Distributed information collaborative unmanned aerial vehicle distribution path planning method, device and medium

Abstract

This invention provides a distributed information-coordinated drone delivery path planning method, device, and medium. The method includes: a detection drone predicting risk information for a future target time period based on acquired perception data; the detection drone acquiring real-time motion status information of each delivery drone within its communication range, and calculating correlation parameters for each delivery drone based on the risk information and the real-time motion status information; the correlation parameters characterizing the spatiotemporal matching degree between the flight trajectory of the delivery drone within the target time period and the risk area corresponding to the risk information; and the detection drone determining a target delivery drone from among the delivery drones that needs to receive risk information based on the correlation parameters, and pushing the risk information to the target delivery drone, enabling the target delivery drone to adjust its delivery path based on the risk information. This achieves optimized resource allocation for a drone swarm while ensuring safe delivery.

Description

Technical Field

[0001] This invention relates to the field of control engineering technology, and in particular to a method, device and medium for distributed information collaboration in unmanned aerial vehicle (UAV) delivery path planning. Background Technology

[0002] With the rapid development of smart logistics in cities, multi-drone collaborative delivery has become an important technological direction for solving the last-mile delivery problem. In practical applications, delivery drones face highly dynamic urban environments, such as real-time changes in weather conditions, the random occurrence of temporary no-fly zones, and communication blind spots caused by building obstructions.

[0003] To address the complexities of flight environments, existing drone swarm delivery processes employ a homogenized swarm architecture. Each drone requires high-performance hardware with comprehensive perception, decision-making, and execution capabilities, consuming significant hardware resources. Optimizing resource allocation within drone swarms while ensuring safe delivery is a crucial issue that the industry urgently needs to address. Summary of the Invention

[0004] This invention provides a distributed information collaborative drone delivery route planning method, device, and medium to optimize the resource allocation of drone swarms while ensuring safe delivery.

[0005] This invention provides a distributed information-coordinated drone delivery route planning method, applied to a drone swarm, wherein the drone swarm includes at least one detection drone for environmental perception and at least one delivery drone for cargo delivery, comprising: The detector predicts risk information for a future target time period based on the acquired sensing data; The detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information; the correlation parameters are used to characterize the spatiotemporal matching degree between the flight trajectory of the delivery machine in the target time period and the risk area corresponding to the risk information. Based on the correlation parameters, the detector determines the target delivery machine from among the delivery machines that needs to receive risk information, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information. The detector is also used to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, to update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0006] According to a distributed information-coordinated drone delivery route planning method provided by the present invention, the process of determining the correlation parameters of the delivery drone includes: Based on the real-time motion status information of the delivery vehicle, predict the flight trajectory of the delivery vehicle within the target time period; Based on the degree of spatial intersection between the flight trajectory and the risk area corresponding to the risk information, a spatial correlation score is determined. Determine the predicted time period for the delivery machine to reach the risk area, and determine the time correlation score based on the degree of overlap between the predicted time period and the target time period; The spatial correlation score and the temporal correlation score are weighted and calculated to obtain the correlation parameter.

[0007] According to the distributed information-coordinated drone delivery route planning method provided by the present invention, the step of weighting the spatial correlation score and the temporal correlation score to obtain the correlation parameter includes: Based on the risk level information, the confidence level of the risk information, and the time decay factor, a weighting parameter is determined; the time decay factor is used to characterize the effectiveness of the risk information over time. Based on the weight parameters, the spatial correlation score and the temporal correlation score are weighted and calculated to obtain the correlation parameters.

[0008] According to a distributed information-coordinated drone delivery route planning method provided by the present invention, the process of determining the confidence level of the risk information includes: The detector sends a cross-verification request for the risk area to other detectors within its communication range; Receive the observation results of the other detectors for the risk area, and verify the authenticity of the observation results based on digital signature verification information to confirm that the observation results have been verified. Determine the statistical dispersion between the observation results of the probe on the risk area and the observation results of the other probes; A consistency score is generated based on the statistical dispersion, and the confidence level of the risk information is determined based on the consistency score.

[0009] According to a distributed information-coordinated drone delivery route planning method provided by the present invention, the process of determining the consistency score includes: Determine the mean and standard deviation of the observation data from the probe and the observation data sent by other probes, and determine the ratio of the standard deviation to the mean; The consistency score is determined based on the difference between the preset constant and the ratio.

[0010] According to the distributed information collaborative drone delivery path planning method provided by the present invention, the feedback information is at least one of the accuracy score, timeliness score, and effectiveness score of the risk information after the target delivery drone flies over the risk area; The feedback information is used to adjust the model parameters of the risk prediction model and the flight prediction model in the probe; the risk prediction model is used to predict risk information; the flight prediction model is used to predict the time period for the delivery aircraft to arrive at the risk area to which the risk information belongs.

[0011] According to a distributed information collaborative drone delivery route planning method provided by the present invention, the step of pushing the risk information to the target delivery drone includes: The detector uses a private key to digitally sign the risk information, generating risk information with a digital signature. The detector sends the risk information with a digital signature to the target delivery machine, so that the target delivery machine can verify the digital signature based on the detector's public key and obtain the risk information after successful verification.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the distributed information collaboration drone delivery path planning method as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the distributed information collaboration drone delivery path planning method as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the distributed information collaboration drone delivery path planning method as described above.

[0015] This invention provides a distributed information-coordinated drone delivery path planning method, device, and medium. By constructing a heterogeneous cluster including a detection drone and a delivery drone, it achieves a specialized division of labor between environmental perception and cargo delivery. This effectively avoids hardware resource redundancy caused by each drone in a homogeneous cluster needing to carry a full set of high-precision perception equipment, thus optimizing the configuration of drone cluster hardware resources. Based on this, the detection drone predicts risk information based on perception data and calculates correlation parameters representing the spatiotemporal matching degree between the flight trajectory and the risk area, combined with the real-time motion status of the delivery drone. This allows for targeted risk push notifications to the selected affected target delivery drones. This ensures that the target delivery drone can obtain timely risk warnings highly correlated with its future trajectory, giving it sufficient time to adjust its delivery path to avoid danger and effectively guaranteeing delivery safety. Simultaneously, targeted push notifications based on correlation parameters avoid the occupation of communication channels by a large amount of irrelevant information and the waste of onboard computing resources of the delivery drone, achieving refined management of communication and computing resources at the software level. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the distributed information collaboration-based drone delivery route planning method provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of the four-layer distributed collaborative network provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the working process of the detector provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the workflow of the delivery machine provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] With the rapid development of smart logistics in cities, multi-drone collaborative delivery has become an important technological direction for solving the last-mile delivery problem. In practical applications, delivery drones face highly dynamic urban environments, including real-time changes in weather conditions, the random occurrence of temporary no-fly zones, and communication blind spots caused by building obstructions. To address these challenges, relevant methods are mainly studied from three dimensions: environmental perception, path planning, and multi-drone collaboration.

[0024] In terms of environmental perception, relevant methods generally adopt a homogeneous cluster-based independent perception model. Each drone undertakes the same perception and decision-making tasks, lacking specialized division of labor. This homogeneous model leads to problems such as redundant configuration of sensor resources, perception range limited to the field of view of a single drone, and inability to utilize the detection results of other drones. More importantly, when multiple drones operate in close proximity, the environmental information independently acquired by each drone may differ or even contradict each other, and the relevant methods lack cross-validation and conflict resolution mechanisms for multi-source information, forming information silos.

[0025] In path planning, relevant methods mainly rely on static maps or historical data for obstacle avoidance. Optimization algorithms are used to generate paths that satisfy constraints. While these methods exhibit good convergence performance at the algorithm level, they are essentially offline planning methods based on prior information and lack the ability to respond in real-time to dynamic environmental information.

[0026] Regarding multi-aircraft collaboration, relevant methods primarily focus on task allocation and formation control. While considering the collaborative relationships between multiple aircraft, the essence of collaboration remains at the task level of load balancing or the spatial level of formation following, without establishing an intelligent collaboration mechanism at the information level. In particular, it lacks targeted delivery of detection information, intelligent sharing of flight experience, and closed-loop optimization capabilities based on actual results.

[0027] In terms of information security, the relevant methods lack a verification mechanism for the authenticity of information sources, making it difficult to prevent system performance degradation or security incidents caused by malicious nodes injecting false information.

[0028] Specifically, the shortcomings of the relevant methods include: The lack of heterogeneous specialization in drone operations leads to resource redundancy. Related methods employ a homogeneous cluster architecture, requiring each drone to be equipped with complete perception, decision-making, and execution capabilities, resulting in redundant resource allocation. When performing delivery missions, the flight trajectory of delivery drones is primarily determined by delivery demand, making it difficult to perform specialized path optimization for environmental detection, leading to low sensor utilization efficiency. The lack of specialized detection capabilities means that delivery drones often only perceive risks when approaching dangerous areas, resulting in extremely short response windows and insufficient safety.

[0029] The crude information transmission process leads to communication waste and information overload. Related methods commonly employ indiscriminate broadcasting or pre-set fixed subscription models during information transmission. Broadcasting results in a large amount of irrelevant information consuming communication bandwidth, requiring the receiver to manually determine the usefulness of the information, thus consuming valuable onboard computing resources. While fixed subscription models can reduce irrelevant information, they cannot adapt to dynamically changing environments and task requirements. Due to the lack of intelligent push mechanisms based on spatiotemporal correlation, the computational burden of information value assessment and transmission decisions rests entirely on the receiving end, resulting in low efficiency.

[0030] The weak fusion capability caused by multi-source information silos stems from the lack of information exchange mechanisms between detection nodes in existing methods. Environmental information acquired by each node is only used by itself or downstream nodes, creating information silos. When a decision-making node receives information describing similar areas from multiple sensing nodes, this information may be inconsistent or even contradictory. Existing methods typically employ simple weighted fusion methods without considering consistency checks between information. The lack of cross-validation mechanisms between sensing nodes allows misjudgments from a single node (sensor malfunction, measurement error, malicious attacks, etc.) to be directly transmitted to the decision-making node, affecting the accuracy of decisions. More seriously, the lack of authentication mechanisms allows malicious nodes to forge information and inject it into the system, posing a systemic security risk.

[0031] The lack of flight experience leads to repeated trial and error. The experience information accumulated by decision nodes during actual execution (deviations between forecasts and actual conditions, successful coping strategies, etc.) is of significant reference value to subsequent nodes. However, the relevant methods lack an experience-sharing mechanism between nodes. Each node needs to execute independently and cannot learn from the experience of other nodes. When the planned paths of multiple nodes may conflict, the relevant methods mainly rely on a central scheduler for conflict resolution, lacking a distributed negotiation mechanism between nodes and failing to consider potential deadlock or livelock issues during the negotiation process.

[0032] The lack of closed-loop learning leads to performance rigidity. In related methods, the detection and information push strategies of sensing nodes are often based on preset algorithm parameters and remain fixed during operation. Without a feedback channel from the decision-making node to the sensing node, the sensing node cannot know whether the information it provides is accurate, timely, or helpful to the decision-making node, and therefore cannot adjust its strategy according to actual results. This open-loop architecture results in rigid system performance, making it unable to adapt to environmental changes and the evolution of task requirements.

[0033] These shortcomings reinforce each other, creating a vicious cycle: resource redundancy leads to information overload, information silos result in weak integration, gaps in user experience lead to repeated trial and error, and the absence of a closed loop leads to performance stagnation. These problems are particularly prominent in scenarios involving collaborative operations of multiple detection and decision-making nodes, severely restricting the overall performance, security, and scalability of drone delivery systems.

[0034] To address the shortcomings of existing methods, this invention provides a distributed information-coordinated drone delivery path planning method, applied to drone swarms. The drone swarm includes at least one detection drone for environmental perception and at least one delivery drone for cargo delivery. Figure 1 This is a flowchart illustrating the distributed information collaboration-based drone delivery route planning method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following: Step 110: The detector predicts risk information for a future target time period based on the acquired sensing data; Step 120: The detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information; the correlation parameters are used to characterize the spatiotemporal matching degree between the flight trajectory of the delivery machine in the target time period and the risk area corresponding to the risk information. Step 130: Based on the correlation parameters, the detector determines the target delivery machine from among the delivery machines that needs to be pushed with risk information, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information; Step 140: The detector is further configured to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0035] Since the detector needs to perceive the environment, it can be equipped with high-performance sensing sensors; while the delivery machine is mainly used for cargo transportation, so it can be equipped with lightweight sensors.

[0036] Specifically, in step 110, the detector predicts risk information for a future target time period based on the acquired sensing data.

[0037] Specifically, when performing environmental reconnaissance missions, the probe aircraft acquires environmental data using onboard weather radar, obstacle recognition sensors, and other equipment. The probe aircraft can employ focused patrols or pre-emptive reconnaissance methods to obtain environmental conditions ahead of the delivery aircraft's flight path. It should be noted that a drone swarm can include multiple probe aircraft and multiple delivery aircraft. The probe aircraft can fly in multiple different directions ahead of the delivery aircraft.

[0038] After acquiring sensing data, the probe can predict the risks for a future target time period based on the risk prediction model deployed in the probe, thus obtaining risk information for the future target time period.

[0039] Risk information includes meteorological risks (such as strong air currents and wind shear), temporary no-fly zones, and dynamic or static obstacles. Risk information typically includes the geographical location of the risk, the risk level, and the confidence level of the information.

[0040] Optionally, to avoid detection blind spots, a distributed collaborative sensing network can be established among the detectors. By exchanging information on detection coverage and already detected areas, blind spots can be identified, and based on a comprehensive assessment of distance cost and remaining energy, detectors can be coordinated to go to the blind spots to perform fill-in detection.

[0041] In step 120, the detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information.

[0042] Specifically, during the execution of their tasks, the delivery drones periodically broadcast their lightweight, real-time motion status information, including position, speed, and direction. By listening to these broadcasts, the detection drone can identify and acquire the real-time motion status information of each delivery drone within its communication range.

[0043] The correlation parameter characterizes the spatiotemporal matching degree between the delivery aircraft's flight trajectory within the target time period and the risk area corresponding to the risk information. It can be used to assess which delivery aircraft is affected by the current risk information. By calculating the correlation parameter, the computational burden of information value assessment is shifted from the delivery aircraft to the detection aircraft.

[0044] The information owner (detector) completes the value assessment, and the information received by the receiver (delivery machine) is highly relevant and can be used directly, significantly reducing the information processing burden on the delivery machine. At the same time, the subsequent targeted push process based on relevance parameters can significantly reduce communication overhead, as each message is only sent to a few highly relevant delivery machines, and the amount of communication data can be greatly reduced compared to the broadcast mode.

[0045] In step 130, the detector determines the target delivery machine from among the delivery machines that needs to be pushed with risk information based on the correlation parameters, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information.

[0046] Specifically, the detector has a preset relevance threshold. When the parameter value corresponding to the relevance information of a certain delivery machine exceeds the relevance threshold, that delivery machine is identified as a target delivery machine. The detector only sends risk information to the target delivery machine, achieving targeted push.

[0047] Upon receiving highly relevant risk information, the target delivery aircraft can conduct a route assessment based on its own mission priority, remaining range, and other constraints. If the assessment indicates that the original route carries a high risk, the route will be replanned to avoid the risk area, or the flight strategy will be adjusted.

[0048] In step 140, the detector is also used to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, to update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0049] Based on the feedback information, the detector updates the model parameters of the prediction model deployed within it; the prediction model is used to predict risk information for future time periods based on the acquired sensing data. The detector can analyze the feedback data, identify problems in the strategy, and automatically adjust the model parameters using methods such as incremental learning.

[0050] Furthermore, during the calibration process, not only can model parameters be calibrated, but adjustments can also be made to sensors or algorithms based on specific problems reflected in the feedback information. Specifically, this can include: If the feedback information indicates a systematic bias, calibrate the sensor model or risk assessment algorithm.

[0051] If the feedback information is not timely enough (e.g., the push is too early or too late), the detector can adjust the prediction time window (e.g., from 60 seconds to 50 seconds, to adapt to scenarios with large speed fluctuations).

[0052] If the feedback information shows low validity (a large amount of useless information is pushed), the detector can increase the relevance threshold (e.g., from 0.30 to 0.35) to reduce the push of edge-related information and avoid disturbance. The relevance threshold is used in subsequent relevance parameter judgment processes.

[0053] Optionally, a four-layer distributed collaborative network can be constructed in a heterogeneous drone swarm to realize the interaction process between the probe and delivery drones. The constructed collaborative network structure can be as follows: Figure 2 The schematic diagram of the four-layer distributed collaborative network provided by this invention is shown.

[0054] Specifically, it includes: The first layer is a sensor network coordinated by detectors.

[0055] Based on the specialized division of labor between detection and delivery, the cluster is divided into detection drones specializing in environmental perception and delivery drones specializing in cargo delivery. Unlike the homogeneous clusters in related methods, this heterogeneous architecture achieves optimized resource allocation and specialized functions.

[0056] Based on this, a distributed collaborative sensing network is established among the detectors, including three functional modules: complementary sensing blind spots, multi-source cross-verification, and cryptographic authentication.

[0057] A mechanism for complementing blind spots in perception based on distributed negotiation: Each probe performs environmental awareness and maintains its own detection coverage description, periodically exchanging coverage and detected area information with neighboring probes. Blind spots not covered by any probe are identified through spatial topology analysis.

[0058] Key elements of a distributed negotiation mechanism include: Exchange of detection intentions: Each probe declares its planned target area for detection; Calculate the conflict degree: Identify conflicts where multiple probes intend to probe the same area; Determine the execution node: Based on a comprehensive evaluation of distance cost (path length to the target area) and remaining energy (energy surplus after completing the detection mission), determine the node most suitable for executing the detection mission; To avoid deadlock (multiple detectors waiting for each other's decisions) or livelock (repeated negotiations failing to converge) during the negotiation process, a timeout forced decision mechanism is introduced: if the negotiation fails to reach an agreement within a preset negotiation timeout threshold (e.g., 5 seconds), the detector closest to the target blind zone is forced to assume responsibility, and other detectors switch to the suboptimal target.

[0059] Multi-source cross-validation and credibility assessment based on statistical dispersion: When the detection ranges of multiple detectors overlap, they may observe the same area's environment. Therefore, a cross-validation mechanism incorporating distributed fault-tolerant principles is proposed: After acquiring environmental information for a certain area, the probe checks if other probes have also observed that area. If so, it actively requests the observation results from these probes for multi-source data comparison. A consistency score based on statistical dispersion is defined to quantify the degree of mutual verification among multiple observation results: the mean and standard deviation of the observations are calculated, and the basic form of the consistency score is: 1 - (standard deviation / mean).

[0060] To ensure the robustness of the formula in the low-risk region (mean close to 0), when the mean is less than a preset minimum threshold, the absolute standard deviation is used as a consistency measure to avoid numerical explosion. A consistency score close to 1 indicates high consistency, while a score close to 0 indicates significant dispersion.

[0061] For an information set with a consistency score below a preset consistency threshold, the deviation of each observation (standardized distance relative to the mean, weighted by confidence level) is calculated. Observations with a deviation exceeding a preset outlier threshold are identified as outliers. The preset outlier threshold is set to tolerate a maximum of [number missing] outliers. A number of malicious or faulty nodes (N being the number of probes participating in the verification) can improve the reliability of environmental perception.

[0062] Cryptographic mechanisms for ensuring the authenticity of information: To ensure the authenticity of information and prevent malicious attacks, the communication between the probes employs a cryptographic authentication and integrity verification mechanism. Each probe has a unique identifier and key pair, and the information sent includes an integrity verification code or digital signature. The receiver verifies the authenticity and integrity of the information source by checking the verification code or signature; information that fails verification is rejected.

[0063] The second layer involves the detection and delivery information flow. When the probe pushes information to the delivery machine, an intelligent relevance assessment and targeted push mechanism is established to shift the computational burden of information value assessment from the receiving end to the sending end, enabling the probe to proactively and accurately push information to the delivery machine based on its understanding of the delivery machine's needs.

[0064] Multidimensional correlation prediction model: The detector maintains a dynamically updated delivery vehicle status table by listening to the periodically broadcast lightweight real-time motion status information (position, speed, direction) from the delivery vehicles. When the detector acquires environmental risk information for a certain location, it performs a correlation prediction.

[0065] Relevance assessment includes the following steps: Based on the current motion status information of the delivery machine, a motion model (default uniform linear model) is used to predict its possible flight trajectory within a preset prediction time window (default 60 seconds); Calculate the degree of spatial intersection between the predicted trajectory and the risk area, and quantify it as spatial correlation. Predict the arrival time of delivery vehicles in high-risk areas and compare it with the information's validity period to quantify it as time relevance; Multidimensional correlation parameters are calculated by combining spatial correlation, temporal correlation, risk level, and information confidence.

[0066] The information is only pushed to delivery machines whose relevance parameters exceed a preset relevance threshold, achieving precise delivery. The push messages also employ cryptographic mechanisms to ensure the authenticity of their source; upon receiving the message, the delivery machine verifies the integrity check code to confirm the information's origin.

[0067] Synergistic effects of targeted push notifications: The information owner (detector) completes the value assessment, and the information received by the receiver (delivery machine) is highly relevant and can be used directly, significantly reducing the information processing burden on the delivery machine. At the same time, targeted push significantly reduces communication overhead, with each message only sent to a few highly relevant delivery machines, resulting in a substantial reduction in communication data volume compared to the broadcast mode.

[0068] Third layer: Delivery machine network: A distributed experience-sharing network and a path conflict resolution mechanism were established among the delivery machines, as detailed below: Intelligent sharing of experiences based on path relevance: The experience information accumulated by delivery aircraft during actual flights includes: flight path location, time, predicted risk, actual risk, deviation assessment, route selection, and actual effects. Sharing of this experience information is based on intelligent matching according to path relevance. Path relevance assessment considers the spatial overlap of the planned routes of two delivery aircraft (calculating the shortest distance or degree of intersection) and the temporal window overlap (comparing the difference in estimated arrival times). Experience information is shared only with delivery aircraft whose path relevance is higher than a preset path relevance threshold (which can be set to 0.40).

[0069] The value of experiential information decays over time. A time decay factor can be introduced for weighting to ensure that delivery machines prioritize the freshness of the experience. The decay coefficient can be adjusted according to the dynamics of the environment: a larger decay coefficient is used for scenarios with high dynamics (such as changing weather), and a smaller decay coefficient is used for scenarios with low dynamics (such as static obstacles).

[0070] Distributed path conflict negotiation mechanism: When the planned routes of multiple delivery vehicles may conflict in time and space, a distributed negotiation mechanism is used to resolve the conflict. The negotiation mechanism includes the following elements: Exchange route information: The delivery machine periodically broadcasts key points of its planned route and estimated arrival time; Conflict detection: Each delivery machine listens to the path information of other delivery machines and calculates the degree of intersection or proximity of the paths in time and space; Negotiation and adjustment plan: The relevant delivery aircraft will conduct point-to-point negotiations to determine which party will adjust the route or stagger the arrival time based on constraints such as mission priority, remaining range, and energy reserves. Forced decision after timeout: If no agreement is reached within the preset negotiation timeout threshold (which can be set to 5 seconds), the delivery machine with higher task priority will maintain the original path, while the delivery machine with lower priority will be forced to execute the alternative path.

[0071] Fourth layer: Delivery-detection feedback-driven adaptive optimization mechanism.

[0072] Establishing a closed-loop feedback channel from the delivery machine to the probe enables continuous learning and adaptive optimization, breaking the performance stagnation.

[0073] Multidimensional quantitative feedback on information quality: After using the information provided by the detector, the delivery machine conducts a multi-dimensional assessment of the information quality. The assessment dimensions include: Accuracy: The degree of deviation between predicted risk and actual risk; Timeliness: The degree to which the arrival time of information matches the actual time of demand; Effectiveness: The degree to which the information substantially helps in path optimization; The delivery machine generates an information quality feedback report and sends it to the probe machine that previously sent it the information. The feedback message uses a cryptographic mechanism to ensure its authenticity, and the probe machine accepts the feedback data after verifying the source of the feedback.

[0074] The probe's parameters are adaptively adjusted: After receiving feedback, the detector accumulates and analyzes multiple feedback data, and identifies problems in the strategy based on statistical analysis: If the accuracy feedback shows a systematic deviation between the forecast and the actual situation, the probe will calibrate the parameters of its sensor model or risk assessment algorithm. If the timely feedback indicates that the information push timing is inappropriate, the detection opportunity will adjust the prediction time window or time decay coefficient in the correlation assessment. If the effectiveness feedback indicates that the pushed information has limited value, the detection opportunity will increase the relevance threshold and reduce the push of low-value information.

[0075] The system automatically learns the optimal values ​​for parameters such as time windows and correlation thresholds based on historical feedback data, achieving adaptive adjustment. The adjustment process employs incremental learning (such as sliding window averaging or exponentially weighted moving average), and the learning rate can be dynamically adjusted according to the stability of the feedback data, avoiding system instability caused by drastic parameter fluctuations.

[0076] Demand-driven probe request mechanism: The delivery vehicle can also initiate proactive probe requests to the detection vehicles. When the delivery vehicle is uncertain about the environmental information of a certain area ahead, it can request a secondary, more precise probe from a nearby detection vehicle. The detection vehicle decides whether to respond based on its own status and the priority of the request. Both request and response messages employ cryptographic mechanisms to ensure communication security.

[0077] By constructing a heterogeneous specialized division of labor and a four-layer intelligent collaborative network, system performance has been improved in multiple dimensions compared to related methods.

[0078] From the perspective of resource allocation efficiency, heterogeneous specialization enables the large-scale amortization of sensor investment. A small number of probe aircraft are equipped with high-performance sensors dedicated to environmental detection, while a large number of delivery aircraft use lightweight sensors dedicated to cargo transportation. Compared to a homogeneous cluster where each delivery aircraft needs a complete sensor system, the overall sensor investment is significantly reduced. More importantly, the flight paths of probe aircraft can be optimized for environmental detection needs, allowing for pre-detection and focused patrols in high-risk areas, greatly improving sensor utilization efficiency. With the reduced sensor burden on delivery aircraft, payload capacity is released, significantly improving the delivery efficiency of a single aircraft and resulting in a significant increase in the overall carrying capacity of the system for the same fleet size.

[0079] From the perspective of information transmission efficiency, the targeted delivery mechanism based on motion trend prediction shifts the computational burden of information value assessment from the receiving end to the sending end. The detector can accurately identify the target delivery aircraft for each piece of environmental information, achieving targeted delivery rather than indiscriminate broadcasting. This mechanism brings two efficiency improvements: significantly improved communication bandwidth utilization, with each piece of information only sent to a few highly relevant delivery aircraft, avoiding a large amount of irrelevant information occupying channel resources; and a significantly reduced information processing burden on the delivery aircraft, as the received information is all related to its own flight path, eliminating the need to filter and judge a large amount of irrelevant information, allowing computational resources to be used more for core tasks such as path planning, resulting in a significant improvement in the overall system response speed.

[0080] From the perspective of environmental perception reliability, the multi-source information intelligent filtering and multi-dimensional fusion mechanism can effectively cope with heterogeneous information provided by multiple detectors. Through cross-validation mechanisms and cryptographic authentication mechanisms at the detector network layer, erroneous or malicious information can be automatically identified and eliminated. A consistency check algorithm based on statistical dispersion can quantitatively assess the degree of mutual verification between information, while an anomaly detection algorithm based on confidence weighting can accurately eliminate erroneous data. This series of technical measures enables the system to extract cross-validated and reliable environmental perception from raw, unverified multi-source observations. Even in the presence of sensor errors, communication noise, or malicious information, the accuracy of environmental assessment remains at a high level. Furthermore, the digital signature verification mechanism effectively prevents malicious nodes from forging information, significantly improving the system's security and reliability.

[0081] From the perspective of path decision-making safety, the advance reconnaissance by the probe aircraft and the path planning by the delivery aircraft form a collaborative mechanism of pre-detection-re-decision. Before actually entering the risk area, the delivery aircraft has already obtained real-time information about the environment ahead through the probe aircraft, allowing sufficient time and distance for path adjustments. Compared to the on-the-fly reconnaissance mode, this significantly advances the moment of danger perception and greatly extends the response window. Through the experience-sharing mechanism of the delivery aircraft network, subsequent delivery aircraft can learn from the flight experience of the preceding delivery aircraft, avoiding repeated trial and error. Based on the experience information, the accuracy of the probe aircraft's forecasts was verified in the field, allowing delivery aircraft to adopt forecast information with greater confidence or remain vigilant against forecasts with significant deviations. This experience transfer mechanism makes the path selection of subsequent flights more accurate. The path negotiation mechanism among delivery aircraft effectively avoids multi-aircraft path conflicts, and the negotiation process, after introducing a timeout-forced decision-making mechanism, can be reliably completed within a limited time, avoiding deadlock problems.

[0082] From a robustness and scalability perspective, the fully distributed architecture eliminates the risk of single points of failure. The failure of any single probe or delivery drone only affects its own functionality and will not paralyze the entire system. It possesses excellent degradation service capabilities: if a probe malfunctions and stops working, the area it is responsible for probing may temporarily lack real-time information, but the delivery drone can still continue operating based on information and historical data provided by other probes. Cryptographic mechanisms ensure that even if some nodes are compromised, malicious information will be identified and rejected by the signature verification mechanisms of other nodes. In terms of scalability, adding new drones only requires configuring key pairs and joining the state broadcast network to automatically coordinate with existing drones, without requiring system reconfiguration or parameter adjustments. The lightweight state-sharing mechanism keeps communication overhead within acceptable limits even when the number of drones is large, allowing the system to smoothly scale to a large-scale cluster.

[0083] From the perspective of continuous optimization capabilities, the probe can continuously optimize its strategy based on actual application results through the delivery-probe closed-loop feedback mechanism. This continuous learning capability ensures that system performance is not fixed at the initial configuration but continuously improves over time. Based on the multi-dimensional feedback from the delivery aircraft, the probe can automatically identify problems in its own strategy and adaptively adjust parameters. In the initial stage of operation, the push strategy is based on preset parameters. After accumulating feedback from a certain number of flight missions, the probe continuously improves the accuracy of the push strategy through adaptive learning. This continuous optimization capability enables the system to adapt to changes in the environment and the evolution of mission requirements, maintaining long-term high-performance operation.

[0084] In summary, through innovative mechanisms such as heterogeneous specialization, intelligent information collaboration, experience sharing, and closed-loop learning, while maintaining a fully distributed architecture, the system has achieved comprehensive improvements in resource allocation efficiency, information transmission efficiency, environmental awareness reliability, path decision security, system robustness, and continuous optimization capabilities, providing a systematic technical solution for the commercial application of drone delivery systems.

[0085] This invention provides a distributed information-coordinated drone delivery path planning method. By constructing a heterogeneous cluster including a detection drone and a delivery drone, it achieves a specialized division of labor between environmental perception and cargo delivery. This effectively avoids hardware resource redundancy caused by each drone in a homogeneous cluster needing to carry a full set of high-precision perception equipment, thus optimizing the configuration of drone cluster hardware resources. Based on this, the detection drone predicts risk information based on perception data and calculates correlation parameters representing the spatiotemporal matching degree between the flight trajectory and the risk area, combined with the real-time motion status of the delivery drone. This allows for targeted risk push notifications to the selected affected target delivery drones. This ensures that the target delivery drone can obtain timely risk warnings highly correlated with its future trajectory, giving it sufficient time to adjust its delivery path to avoid danger and effectively guaranteeing delivery safety. Simultaneously, targeted push notifications based on correlation parameters avoid the occupation of communication channels by a large amount of irrelevant information and the waste of onboard computing resources of the delivery drone, achieving refined management of communication and computing resources at the software level.

[0086] In one embodiment, the process of determining the relevant parameters of the delivery machine includes: Based on the real-time motion status information of the delivery vehicle, predict the flight trajectory of the delivery vehicle within the target time period; Based on the degree of spatial intersection between the flight trajectory and the risk area corresponding to the risk information, a spatial correlation score is determined. Determine the predicted time period for the delivery machine to reach the risk area, and determine the time correlation score based on the degree of overlap between the predicted time period and the target time period; The spatial correlation score and the temporal correlation score are weighted and calculated to obtain the correlation parameter.

[0087] Specifically, the probe can use a motion model (such as the default uniform linear motion model or a higher-order motion model based on historical trajectories) to predict the delivery aircraft's flight trajectory within a future target time period, based on the current position, speed, and flight direction of the delivery aircraft.

[0088] The detector calculates the minimum distance or overlap area between the predicted trajectory and the risk area. For example, if the predicted trajectory passes directly through the risk area, or the distance to the edge of the risk area is less than the preset safety radius (e.g., 30 meters), a higher spatial correlation score (e.g., 0.90) is assigned; if the distance is greater (e.g., 120 meters), a very low score (e.g., 0.05) is assigned.

[0089] In addition to spatial proximity, temporal synchronization must also be considered. The detector calculates the estimated arrival time of the delivery vehicle in the risk area. If this time falls within the effective lifecycle of the risk information (target time period), the time relevance is considered high. For example, if the risk is a gust of wind lasting one minute, and the delivery vehicle is expected to arrive in 15 seconds, the time relevance score is high; if the delivery vehicle is expected to arrive in 10 minutes, by which time the gust may have dissipated, the time relevance score is low.

[0090] The scores can be combined using multiplication or weighted summation. For example, the correlation parameter = spatial correlation score × temporal correlation score. Only when both the spatial and temporal dimensions are highly matched will the final correlation parameter be high.

[0091] In one embodiment, the weighted calculation of the spatial correlation score and the temporal correlation score to obtain the correlation parameter includes: Based on the risk level information, the confidence level of the risk information, and the time decay factor, a weighting parameter is determined; the time decay factor is used to characterize the effectiveness of the risk information over time. Based on the weight parameters, the spatial correlation score and the temporal correlation score are weighted and calculated to obtain the correlation parameters.

[0092] Weighting parameters are determined based on the risk level information, the confidence level of the risk information, and the time decay factor. The risk level information reflects the severity of the environmental threat, and the confidence level reflects the degree of confirmation the detection result has by the detection aircraft. The core function of introducing the time decay factor is to endow the system with the ability to perceive the timeliness of the dynamic environment and eliminate the interference of outdated risk information on path planning. In drone delivery scenarios, many risks (such as sudden gusts of wind and temporary moving obstacles) are highly transient. If decisions are made solely based on space and confidence level, outdated risk data may cause the delivery aircraft to unnecessarily avoid areas that have already been restored to safety, severely reducing delivery efficiency. Through the time decay factor, the weight of historical data is automatically reduced over time, ensuring that the decision-making model always focuses on the latest and most effective environmental conditions.

[0093] High-risk levels and high-confidence levels should be given higher transmission priority. The formula for calculating the relevance parameter can be expressed as: Relevance parameter = Spatial relevance × Temporal relevance × Risk level coefficient × Information confidence level × Time decay factor.

[0094] For example, if the probe's predicted trajectory is 18m away from the risk area (spatial correlation 0.90), and it is expected to arrive in 15 seconds (temporal correlation 0.86), with a risk level of 0.75, a confidence level of 0.95, and a time decay factor of 0.5, then the final weighted correlation parameter is: 0.90 × 0.86 × 0.75 × 0.95 × 0.5 ≈ 0.275. If this value is higher than a preset threshold (e.g., 0.20), a push notification will be triggered.

[0095] In one embodiment, the process of determining the confidence level of risk information includes: The detector sends a cross-verification request for the risk area to other detectors within its communication range; Receive the observation results of the other detectors for the risk area, and verify the authenticity of the observation results based on digital signature verification information to confirm that the observation results have been verified. Determine the statistical dispersion between the observation results of the probe on the risk area and the observation results of the other probes; A consistency score is generated based on the statistical dispersion, and the confidence level of the risk information is determined based on the consistency score.

[0096] When a detector (such as S1) detects a risk in a certain area, in order to prevent single-point sensor failure or misjudgment, S1 will check whether the detection range of other detectors (such as S2) also covers the area. If so, it will actively initiate interaction.

[0097] Upon receiving the request, S2 returns its observation data for the same area.

[0098] Specifically, before sending observation results, each delivery machine encrypts the data digest using its pre-set private key to generate a digital signature, and broadcasts the original observation data, digital signature, and device ID together. Upon receiving the data, the probe decrypts the signature using the sender's public key and compares the decrypted digest with a digest recalculated based on the received data. This process not only achieves identity authentication, ensuring that the data originates from legitimate network nodes and not external attackers, but also ensures data integrity, preventing information from being tampered with during transmission. Only observation results verified by the signature (i.e., from a genuine source and untampered) are included in the subsequent statistical dispersion calculation pool. This dual filtering mechanism of prior verification (true then accurate) effectively prevents malicious nodes from forging identities and injecting false data to disrupt the collaborative network, significantly improving security in open environments.

[0099] By determining the statistical dispersion between the observation results of the probe on the risk area and the observation results of other probes, the differences between multiple observation results can be quantified.

[0100] A consistency score is generated based on the statistical dispersion, and the confidence level of the risk information is determined based on the consistency score. The consistency score measures the degree of mutual verification of multi-source information. The higher the consistency, the higher the confidence level. The detector can weightedly fuse the initial confidence level with the consistency score to obtain the final improved confidence level.

[0101] Furthermore, after verifying the authenticity of the information source based on digital signatures, cross-validation based on statistical dispersion is performed, and abnormal observations are eliminated based on consistency scores, thereby ensuring the confidence level of risk information even in the presence of malicious nodes.

[0102] Furthermore, the process of determining the consistency score includes: Determine the mean and standard deviation of the observation data from the probe and the observation data sent by other probes, and determine the ratio of the standard deviation to the mean; The consistency score is determined based on the difference between the preset constant and the ratio.

[0103] The specific formula for calculating the consistency score is: Consistency Score = Preset Constant - (Standard Deviation / Mean). The preset constant can be set to 1. For example, the risk value for observation S1 is 0.78, and for observation S2 it is 0.72. Mean μ =0.75, standard deviation σ ≈0.04. Therefore, the consistency score = 1 - (0.04 / 0.75) ≈ 0.95. This indicates that the two are highly consistent and have extremely high credibility.

[0104] It should be noted that, in order to ensure the robustness of the formula in the low-risk region (where the mean is close to 0), when the mean is less than the preset minimum threshold, the absolute standard deviation can be directly used as a consistency measure.

[0105] Furthermore, anomaly detection can be performed based on this mechanism. For cases where the consistency score is below a threshold, the deviation (e.g., Z-score) of each observation is calculated. If the deviation of an observation from a certain detector (e.g., S5) exceeds the anomaly threshold, it is identified as an anomaly (possibly a sensor malfunction or a malicious node) and removed, thereby achieving distributed fault tolerance.

[0106] In one embodiment, the feedback information is at least one of an accuracy score, a timeliness score, and an effectiveness score of the risk information after the target delivery aircraft has flown over the risk area; The feedback information is used to adjust the model parameters of the risk prediction model and the flight prediction model in the probe; the risk prediction model is used to predict risk information; the flight prediction model is used to predict the time period for the delivery aircraft to arrive at the risk area to which the risk information belongs.

[0107] Accuracy score: reflects the degree of deviation between the predicted risk and the actual risk measured by the delivery machine.

[0108] Timeliness score: reflects whether the information arrives at the appropriate time (e.g., whether it is too late to avoid, or too early to make the information expire).

[0109] Effectiveness score: Reflects whether the information is substantially helpful for path optimization.

[0110] Based on the feedback information, the detector updates the model parameters of the prediction model deployed within it; the prediction model is used to predict risk information for future time periods based on the acquired sensing data. The detector can accumulate and analyze multiple sets of feedback data to identify problems in the strategy and automatically adjust parameters using methods such as incremental learning.

[0111] If feedback indicates a systematic bias, calibrate the sensor model or risk assessment algorithm.

[0112] If the feedback is not timely enough (e.g., the push is too early or too late), the detector can adjust the prediction time window (e.g., from 60 seconds to 50 seconds, to adapt to scenarios with large speed fluctuations).

[0113] If the feedback shows low information validity (a large amount of useless information is pushed), the detector can increase the relevance threshold (e.g., from 0.30 to 0.35) to reduce the push of edge-related information and avoid disturbance.

[0114] It should be noted that the models deployed in the probe aircraft include at least a risk prediction model and a flight prediction model. The probe aircraft utilizes posterior data (feedback information) generated after the delivery aircraft actually flies over or avoids risk areas to drive the optimization of the front-end perception and computing models. Specifically, for the risk prediction model, based on the accuracy score of the feedback (such as the deviation between the intensity of the actual risk encountered and the predicted value), incremental learning algorithms are used to adjust feature weights or threshold parameters, correcting the model's sensitivity to specific environmental features (such as sudden gusts of wind or obstacle movement patterns), thereby improving the accuracy of future risk identification. Simultaneously, for the flight prediction model, combined with the timeliness score of the feedback (such as the deviation between the actual arrival time and the prediction window), the model's dynamic parameters (such as the influence coefficient of load on speed) and environmental drag coefficient are corrected. This parameter tuning strategy not only corrects prediction errors in a single dimension but also resolves the discrepancies in time and space judgment between the perception end (probe aircraft) and the execution end (delivery aircraft), enabling the entire cluster to dynamically adapt to environmental changes as the number of missions increases.

[0115] In one embodiment, pushing the risk information to the target delivery machine includes: The detector uses a private key to digitally sign the risk information, generating risk information with a digital signature. The detector sends the risk information with a digital signature to the target delivery machine, so that the target delivery machine can verify the digital signature based on the detector's public key and obtain the risk information after successful verification.

[0116] Each drone node (detector and delivery drone) is pre-configured with a unique identifier and key pair (public and private keys). Before sending risk information, the detector uses its own private key to encrypt the message digest and generate a digital signature (such as using the ECDSA algorithm).

[0117] After receiving the information, the delivery machine uses the corresponding public key to decrypt the signature and verify its integrity. If the verification fails, it indicates that the information may have been tampered with or originated from a malicious node (such as a forged probe), and the delivery machine will directly refuse to receive the information. Similarly, the cross-verification requests between probes and the feedback information sent by the delivery machine to the probes mentioned in the above embodiments all use the same digital signature verification mechanism.

[0118] This invention also provides an execution flow of the distributed information collaboration-based drone delivery path planning method of this invention, specifically including the workflow of the probe and the workflow of the delivery drone. The workflow of the probe can be as follows: Figure 3 The schematic diagram of the detection machine's workflow provided by this invention shows that the delivery machine's workflow can be as follows: Figure 4The workflow diagram of the delivery machine provided by this invention is shown. The specific implementation process based on the detector and the delivery machine includes: In a city delivery scenario, a drone swarm consists of 5 probe drones {S1, S2, S3, S4, S5} and 3 delivery drones {D1, D2, D3}. The probe drones are equipped with weather radar and obstacle recognition sensors, while the delivery drones are equipped with basic positioning and communication modules. All drones are equipped with encrypted communication modules (using ECDSA digital signatures) and digital signature verification capabilities.

[0119] Collaborative sensing of a network of detectors: Detector S1 detected a region of strong airflow at location (200m, 100m), with a risk level of 0.78 and a confidence level of 0.85. Detector S2 also detected the same region at location (235m, 115m), but with a risk level of 0.72 and a confidence level of 0.80.

[0120] Cross-validation process: After acquiring the information, S1 checks if the detection range of any other nearby detectors overlaps with the area. If it finds that S2's detection range covers the area, it sends an information exchange request to S2 (the message contains S1's digital signature). After verifying S1's signature, S2 returns its observation results (along with S2's digital signature).

[0121] After S1 verifies the authenticity of S2's signature confirmation information, it calculates the consistency between the two observations: the mean. =0.75, standard deviation σ=0.04. Since the mean is greater than the minimum threshold ε (set to 0.05), the standard consistency scoring formula is used: ; If the consistency score is higher than the preset consistency threshold (0.70), S1 marks the information as highly credible and obtains a final risk assessment of about 0.75 based on the confidence weighted fusion. The confidence of the fused information is increased to 0.95.

[0122] Anomaly detection scenarios: Suppose that detector S5, due to sensor malfunction, reports a significantly lower risk value of 0.15 for the same area, with a confidence level of 0.50. When S1 collects and compares the observations from S1, S2, and S5: ; ; ; If the consistency score falls below the preset consistency threshold (0.70), anomaly detection is triggered. The weighted deviation of each observation is calculated: ; S5's Z-score significantly exceeded the preset outlier threshold (2.0), and it was identified as an outlier and removed. Ultimately, only the observations from S1 and S2 were used for fusion.

[0123] If the detector S6 reports a completely inconsistent risk value but fails digital signature verification, S1 directly rejects the information and does not participate in subsequent consistency calculations, thus ensuring the system's ability to defend against malicious attacks.

[0124] Precise delivery from the probe to the delivery vehicle: The probe S1 obtains the fused risk information (location (220m, 110m), risk 0.75, confidence level 0.95) and needs to decide which delivery machines to push it to.

[0125] Query the delivery machine status table to obtain the current status of D1: position (120m, 85m), speed (8m / s, 3m / s), and direction 20.6°.

[0126] Based on the uniform linear motion model, the trajectory of D1 within the prediction time window (60 seconds) is predicted. The minimum distance between the predicted trajectory and the risk area is calculated to be approximately 18m, which occurs after 15 seconds. This distance is less than the risk impact radius (30m), indicating a high spatial correlation (approximately 0.90).

[0127] Calculate the time matching degree: D1 is expected to arrive in 15 seconds, with a time correlation of approximately 0.86.

[0128] The calculated comprehensive correlation parameter = spatial correlation × temporal correlation × risk level coefficient × information confidence level = 0.90 × 0.86 × 0.75 × 0.95 = 0.55.

[0129] If the parameter value is higher than the preset correlation threshold (0.30), S1 pushes the information to D1 (the message contains S1's digital signature). D1 accepts the information after verifying the signature. Furthermore, a time decay factor can be considered when calculating the comprehensive correlation parameter.

[0130] For delivery aircraft D2, assuming its flight direction deviates significantly from the risk area, the minimum predicted trajectory distance is 120m, the spatial correlation is only about 0.05, and the comprehensive score is about 0.04, which is below the threshold, so no push is made.

[0131] Experience sharing and conflict resolution in delivery machine networks: Delivery machine D1 successfully bypassed the risk area by following the replanned path and generated an experience information unit: location (220m, 110m), time T=35s, predicted risk 0.75, actual risk 0.80, deviation +0.05, detour path adopted, and the path effect is good.

[0132] D1 queries the planned routes of other delivery machines and finds that D3's route will also pass through the area, and it is expected to arrive in 10 minutes. Path correlation is calculated: the spatial overlap is high, the time windows are similar, and the correlation parameter is approximately 0.65, which is higher than the preset path correlation threshold (0.40).

[0133] D1 sends the experience information to D3 (containing D1's digital signature). After verifying the signature, D3 accepts the experience information and refers to D1's experience when planning the route: since D1 verified that the predicted risk was basically accurate (deviation only +0.05) and the detour route was feasible, D3 adopts a similar detour strategy to avoid repeated risk assessment and trial and error.

[0134] Additionally, assuming the planned paths of delivery machines D1 and D2 intersect in a certain airspace (180m, 120m), with estimated arrival times of T=45s and T=47s respectively, a time difference of only 2 seconds, there is a risk of collision. The system detects the potential conflict, and D1 and D2 enter a negotiation process.

[0135] Negotiated inputs: Task priority D1 is high (emergency delivery), remaining range 120m; Task priority D2 is medium, remaining range 150m. Constraints: Avoid collisions, minimize path adjustment costs.

[0136] Negotiation process: D1 and D2 exchange their priorities and path adjustment costs. Based on the priority rules, D1 maintains its original path and speed, while D2 chooses to reduce its flight speed by 5%, delaying its arrival time to T=49.5s to ensure it avoids D1's arrival time.

[0137] If an agreement is reached within 3 seconds, D2 will execute the adjusted path. If no agreement is reached within the preset negotiation timeout threshold (5 seconds), a forced timeout decision will be initiated: D1, as the high priority party, will maintain the original path, while D2 will be forced to execute the alternative path (a detour, increasing the flight distance by approximately 20m).

[0138] Feedback and adaptive adjustments from the delivery vehicle to the detector: After delivery aircraft D1 completes its flight, it conducts a multi-dimensional quality assessment of the information provided by probe aircraft S1: Accuracy score: 0.95 (predicted 0.75, actual 0.80, small deviation); Timeliness score: 0.90 (Information arrives at the right time, with sufficient time to respond); Usefulness score: 0.85 (The information helped optimize the path and avoid risks); D1 generates a feedback report and sends it to S1 (containing D1's digital signature). S1 verifies the signature and then accepts the feedback data.

[0139] Adaptive adjustment of the probe: After receiving feedback from 30 flight missions, S1 analyzed the statistical data: Average accuracy score: 0.92 (sensors and risk assessment models are relatively accurate); Timeliness average score: 0.85 (some feedback indicates that the push notification was delivered slightly earlier or later); Average usefulness score: 0.80 (Most of the information is useful, but there is still room for improvement); Based on feedback data, S1 identified patterns of insufficient timeliness: when the actual flight speed of the delivery aircraft is higher than the predicted speed, the information is pushed out too early; when the delivery aircraft encounters temporary obstacles and slows down, the information is pushed out too late.

[0140] S1 execution parameters are adaptively adjusted: The prediction time window was changed from a fixed 60 seconds to a dynamic window: based on the historical speed variance of the delivery machine, a 60-second window was used for delivery machines with stable speeds, and a 50-second window was used for delivery machines with large speed fluctuations. The relevance threshold was slightly increased from 0.30 to 0.35 to reduce the push of marginal relevance information.

[0141] It should be noted that for embedded platforms with limited computing resources, a simplified verification model can be used for cross-validation. When the differences in observations of the same area by multiple probes are within a preset difference threshold, the observations are considered consistent, and the weighted average is taken as the fusion result. If the difference exceeds the preset difference threshold, it is marked as questionable, and more probes are requested to supplement the observations or on-site verification by a delivery probe is required. Although this simplified model compromises on theoretical rigor, it significantly reduces computational complexity and is suitable for resource-constrained application scenarios.

[0142] In risk field modeling, a hybrid geometric approach can be employed to more accurately describe different types of risks. For continuous diffusion phenomena such as meteorological risks, a Gaussian distribution is used for spatial modeling, while discrete risks such as obstacles are represented by polygonal or rectangular bounding boxes. The appropriate geometric representation is automatically selected based on the risk type: a Gaussian distribution is used for meteorological risks, polygons for static obstacles, and circular buffer zones for temporary no-fly zones. This hybrid modeling method can more accurately characterize the spatial distribution features of different types of risks, improving the accuracy of path planning.

[0143] Regarding the information delivery mechanism, a hybrid model combining push and subscription can be adopted. At the start of the mission, the delivery aircraft publishes subscription requests based on the areas of interest along its flight path, specifying the range of areas of interest and the type of risk. The detection aircraft maintains a subscription table and proactively pushes information to subscribers when it detects information in that area, while also retaining a correlation-based proactive push mechanism to handle sudden or urgent information. This hybrid model combines the predictability of the subscription mechanism with the flexibility of proactive push, making it suitable for application scenarios with a wide range of dynamic environmental changes.

[0144] Regarding communication mechanisms, a layered communication mechanism can be adopted to avoid channel congestion in high-density drone operation scenarios. Status broadcasts and risk alerts can use different communication frequency bands or time slots, or status broadcasts can use low-frequency, high-penetration channels to ensure global coverage, while risk alerts can use high-frequency, high-bandwidth channels to support rapid transmission of detailed information. This layered design effectively avoids mutual interference between different types of communication, improving the overall communication efficiency of the system.

[0145] In terms of computing architecture, for drone platforms with extremely limited computing resources, some computationally intensive tasks can be offloaded to ground-based edge computing nodes. The probe sends raw risk data and delivery vehicle status data to the edge nodes, which then perform computationally intensive tasks such as correlation assessment and trajectory prediction, and return the results to the probe. This cloud-edge collaborative architecture can overcome the limitations of single-machine computing power and support more complex algorithm models, but the impact of communication latency on real-time performance needs to be considered.

[0146] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a distributed information-coordinated drone delivery path planning method, which includes: the detector predicting risk information for a future target time period based on acquired perception data; The detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information; the correlation parameters are used to characterize the spatiotemporal matching degree between the flight trajectory of the delivery machine in the target time period and the risk area corresponding to the risk information. Based on the correlation parameters, the detector determines the target delivery machine from among the delivery machines that needs to receive risk information, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information. The detector is also used to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, to update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0147] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, the computer program being executed by a processor, the computer being able to execute the distributed information collaborative drone delivery path planning method provided by the above methods, the method including: the detector predicting risk information for a future target time period based on the acquired perception data; The detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information; the correlation parameters are used to characterize the spatiotemporal matching degree between the flight trajectory of the delivery machine in the target time period and the risk area corresponding to the risk information. Based on the correlation parameters, the detector determines the target delivery machine from among the delivery machines that needs to receive risk information, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information. The detector is also used to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, to update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0149] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a distributed information-coordinated drone delivery path planning method provided by the above methods, the method comprising: the detector predicting risk information for a future target time period based on acquired perception data; The detector acquires real-time motion status information of each delivery machine within the communication range, and calculates the correlation parameters of each delivery machine based on the risk information and the real-time motion status information; the correlation parameters are used to characterize the spatiotemporal matching degree between the flight trajectory of the delivery machine in the target time period and the risk area corresponding to the risk information. Based on the correlation parameters, the detector determines the target delivery machine from among the delivery machines that needs to receive risk information, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts its delivery route based on the risk information. The detector is also used to receive feedback information from the target delivery aircraft after it flies over the risk area, and based on the feedback information, to update the model parameters of the prediction model deployed in the detector using an incremental learning algorithm, so as to adjust the risk prediction process of the detector for the next time period of the target time period.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for planning a delivery path of a UAV by distributed information coordination, applied to a UAV cluster, the UAV cluster comprising at least one probe UAV for environment perception and at least one delivery UAV for cargo delivery, characterized in that, The method comprises the following steps: The detection machine predicts risk information of a future target time period based on the obtained perception data; The detection machine obtains real-time motion state information of each delivery machine within a communication range, and calculates a correlation parameter of each delivery machine based on the risk information and the real-time motion state information; the correlation parameter is used to represent a degree of spatiotemporal matching between a flight trajectory of a delivery machine in the target time period and a risk region corresponding to the risk information; The detection machine determines a target delivery machine that needs to be pushed with risk information from the delivery machines based on the correlation parameter, and pushes the risk information to the target delivery machine, so that the target delivery machine adjusts a delivery path based on the risk information; The detection machine is also used to receive feedback information of the target delivery machine after flying through the risk region, and update model parameters of a prediction model deployed in the detection machine using an incremental learning algorithm based on the feedback information, so as to adjust a risk prediction process of the detection machine for a next time period of the target time period. 2.The method of claim 1, wherein, The determination process of the correlation parameter of the delivery machine comprises the following steps: predicting a flight trajectory of the delivery machine in the target time period based on real-time motion state information of the delivery machine; determining a spatial correlation score based on a degree of spatial intersection between the flight trajectory and a risk region corresponding to the risk information; determining a predicted time period for the delivery machine to arrive at the risk region, and determining a time correlation score based on an overlapping degree of the predicted time period and the target time period; performing weighted calculation on the spatial correlation score and the time correlation score to obtain the correlation parameter. 3.The method of claim 2, wherein, The weighted calculation on the spatial correlation score and the time correlation score to obtain the correlation parameter comprises the following steps: determining a weight parameter based on risk level information of the risk information, a confidence of the risk information, and a time decay factor; the time decay factor is used to represent an effective degree of the risk information over time; performing weighted calculation on the spatial correlation score and the time correlation score based on the weight parameter to obtain the correlation parameter. 4.The method of claim 3, wherein, The determination process of the confidence of the risk information comprises the following steps: The detection machine sends a cross-validation request of the risk region to other detection machines within a communication range; receiving observation results of the risk region returned by the other detection machines, and verifying authenticity of the observation results based on digital signature verification information to confirm that the observation results pass the verification; determining a statistical dispersion between observation results of the risk region by the detection machine and observation results of the other detection machines; generating a consistency score based on the statistical dispersion, and determining the confidence of the risk information based on the consistency score. 5.The method of claim 4, wherein, The determination process of the consistency score comprises the following steps: determining a mean value and a standard deviation of observation data of the detection machine and observation data sent by the other detection machines, and determining a ratio of the standard deviation to the mean value; determining the consistency score based on a difference between a preset constant and the ratio. 6.The method of claim 2, wherein, The feedback information is at least one of an accuracy score, a timeliness score, and an effectiveness score of the risk information after the target delivery drone flies through the risk area; The feedback information is used to adjust model parameters of a risk prediction model and model parameters of a flight prediction model in the probe drone; the risk prediction model is used to predict risk information; and the flight prediction model is used to predict a predicted time period for the delivery drone to reach a risk area to which the risk information belongs.

7. The method of claim 1, wherein, The pushing of the risk information to the target delivery drone comprises: The probe drone digitally signs the risk information based on a private key to generate risk information with a digital signature; The probe drone sends the risk information with the digital signature to the target delivery drone, so that the target delivery drone verifies the digital signature based on a public key of the probe drone and acquires the risk information after verification.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to implement the method for planning a delivery path of a drone in a distributed information coordination manner as claimed in any one of claims 1 to 7. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method for planning a delivery path of a drone in a distributed information coordination manner as claimed in any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method for planning a delivery path of a drone in a distributed information coordination manner as claimed in any one of claims 1 to 7.

Images