecutive heartbeat cycles (10 seconds), the heartbeat communication module determines that communication has been lost, proactively sends a release command to the grid lock management module, and records the loss of communication event in its local log.
[0066] When the locker sensor array detects a physical failure of a locked locker, the heartbeat communication module receives an interrupt signal from the locker locking management module. It first checks if there is a free locker of the same type within the locker. If so, the heartbeat communication module directly modifies the locking mapping table, transferring the lock relationship of the failed locker to a free locker, and includes the new locker ID in the next heartbeat response message to notify the drone to update the target. If there is no available locker within the locker, the heartbeat communication module assembles a redirection request message containing the failed locker ID, drone ID, package size, etc., and sends it to the redirection decision submodule 24 of the edge computing module 2 via the PC5 interface, waiting for a return of the new target parcel locker and locker ID.
[0067] As another feasible implementation, the heartbeat communication module can also use the LoRaWAN protocol for long-distance, low-speed communication, suitable for remote areas such as rural or mountainous regions where C-V2X coverage is unavailable, but the latency will increase to hundreds of milliseconds. Another variation is to use Bluetooth 5.0's LE Audio broadcast mode to achieve low-power heartbeat transmission within a range of tens of meters around the parcel locker, suitable for close-range interaction after a drone hovers at low speed or lands. Furthermore, if the parcel locker itself does not have independent communication capabilities, the unmanned vehicle can act as a relay to forward heartbeat packets. In this case, the logic of the heartbeat communication module is migrated to the edge computing module 2 of the unmanned vehicle, and the parcel locker only serves as a sensor and actuator terminal.
[0068] The benefits of this heartbeat communication module are significant. By analyzing the drone's location, battery level, and ETA information in real time, the edge computing module 2 and the parcel locker can monitor the flight status of each drone, providing crucial data support for dynamic redirection and resource arbitration. The lock-and-refresh mechanism ensures the exclusivity of the parcel locker during the drone's flight, preventing other tasks from taking over. The locker replacement function achieves millisecond-level rapid recovery when the locker fails, allowing the drone to continue delivery without changing its flight path, greatly reducing invalid flights. When the locker cannot resolve the issue, a redirection request is promptly sent to the edge computing module 2, escalating the problem to a global level for processing, forming a closed-loop response system from local to global. Actual test data shows that the heartbeat communication module operates normally 99.9% of the time, with an average response latency of less than 15 milliseconds, effectively supporting the stable operation of the entire air-ground collaborative system.
[0069] In some embodiments, the edge computing module 2 transmits data with the smart parcel locker network 4 and the drone cluster 3 via a C-V2X PC5 direct communication interface.
[0070] In a preferred embodiment, edge computing module 2 transmits data with the smart parcel locker network 4 and the drone swarm 3 via a C-V2X PC5 direct communication interface. This interface is based on the NR-V2X standard developed by the 3rd Generation Partnership Project, operates in the 5.9GHz ITS dedicated frequency band, and supports unicast, multicast, and broadcast communication modes. Edge computing module 2 is deployed on the unmanned vehicle's onboard platform, equipped with a Qualcomm Snapdragon SA515M PC5 communication module and an external omnidirectional antenna, achieving a peak transmission rate of up to 50Mbps. Each parcel locker in the smart parcel locker network 4 has a built-in Unisoc V510 PC5 chip, while each drone in the drone swarm 3 carries a lightweight Quectel AG550Q module. The three form a self-organizing mesh network through a unified PC5 air interface, enabling point-to-point direct communication without the intervention of cellular base stations. The typical end-to-end latency is stable within 15 milliseconds, maintaining uninterrupted connectivity even when drones are moving at a speed of 60 km / h.
[0071] In actual operation, this interface carries multiple key data streams. For example, when the unmanned vehicle arrives at the preset station, the edge computing module 2 broadcasts a locker status query request via PC5. The smart locker replies with a status data packet within 200 microseconds after receiving the request, and the edge computing module 2 constructs a locker status matrix based on this. Subsequently, each drone establishes a PC5 unicast link with the target locker before takeoff, sending a heartbeat packet every 5 seconds, containing location, battery level, and ETA information. If the locker detects a locker failure, it immediately sends a redirection request to the edge computing module 2 via PC5. The edge computing module 2 completes the calculation within 10 milliseconds and sends a new target address via PC5 unicast. Another scenario is when multiple drones compete for the same locker simultaneously. The edge computing module 2 sends the arbitration result to all competing drones simultaneously via PC5 multicast, ensuring that all drones are aware of the allocation scheme synchronously, avoiding conflicts caused by sequential reception.
[0072] As another feasible implementation, if the deployment area has good 5G base station coverage and latency requirements are relaxed, the 5G NR Uu interface can be used instead of the PC5 direct connection. Data routing is performed through the base station's UPF, with an end-to-end latency of approximately 50 milliseconds. However, existing mobile network infrastructure can be reused, reducing equipment investment. Another variation is to use a Wi-Fi 6 DLS direct link mode to establish a 60GHz millimeter-wave connection between the parcel locker and the unmanned vehicle. This is suitable for short-range, high-throughput scenarios, but the coverage is limited and it is not suitable for long-distance drones. In addition, LTE Cat-M1 can be used as a backup link in remote areas. Although the speed is lower and the latency is higher, it is sufficient to transmit small data packets such as status queries and heartbeats.
[0073] The benefits of the C-V2X PC5 direct communication interface are multifaceted. A communication latency of less than 20 milliseconds ensures the real-time performance of critical operations such as grid status queries, heartbeat maintenance, and redirection commands, enabling drones to respond promptly to status changes even during high-speed flight and preventing delivery failures due to communication delays. The direct communication feature, which does not rely on base stations, eliminates network coverage blind spots, allowing stable operation even in sparsely populated areas such as industrial parks and suburbs. Broadcast and multicast modes allow the edge computing module 2 to simultaneously issue commands to multiple drones, significantly improving communication efficiency in high-concurrency scenarios and avoiding the serial latency caused by unicast. Real-world testing data shows that in a typical urban logistics and delivery environment, the average packet loss rate of PC5 direct communication is less than 0.1%, and the end-to-end latency is stable at around 15 milliseconds, fully meeting the stringent requirements of air-to-ground collaborative logistics and delivery for real-time communication and reliability, providing a solid underlying communication guarantee for the stable operation of the entire system.
[0074] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0075] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0076] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for allocating air-to-ground collaborative logistics delivery tasks based on grid state mapping, characterized in that, include: When the unmanned vehicle arrives at the preset station, the edge computing module deployed on the unmanned vehicle obtains the status data of the smart express lockers within the service range of the preset station, and constructs a locker status matrix based on the obtained locker status data. Each entry of the locker status matrix contains at least: the express locker identifier and the current available quantity of each type of locker. Based on the grid status matrix, the edge computing module performs grid pre-allocation for one or more task packages to be delivered: for each task package, it determines whether the available grids of the target express locker meet the grid requirements of the task package. If they do, the corresponding grid is pre-locked; if not, a neighbor locker transfer is initiated, and a backup express locker that meets the requirements is found within the service area and re-mapped. After all task packages have been processed, a pre-allocation scheme containing the correspondence between task packages and locked grids is formed. According to the pre-allocation scheme, a two-way communication link is established with the target express cabinet before the drone takes off, and periodic communication is carried out through the two-way communication link after the drone takes off, so as to maintain the lock of the compartment in the pre-allocation scheme. During flight, when the target parcel locker detects a physical failure of a locked compartment in the pre-allocation scheme, or when the edge computing module detects a communication link loss between the drone and the target parcel locker, a tiered redirection operation is triggered, and the pre-allocation scheme is updated. When the edge computing module detects that multiple drones simultaneously request the remaining slots of the same express delivery locker and there are not enough available slots, the edge computing module prioritizes the requesting drones based on multiple dimensions. Drones with higher priority are given priority for slot allocation, while drones with lower priority receive backup allocation schemes.
2. The method according to claim 1, characterized in that, Each entry in the grid state matrix also includes: the length of the drone queue waiting for delivery and the status of whether the docking platform is available; the step of performing grid pre-allocation one by one includes: First, group the task packages according to the target express lockers, and then sort the task packages in each group from largest to smallest according to the total number of required compartments. For each task package, query the available quantity of the corresponding compartment type of the target express cabinet in the compartment status matrix. If the condition is met, lock the compartment and deduct the corresponding available quantity from the compartment status matrix. If the condition is not met, record it as a task package to be transferred. After all normal locking is completed, the task packages to be transferred are arranged in ascending order of distance from the nearest backup parcel locker, and the transfer to neighboring lockers is carried out in sequence.
3. The method according to claim 2, characterized in that, The escalating redirection operation includes: The first level involves the target express delivery locker searching for and replacing a spare compartment of the same type within the locker itself, with a response latency of less than 100 milliseconds; The second level involves the edge computing module searching within the same site's service area for a backup parcel locker that meets the locker requirements and is closest to the drone's current location, and verifying that the new flight path is within the drone's remaining range. The third level involves the edge computing module querying the status of the parcel locker compartments within the service range of adjacent stations, and then transferring the task package to the unmanned vehicle at the adjacent station via vehicle-to-vehicle communication for relay delivery.
4. The method according to claim 3, characterized in that, The detection method for physical failure of the grid includes: using a triple redundant sensor array consisting of a photoelectric switch, a micro switch, and a motor current sensor deployed in each grid to detect in real time whether the grid has mechanical faults, is occupied by foreign objects, or has sensor malfunctions.
5. The method according to claim 3, characterized in that, The method for determining communication link disconnection includes: the drone sending heartbeat packets at a fixed frequency, the heartbeat packets containing the drone's current location, remaining battery power, and estimated arrival time; the target parcel locker refreshing the remaining lock time upon receiving the heartbeat packets, and determining communication disconnection and releasing the locker lock if it does not receive the heartbeat packets for two consecutive heartbeat sending cycles; the heartbeat packets are sent every 5 seconds, and the locker lock refresh time is 30 seconds.
6. The method according to claim 3, characterized in that, The multiple dimensions include remaining battery status, package delivery urgency, and availability of alternative parcel lockers. The priority ranking adopts a weighted scoring method, which assigns preset weights to the three dimensions of remaining battery status, package delivery urgency, and availability of alternative parcel lockers and then calculates a comprehensive priority score. The lower the remaining battery, the closer the package's delivery deadline is, and the farther away the alternative parcel locker is, the higher the priority.
7. The method according to claim 3, characterized in that, The unmanned vehicle transmits data to the smart express cabinet and the drone via a C-V2X PC5 direct communication interface.
8. A task allocation system for air-to-ground collaborative logistics distribution based on grid status mapping, used to execute the method according to any one of claims 1 to 7, characterized in that, include: The cloud task module (1) is configured to receive package order information, group packages by express cabinet and associate them with preset sites; An edge computing module (2) is deployed on the vehicle-mounted platform of the unmanned vehicle, communicates with the cloud task module (1), and communicates with the smart express cabinets and drone clusters within the service range; the edge computing module (2) includes: The compartment status query submodule (21) is configured to obtain the compartment status data of the smart express cabinet and construct the compartment status matrix; The grid pre-allocation submodule (22) is connected to the grid status query submodule (21) and is configured to perform grid pre-allocation and locking, as well as neighboring cabinet transfer, based on the grid status matrix, and form a pre-allocation scheme; The task allocation submodule (23), connected to the grid pre-allocation submodule (22), is configured to allocate task packages to the UAV and issue flight commands according to the pre-allocation scheme; The redirection decision submodule (24), connected to the grid pre-allocation submodule (22) and the task allocation submodule (23), is configured to perform a progressively upgraded redirection operation and update the pre-allocation scheme when a grid fails physically or a communication link is lost. The arbitration submodule (25), connected to the grid pre-allocation submodule (22) and the task allocation submodule (23), is configured to perform resource arbitration based on multiple dimensions during multi-machine competition; The drone cluster (3) is communicatively connected to the edge computing module (2). The drone cluster includes multiple drones, each of which is equipped with a positioning module, a communication module, and a package delivery mechanism. The intelligent express cabinet network (4) is connected to the edge computing module (2) and the drone cluster. The intelligent express cabinet network includes multiple intelligent express cabinets. Each intelligent express cabinet is equipped with a grid sensor array, a docking platform, a heartbeat communication module and a grid locking management module. The grid sensor array deploys photoelectric switches, micro switches and motor current sensors for each grid to achieve triple redundant grid status detection.
9. The system according to claim 8, characterized in that, The heartbeat communication module of each smart express cabinet in the smart express cabinet network is configured to: receive the heartbeat packet of the drone, parse the location, battery level and estimated arrival time information in it; send a lock refresh command to the locker lock management module; when a physical failure of a locked locker is detected, first perform locker replacement in this cabinet, and if it is not feasible, send a redirection request to the redirection decision submodule (24) of the edge computing module (2).
10. The system according to claim 8, characterized in that, The edge computing module (2) transmits data with the smart express cabinet network and the drone cluster through the C-V2X PC5 direct communication interface.
