Wide-area reconnaissance and hovering pinpointing coordination method and system for fixed-wing multi-rotor drop

Abstract

This application relates to the field of unmanned aerial vehicle (UAV) control technology, and particularly to a collaborative method and system for wide-area reconnaissance and hovering detailed investigation of fixed-wing UAVs deployed to multi-rotor aircraft. The proposed scheme involves a master UAV performing a wide-area reconnaissance mission, generating detailed investigation tasks based on wide-area clues and encoding them into multiple task seeds. Each task seed contains task skeleton information and rule information. A complementary codebook maps the task seeds to complementary evidence collection combinations, and wingman capabilities are combined to form task wingman allocation. Wingmen pre-compile tasks before deployment and, after deployment, unfold and execute hovering detailed investigation based on rules. This application enables complementary evidence collection for detailed investigation under multi-wingman parallel conditions, reducing redundancy and improving resource utilization efficiency and mission execution reliability.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for wide-area reconnaissance and hovering precision reconnaissance collaboration for fixed-wing deployment of multi-rotor drones. Background Technology

[0002] In the field of UAV collaborative reconnaissance, existing technologies typically employ fixed-wing UAVs for large-scale searches and multi-rotor UAVs for detailed local reconnaissance, aiming to balance coverage efficiency and observation accuracy. However, in practical applications, this collaborative approach still has shortcomings in terms of mission organization and execution effectiveness.

[0003] On the one hand, existing technologies, after triggering detailed investigation tasks in wide-area reconnaissance, usually directly assign detailed investigation tasks to multi-rotor UAVs in the form of static coordinates, fixed tracks, or a single execution script. This makes it difficult to form differentiated task expressions for different detailed investigation needs, resulting in highly similar observation data collected by multiple multi-rotor UAVs during the execution process. There is obvious redundancy in detailed investigation evidence, while evidence within key observation angles, altitude layers, or time windows is missing, affecting the completeness and reliability of the final judgment results.

[0004] On the other hand, existing technologies generally rely on single indicators such as distance, remaining battery power, or payload type as the main basis for the selection and allocation of multi-rotor drones, lacking overall constraints on the combination of detailed evidence. This makes it difficult to ensure that the execution behavior of different multi-rotor drones complements each other in multi-task parallel scenarios, resulting in low overall resource utilization efficiency and limited mission completion quality.

[0005] To address the above issues, this application presents a collaborative method and system for wide-area reconnaissance and hovering precision reconnaissance using fixed-wing aircraft deploying multi-rotor drones. Summary of the Invention

[0006] The technical problem to be solved by this application is to address the shortcomings of existing technologies by providing a collaborative method and system for wide-area reconnaissance and hovering detailed investigation of fixed-wing aircraft deploying multi-rotor aircraft. The mother aircraft performs a wide-area reconnaissance mission, generates detailed investigation missions based on wide-area clues and encodes them into multiple mission seeds. The mission seeds contain mission skeleton information and rule information. The mission seeds are mapped to complementary evidence collection combinations through a complementary codebook, and mission wingman allocation is formed by combining wingman capabilities. The wingman completes mission pre-compilation before deployment and unfolds and performs hovering detailed investigation based on rules after deployment.

[0007] To achieve the above objectives, this application provides the following technical solution:

[0008] A method for coordinated wide-area reconnaissance and hovering detailed reconnaissance using fixed-wing aircraft to deploy multi-rotor drones is applied to an airborne homeport system comprising a mother aircraft and multiple wingmen operating around the mother aircraft. The mother aircraft corresponds to a fixed-wing UAV, and the wingmen correspond to multi-rotor UAVs. The mother aircraft performs wide-area reconnaissance missions and loads detailed reconnaissance missions to be performed onto the multiple wingmen. The method includes:

[0009] The detailed investigation task is encoded to generate multiple task seeds, wherein the task seeds include task skeleton information for describing the detailed investigation task and rule information for expanding the task skeleton information into detailed investigation actions.

[0010] Based on the multiple task seeds, select the corresponding task wingman and output a preprocessing completion indication to the master machine;

[0011] Upon receiving the preprocessing completion instruction, the mission wingman is deployed.

[0012] The aforementioned wide-area reconnaissance mission includes:

[0013] Acquire wide-area reconnaissance data, wherein the wide-area reconnaissance data includes at least one of the image data, video data, radar data, and infrared data acquired by the sensors mounted on the mothership;

[0014] The wide-area reconnaissance data is preprocessed to obtain corresponding wide-area clue data, wherein the preprocessing includes time synchronization, geographic registration and target area segmentation;

[0015] Based on the wide-area clue data, at least one suspected target clue is identified, and a detailed investigation task to be executed is generated for the suspected target clue. The identification of at least one suspected target clue includes: extracting suspected target candidates from the wide-area clue data through a preset target classification model; performing multi-frame association on the suspected target candidates to obtain a suspected target set; generating a corresponding suspected target clue for each suspected target in the suspected target set. The suspected target clue includes the target center position, position uncertainty parameters, and target motion hypothesis parameters. The target classification model is constructed using historical training samples.

[0016] The detailed investigation task is coded to generate multiple task seeds, including:

[0017] The suspected target clues corresponding to the detailed investigation task are converted into a task parameter set, wherein the task parameter set includes detailed investigation area range parameters, detailed investigation time window parameters, target prior category parameters, target motion constraint parameters, and evidence collection requirement parameters;

[0018] Based on the set of task parameters, a task skeleton information describing the detailed investigation task is generated by a task state machine generation algorithm. The task skeleton information includes at least one task node and the transition relationship between task nodes. The task node includes at least one of an entry node, a search node, a confirmation node, a tracking node, and an exit node. Each task node is associated with corresponding spatial constraint parameters, height constraint parameters, and speed constraint parameters.

[0019] Based on the task parameter set, rule information is generated to expand the task skeleton information into detailed investigation actions, wherein the rule information includes a mapping relationship between trigger conditions and action templates;

[0020] The task parameter set is divided according to a preset task differentiation strategy, and corresponding task skeleton information and rule information are assigned according to the division results. The task differentiation strategy includes observation angle differentiation, height layer differentiation, sensor mode differentiation and time window differentiation.

[0021] The suspected target clues corresponding to the detailed investigation task are converted into a set of task parameters, including:

[0022] Elliptic thresholding is performed on the target center position and position uncertainty parameters in the suspected target clues to determine the range parameters of the detailed investigation area;

[0023] Extrapolate the motion assumption parameters of the suspected target clues in the time domain to determine the detailed investigation time window parameters;

[0024] The confidence level of the prior category parameters of the suspected target clues is normalized to determine the parameters required for evidence collection.

[0025] Based on the multiple task seeds, select the corresponding task wingman, including:

[0026] Obtain a complementary codebook, wherein the complementary codebook includes multiple codewords, each codeword being used to represent a set of detailed evidence collection combinations, the detailed evidence collection combinations including the combination requirements of observation angle, height layer, sensor mode and time window;

[0027] For each task seed, the corresponding target codeword is determined in the complementary codebook based on the task difference information corresponding to the task seed. The target codewords corresponding to different task seeds are complementary to each other in the dimensions corresponding to the observation angle, height layer, sensor mode and time window.

[0028] Obtain the set of support codes for each wingman, wherein the support codes are used to characterize the combination of detailed evidence collection corresponding to the codes that the wingman can execute under the preset constraints of load capacity, flight performance and mission preparation delay.

[0029] Based on the target codeword and the set of supporting codewords, a feasible pairing relationship between task seeds and wingmen is determined by a preset wingman allocation algorithm, thereby obtaining the task wingman corresponding to each task seed.

[0030] The complementary codebook is constructed in the following ways:

[0031] Based on historical data, a discrete set of evidence dimensions is determined to characterize the combination of evidence collected in the detailed investigation. The discrete set of evidence dimensions includes a discrete set of observation angles, a discrete set of height layers, a discrete set of sensor modes, and a discrete set of time windows.

[0032] A candidate codeword set is generated based on the discrete set of evidence dimensions, wherein each candidate codeword is used to represent at least one discrete value selected from the discrete set of observation angles, the discrete set of height layers, the discrete set of sensor modes, and the discrete set of time windows.

[0033] The similarity of the candidate codewords in the candidate codeword set is calculated for each pair of candidate codewords. The similarity of the codewords is used to characterize the degree of overlap of the detailed evidence collection combination corresponding to the two candidate codewords in terms of observation angle, height layer, sensor mode and time window.

[0034] A complementary codebook is obtained by filtering from the candidate codeword set according to a preset similarity threshold. The filtering includes: determining candidate codewords to be selected one by one according to a preset codeword generation order; discarding the candidate codeword when the codeword similarity between the candidate codeword and any selected codeword in the complementary codebook is greater than the similarity threshold; and adding the candidate codeword to the complementary codebook when the codeword similarity between the candidate codeword and each selected codeword in the complementary codebook is less than or equal to the similarity threshold.

[0035] The step of determining the corresponding target codeword in the complementary codebook based on the task difference information corresponding to the task seed includes:

[0036] Obtain the set of task parameters associated with the task seed, and extract task differential labels from the set of task parameters. The task differential labels include observation angle differential labels, height layer differential labels, sensor mode differential labels, and time window differential labels.

[0037] The dimension value combination of the target codeword is determined based on the task differential label, wherein the dimension value combination includes: the target observation angle value determined by the observation angle differential label, the target height layer value determined by the height layer differential label, the target sensor mode value determined by the sensor mode differential label, and the target time window value determined by the time window differential label.

[0038] Search the complementary codebook for codewords that match the combination of the dimension values ​​to determine the corresponding target codeword.

[0039] The step of determining feasible pairings between task seeds and wingmen through a preset wingman allocation algorithm to obtain the task wingman corresponding to each task seed includes:

[0040] Based on the target codeword and the support codeword set, a candidate pairing set between the task seed and the wingman is established. When the support codeword set of the wingman includes the target codeword of the task seed, the corresponding wingman is determined as a candidate wingman of the task seed, and the corresponding candidate pairing relationship is generated.

[0041] The pairing cost is calculated based on the candidate pairing relationship, wherein the pairing cost is used to characterize the resource consumption and task risk when the corresponding wingman executes the corresponding task seed, and the pairing cost includes energy cost, time cost and communication cost;

[0042] Based on the candidate pairing relationships and the corresponding pairing costs, the initial pairing results are determined under the condition of satisfying the wingman resource constraints, wherein the wingman resource constraints include the wingman available energy threshold, the available time window threshold, and the available communication bandwidth threshold.

[0043] The initial pairing results are subjected to complementarity verification, and the initial pairing results are subjected to conflict resolution to obtain the final pairing results. The conflict resolution includes: replacing the conflicting task seed with another candidate wingman until the final pairing results satisfy the complementarity verification.

[0044] Output the final pairing result as the wingman for each task seed.

[0045] After the mission wingman is deployed, the method further includes:

[0046] The task seed is activated when the deployment trigger condition is detected to be met, wherein the deployment trigger condition includes off-board detection and the distance to the mother machine reaching a preset distance threshold.

[0047] After activating the task seed, the task wingman expands the task skeleton information according to the rule information to generate candidate detailed investigation plans.

[0048] Based on the local environmental information and wingman status information acquired by the sensors, the candidate detailed inspection schemes are constrained and screened to determine the target detailed inspection scheme, and the hovering detailed inspection task is executed according to the target detailed inspection scheme.

[0049] A wide-area reconnaissance and hovering precision reconnaissance collaborative system for deploying multi-rotor aircraft from a fixed-wing aircraft, the system comprising:

[0050] The task generation module is used to execute wide-area reconnaissance tasks and acquire wide-area reconnaissance data, determine suspected target clues based on the wide-area reconnaissance data and generate detailed investigation tasks, and encode the detailed investigation tasks to generate multiple task seeds.

[0051] The complementary allocation module is used to obtain the complementary codebook and, based on the target codewords corresponding to each task seed and the support codeword set corresponding to each wingman, execute the wingman allocation algorithm to determine the feasible pairing relationship between the task seed and the wingman.

[0052] The task execution module is used to load the corresponding task seed into the selected task wingman and deploy the task wingman; the task wingman is used to pre-compile the task seed before deployment to generate a task execution structure to be activated, and to activate the task execution structure after deployment to execute the hovering detailed investigation task and send the detailed investigation results back to the host machine.

[0053] Compared with the prior art, the beneficial effects of this application are:

[0054] This application encodes detailed investigation tasks into differentially divisible task seeds and introduces complementary codebooks to constrain the evidence collection combinations of multiple wingmen. This enables parallel multi-rotor UAVs to complement each other in dimensions such as observation angle, altitude layer, sensor mode, and time window, avoiding data redundancy and missing key evidence. Simultaneously, by utilizing the preparation window of the wingmen located within the mother aircraft, task loading and pre-compilation are completed before deployment, improving task executability and response timeliness. Without increasing execution complexity, this enhances the completeness of detailed investigation evidence and resource utilization efficiency, thereby strengthening the stability and reliability of wide-area reconnaissance and hovering detailed investigation collaborative tasks. Attached Figure Description

[0055] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0056] Figure 1 An exemplary application scenario diagram provided for an embodiment of this application;

[0057] Figure 2 A schematic diagram of the modules of the wide-area reconnaissance and hovering detailed reconnaissance collaborative system for fixed-wing deployment of multi-rotors provided in the embodiments of this application;

[0058] Figure 3 This is a flowchart illustrating the collaborative method for wide-area reconnaissance and hovering detailed reconnaissance of a fixed-wing multi-rotor aircraft provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0060] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] The method for coordinated wide-area reconnaissance and hovering detailed reconnaissance of fixed-wing multi-rotor aircraft deployed in this application is applicable to an aerial homeport-type UAV system.

[0062] A fixed-wing mothership with long endurance and high-speed coverage capabilities undertakes wide-area reconnaissance and situational awareness missions. Multiple multi-rotor wingmen are housed in cavities inside the mothership during the early stages of the mission and maintain communication, standby, and mission preparation around the mothership during flight. When the mothership discovers suspected target clues or areas to be confirmed during wide-area reconnaissance, it loads the corresponding detailed investigation tasks onto selected wingmen and deploys them. The wingmen then conduct detailed investigations and evidence collection in local airspace using hovering, low-speed maneuvers, and other methods, and transmit the results back to the mothership to achieve a closed-loop system for the clues.

[0063] This system is particularly suitable for application scenarios with dense ground targets, significant terrain obstruction, varied target shapes, and the need for coverage before confirmation, including but not limited to:

[0064] It can be used for long-distance linear facility inspection and anomaly verification, wide-area border / coastline patrol and close-range identification of suspicious targets, rapid situational awareness acquisition at disaster sites and precise search and rescue at key locations; it can also be applied to any mission organization scenario where stable hovering observation is difficult to achieve due to the flight attitude and speed constraints of fixed-wing platforms, while multi-rotor platforms are limited by endurance and communication conditions and cannot independently complete large-scale searches.

[0065] Understandably, the challenges faced by the air homeport system in engineering applications are not just about how to deploy wingmen, but more importantly, how to transform broad-area clues into a set of executable, complementary, and fusion-based detailed investigation actions within a controllable time window under conditions of multiple tasks and multiple wingmen operating in parallel.

[0066] Wide-area reconnaissance data acquired by fixed-wing mothership aircraft from an external perspective is typically characterized by rapid coverage but insufficient detail.

[0067] Due to the influence of flight altitude, field of view, relative observation geometry, and obstruction conditions, suspected targets often only exhibit low-confidence local anomalies, and the same suspected target shows differences at different times and angles, making it difficult for the mother aircraft to directly form verifiable detailed investigation conclusions.

[0068] In exemplary techniques, a common practice is to directly assign detailed investigation tasks to wingmen as coordinate points or bounding boxes, or to use a fixed threshold triggering deployment and allocation strategy on the master server, allowing wingmen to perform detailed investigations based on simple principles of proximity, redundancy, and load matching. However, in real-world tasks, this approach easily leads to typical dilemmas:

[0069] First, multiple wingmen form similar observation geometry near the same target, collecting highly correlated evidence fragments, resulting in bandwidth and energy consumption being occupied by redundant data, while evidence of key discrimination angles, key altitude layers or key time windows is missing.

[0070] Secondly, the mission may still need to be adjusted as the situation changes after it is issued. However, due to communication interruptions, preparation delays and differences in payload capacity, the mission may be issued but not be able to be completed with high quality.

[0071] Third, when the back-end of the master machine is integrated, it is difficult to ensure that the evidence from different wingmen is naturally complementary. This often requires additional manual rules or secondary scheduling, which reduces the real-time performance and reliability of the wide-area-detailed investigation closed loop.

[0072] Given the aforementioned difficulties, the core logic of the method in this application lies in:

[0073] The detailed investigation task is transformed from a traditional static task into a task representation that can be carried out on the wingman's side, enabling the main machine to organize detailed investigation behavior in a more abstract and composable form, and explicitly introducing complementary evidence collection constraints and objectives in the allocation process.

[0074] Specifically, after the master machine generates a detailed investigation task for suspected target clues, it encodes the task into multiple task seeds. Each task seed contains at least a task skeleton describing the detailed investigation process and rule information for expanding the task skeleton into detailed investigation actions. Task seeds are differentiated through dimensions such as observation angle, height layer, sensor mode, and time window, ensuring that multiple task seeds naturally correspond to different evidence collection focuses. To enable this complementarity to be implemented in an engineered manner, the master machine further introduces a complementary codebook organizational structure:

[0075] The codewords in the codebook are used to represent a set of detailed evidence collection requirements. Different codewords meet the preset low overlap constraint, thus providing a calculable and verifiable basis for how to avoid redundancy and how to supplement key evidence in the parallel evidence collection of multiple wingmen.

[0076] The mothership can map the mission seed to the target codeword and combine the support capabilities of each wingman for the codeword in terms of payload capacity, flight performance, mission preparation latency, etc., to complete the selection and pairing of wingmen for complementary evidence collection targets, so that the final set of wingmen deployed can form complementary observations at the execution level, rather than random superposition of repeated observations.

[0077] Furthermore, since the wingman is located inside the mothership before deployment, this application utilizes this built-in mission bay window in a specific embodiment to complete the pre-compilation and preparation of the mission seed:

[0078] Before leaving the aircraft, the wingman pre-compiles the loaded mission seed to generate the mission execution structure to be activated and outputs a pre-processing completion instruction to the mother aircraft. The mother aircraft then releases the mission after confirming that the wingman has the conditions to execute.

[0079] In this way, the complementary codebook-driven allocation strategy can be implemented more stably, while also reducing the risk of missing the time window due to differences in wingman preparation delays.

[0080] It should be noted that the complementary evidence collection and pseudo-redundancy suppression mechanism described in this application does not rely on fixed partitioning of the mission area, nor is it based on a specific wingman model or a specific payload series; it is oriented towards a more general cooperative constraint.

[0081] In the chain of wide-area reconnaissance triggering detailed investigation, limited by platform maneuverability, observation geometry, and resource constraints, it is necessary to achieve a balance between the completeness of evidence and efficiency in multi-wingman parallel detailed investigation through composable task representations and computable complementary constraints. The above understanding provides the necessary engineering context and principle source for those skilled in the art to implement the various embodiments of this application after reading it.

[0082] refer to Figure 1 , Figure 1 This is an exemplary application scenario diagram provided for an embodiment of this application.

[0083] like Figure 1 As shown, the airborne homeport system includes a mother aircraft and at least one wingman. The mother aircraft is a fixed-wing UAV used to perform wide-area reconnaissance missions over a large airspace. The wingman is a multi-rotor UAV, which is stored and deployed in the internal mounting area or mission cavity of the mother aircraft before the mission is carried out, so as to form an integrated airborne homeport structure.

[0084] During flight, the mothership continuously scans the wide-area scene based on its onboard sensors to obtain wide-area scene information with a large coverage area. When a suspected target or key area that needs further confirmation is identified in the wide-area scene, the mothership, while maintaining its own flight attitude and trajectory stability, determines the area where the suspected target is located as the detailed investigation area.

[0085] It is understandable that wide-area scenarios typically involve large areas with uneven target distribution and low information density. The mothership can quickly achieve coverage through high-speed cruising and wide-field-of-view imaging. However, due to limitations in flight altitude, speed, and observation geometry, wide-area reconnaissance results can often only form preliminary clues, and it is difficult to directly obtain detailed information that can be used to determine the attributes or status of targets.

[0086] Correspondingly, the detailed survey area is usually a local area in a wide-area scene. Its spatial range is relatively small, but it has higher requirements for observation resolution, observation angle stability and multidimensional evidence integrity. It is suitable for multi-rotor wingmen to perform detailed survey tasks in hovering or low-speed maneuvering mode.

[0087] It should be noted that during wide-area reconnaissance, once the mothership detects a detailed reconnaissance area within the wide-area scene, it can generate a detailed reconnaissance task based on the corresponding task requirements of that area, and load the detailed reconnaissance task as a task seed into the wingman located inside the mothership. In a specific embodiment of this application, the wingman is always within the mounting area inside the mothership before deployment. The mothership can complete task loading, task preparation status confirmation, and necessary task verification before deployment, thereby ensuring that the wingman can directly enter the task execution state after being deployed to the detailed reconnaissance area.

[0088] Those skilled in the art will understand that there is usually a spatial distance between the mother aircraft and the detailed survey area. The mother aircraft can organize and schedule the detailed survey task without entering the airspace above the detailed survey area. After being deployed, the wingman flies independently away from the mother aircraft and performs hovering detailed survey operations within the detailed survey area, transmitting the acquired detailed reconnaissance results back to the mother aircraft. The mother aircraft then combines wide-area scene information with the detailed survey results from the wingman to achieve collaborative cognition and information fusion of the wide-area scene and the detailed survey area. In this way, the flight safety issues caused by the mother aircraft entering complex or high-risk areas are avoided, while fully leveraging the advantages of multi-rotor wingmen in local detailed observation, thus forming a wide-area reconnaissance and hovering detailed survey collaborative application system suitable for complex airspace and multi-mission scenarios.

[0089] refer to Figure 2 , Figure 2 A schematic diagram of the modules of the wide-area reconnaissance and hovering precision reconnaissance collaborative system for deploying multi-rotors from a fixed wing, provided in an embodiment of this application.

[0090] like Figure 2 As shown, the system includes:

[0091] The task generation module is used to execute wide-area reconnaissance tasks and acquire wide-area reconnaissance data, determine suspected target clues based on the wide-area reconnaissance data and generate detailed investigation tasks, and encode the detailed investigation tasks to generate multiple task seeds.

[0092] The complementary allocation module is used to obtain the complementary codebook and, based on the target codewords corresponding to each task seed and the support codeword set corresponding to each wingman, execute the wingman allocation algorithm to determine the feasible pairing relationship between the task seed and the wingman.

[0093] The task execution module is used to load the corresponding task seed into the selected task wingman and deploy the task wingman; the task wingman is used to pre-compile the task seed before deployment to generate a task execution structure to be activated, and to activate the task execution structure after deployment to execute the hovering detailed investigation task and send the detailed investigation results back to the host machine.

[0094] Next, with reference to the accompanying drawings, the method for wide-area reconnaissance and hovering detailed reconnaissance coordination of fixed-wing multi-rotor deployment provided in this application embodiment will be further described. Figure 3 The method shown is applied to an airborne homeport system comprising a mother aircraft and multiple wingmen operating around the mother aircraft, wherein the mother aircraft corresponds to a fixed-wing UAV, the wingmen correspond to multi-rotor UAVs, the mother aircraft is used to perform wide-area reconnaissance missions, and loads detailed reconnaissance missions to be performed onto the multiple wingmen. The method includes:

[0095] S1: Encode the detailed investigation task to generate multiple task seeds;

[0096] The task seed includes task skeleton information for describing the detailed investigation task and rule information for expanding the task skeleton information into detailed investigation actions;

[0097] In this embodiment, after the mothership generates a detailed investigation task based on the acquired wide-area clue data during wide-area reconnaissance, it does not directly issue the detailed investigation task in the form of a fixed track or specific control commands. Instead, it first performs task encoding processing on the detailed investigation task, forming several mutually distinguishable task seeds. Each task seed describes the overall structure and execution flow of the detailed investigation task in the form of task skeleton information, such as the stages of entering the detailed investigation area, searching, confirming, and exiting. The task skeleton is further defined by rule information to determine how to unfold the above task skeleton into specific detailed investigation actions under different observation conditions, operating states, or constraints.

[0098] Understandably, the aforementioned methods enable the detailed investigation task to have a certain degree of scalability and flexibility while maintaining consistency in the overall objectives, thereby avoiding rigid failure of the task due to changes in the external environment or differences in operating status.

[0099] Those skilled in the art will understand that the specific organization of task skeleton information and rule information can be designed according to actual application needs, as long as it can generate corresponding detailed investigation behavior on the wingman side based on the operation data. This application does not impose further limitations.

[0100] S2: Select the corresponding task wingman based on the multiple task seeds, and output a preprocessing completion instruction to the master machine;

[0101] In this embodiment, for multiple task seeds obtained from the same detailed investigation task encoding, the master machine does not randomly or simply select wingmen based on spatial location. Instead, it combines the differential characteristics between task seeds in dimensions such as observation angle, altitude layer, sensor mode, and time window, and selects wingmen that can support different task seeds accordingly to form a complementary detailed investigation execution combination.

[0102] In some optional implementations, after determining the target codeword for the mission seed, the master aircraft selects wingmen capable of supporting the corresponding detailed evidence collection combination as mission wingmen based on their payload capacity, flight performance, and mission preparation latency. The selected wingmen then send a preprocessing completion indication back to the master aircraft. By completing mission preparation before deployment, the feasibility of the wingmen for the mission can be verified in advance without increasing communication burden, reducing the risk of mission interruption due to insufficient preparation or capability mismatch after deployment, thereby improving the overall reliability of parallel detailed investigation by multiple wingmen.

[0103] S3: Upon receiving the preprocessing completion instruction, deploy the mission wingman;

[0104] In this embodiment, after receiving a preprocessing completion instruction from the corresponding task wingman, the host machine triggers the deployment operation for that task wingman. Since the task wingman has completed pre-compilation processing of the task seed before deployment, the wingman can directly activate the generated task execution structure and quickly enter the detailed investigation execution state after leaving the host machine when the deployment trigger conditions are met, thereby shortening the response time between wide-area clue discovery and detailed investigation implementation. By moving task organization, wingman selection, and task preparation to before deployment, the uncertainties caused by environmental changes, communication interruptions, or accumulated latency can be effectively reduced, creating a closer and more stable collaborative relationship between wide-area reconnaissance and hovering detailed investigation.

[0105] Those skilled in the art will understand that the delivery method, delivery triggering conditions, and specific flight control strategies after the wingman leaves the cabin can be adjusted according to different platform conditions, as long as the basic requirements for the wingman to perform the corresponding detailed investigation task after delivery are met. This application does not impose any further limitations on this.

[0106] Before detailing the specific technical aspects of the steps, this application's embodiments need to reiterate:

[0107] The collaborative object addressed in this application is not a simple combination of a fixed-wing aircraft carrying a multi-rotor aircraft in the traditional sense, but rather an airborne homeport operation mode with an internal mission module structure:

[0108] Multiple wingmen are located in the mounting area inside the mother aircraft at the beginning of the mission, with stable power supply, controllable communication links and a protected environment; at the same time, the mother aircraft is in a state of continuous maneuver during the wide-area reconnaissance phase, and its external observation geometry, ground distance and flight path strategy are constantly changing with the situation.

[0109] The aforementioned structural characteristics determine a contradiction that is difficult to resolve using conventional scheduling methods: the organization of detailed investigation tasks must simultaneously meet two requirements—rapid response and complementary evidence.

[0110] If priority is given to response speed, there is usually a tendency to select deployable wingmen and issue simplified tasks immediately after discovering clues, resulting in multiple wingmen repeatedly collecting evidence from the same perspective, which creates redundancy. If priority is given to complementarity, it is necessary to comprehensively evaluate the differences in capabilities, preparation time, and changes in the target situation among multiple wingmen, which often makes it difficult to complete reliable allocation in a short period of time, and ultimately misses the window for detailed investigation.

[0111] Because the structural feature of the wingman being inside the mother aircraft provides a pre-departure preparation window, the core logic of this application can transform the allocation problem from dynamic correction after deployment to executable locking before deployment without introducing additional complex external facilities, and further solidify the complementary evidence collection requirements into a calculable and verifiable task expression.

[0112] It should be noted that this application does not treat the detailed investigation task merely as an objective to be executed, but rather transforms it into a set of evidence collection constraints that can be combined and separated across multiple dimensions. To avoid situations where multiple wingmen appear to have different roles at the execution level but are actually highly correlated in their observations, this application uses a complementary codebook as an organizational carrier. Factors such as observation angle, altitude layer, sensor mode, and time window are expressed as codewords in a discrete combination, and the degree of overlap between codewords is controlled by a preset similarity threshold. Through this organizational method, the complementary constraints are brought forward to the task allocation stage, eliminating the need for subsequent post-hoc screening of data returned by wingmen or repeated adjustments of the mother aircraft's flight path in the air to fill evidence gaps. In other words, the complementary codebook transforms empirical rules about what evidence should be collected and which evidence should not be collected repeatedly into structured conditions that can be directly used for matching decisions.

[0113] Those skilled in the art will understand that the discrete granularity of codewords, similarity threshold, and codebook size can be set according to the task area scale, target motion characteristics, and payload capacity, as long as it can at least ensure that different codewords form differentiated evidence collection requirements in at least one dimension. This application does not impose any further limitations.

[0114] Furthermore, since the wingman can complete data loading and status interaction with the mother aircraft under controlled conditions before deployment, task allocation no longer depends solely on the static claims of the wingman's capability parameters, but can be solidified into a verifiable execution preparation state through pre-departure preparation behavior.

[0115] For example, after determining the target codeword corresponding to the task seed, it is possible to prioritize selecting wingmen that support the codeword and whose preparation latency meets the corresponding time window. Before deployment, the loaded content is checked for consistency and preprocessed for confirmation. This makes the evidence collection combination required by the codeword feasible at the execution level, which is substantially different from the conventional external mounting or long-distance wireless distribution methods.

[0116] The latter often cannot obtain reliable preparation completion instructions after the task is issued, and can only correct the deviation through trial operation or failure rollback after deployment, which easily leads to the waste of the detailed investigation window; while under the implementation logic of this application, the deployment action itself is equivalent to confirming that the complementary evidence collection combination has been correctly bound to the corresponding wingman and has the conditions for execution, thereby significantly reducing the probability of task interruption caused by duplicate evidence collection, wrong window evidence collection and capability mismatch that are common in multi-wingman parallel detailed investigation.

[0117] Furthermore, complementary codebooks do not express strong dependencies on specific terrain, specific target categories, or fixed partitions, but rather an organizational method oriented towards the uncertainty of wide-area clues.

[0118] Suspected target leads generated during the wide-area reconnaissance phase often exhibit characteristics such as uncertain location, uncertain category, or uncertain movement. If a single detailed investigation script is uniformly issued, inefficient evidence collection is likely to occur when the target shifts or changes in obstruction. Allowing wingmen to adjust freely may lead to convergence in behavior and repetitive aggregation among multiple aircraft. With codebook constraints, even when multiple wingmen adaptively deploy their missions under local environmental changes, they will still be limited by the boundaries of evidence combinations defined by the codewords, thus naturally differentiating themselves in terms of time, angle, altitude, or pattern. Therefore, the collaboration between wide-area reconnaissance and hovering detailed investigation no longer relies on direct locking of a single detailed investigation, but rather on a set of complementary evidence combinations to improve the robustness and verifiability of the judgment under limited resource conditions. This is also an important reason why the proposed solution can adapt to complex airspace, concurrent multiple leads, and unstable communication conditions.

[0119] Next, we will further elaborate on the technical aspects of the method in this application regarding wide-area reconnaissance missions.

[0120] In one example, performing a wide-area reconnaissance mission includes:

[0121] Acquire wide-area reconnaissance data, wherein the wide-area reconnaissance data includes at least one of the image data, video data, radar data, and infrared data acquired by the sensors mounted on the mothership;

[0122] Specifically, wide-area reconnaissance missions are not simply about acquiring data from a single type of sensor. Instead, they revolve around the goal of generating stable cues that can be used for subsequent detailed investigations under limited observation time and constantly changing flight attitude conditions. During flight, the mothership can activate one or more sensors for data acquisition as needed. Different types of wide-area reconnaissance data differ in temporal resolution, spatial resolution, and emphasis on target sensitivity. For example, image and video data are better suited for extracting the appearance and contour features of targets, radar data is better suited for providing cues of target presence under obstructed or complex weather conditions, and infrared data can highlight unusual thermal features in low light or complex background conditions.

[0123] Those skilled in the art will understand that the acquisition of the above data can be carried out continuously as the flight progresses, or it can be collected intermittently within a preset flight segment. The specific sensor combination and sampling strategy can be flexibly configured according to mission requirements, as long as it can provide basic observation information for subsequent clue extraction.

[0124] The wide-area reconnaissance data is preprocessed to obtain corresponding wide-area clue data, wherein the preprocessing includes time synchronization, geographic registration and target area segmentation;

[0125] It is understood that the preprocessing steps in this application are conventional choices that can be easily reproduced by those skilled in the art, and will not be described in detail here.

[0126] Based on the wide-area clue data, at least one suspected target clue is identified, and a detailed investigation task to be executed is generated for the suspected target clue. The identification of at least one suspected target clue includes: extracting suspected target candidates from the wide-area clue data through a preset target classification model; performing multi-frame association on the suspected target candidates to obtain a suspected target set; generating a corresponding suspected target clue for each suspected target in the suspected target set. The suspected target clue includes the target center position, position uncertainty parameters, and target motion hypothesis parameters. The target classification model is constructed using historical training samples.

[0127] In this embodiment, the determination of suspected target clues is not a one-time judgment result, but is gradually formed through candidate extraction, multi-frame association and parameterized modeling, thereby providing a stable and interpretable basis for the generation of detailed investigation tasks.

[0128] Specifically, the first step is to extract potential target candidates from the wide-area cue data using a pre-defined target classification model. This target classification model can be constructed based on historical training samples, which can be derived from historical task data, simulation data, or manually labeled data, and are used to characterize the typical features of the target of interest in the task under different observation conditions.

[0129] Those skilled in the art will understand that target classification models are not limited to a specific implementation. For example, they can be classification models based on traditional features, or detection or segmentation models based on deep learning, as long as they can output candidate regions, candidate points, or candidate point sets that match the target features in wide-area cue data. In actual operation, to avoid prematurely discarding potential targets, the suspected target candidates output by the model are usually set with a relatively loose confidence threshold, so that the suspected target candidates may include both real targets and false candidates introduced by background interference.

[0130] In one optional implementation, the target classification model is a two-stage detection model or a single-stage detection model. Its input is a geo-registered image frame, video keyframe, or its cropped block, and its output is a candidate target box, a candidate category label, and a corresponding confidence score. When the sensor is radar or infrared, the output can be the geometric envelope and confidence score of the candidate echo cluster or hot spot region.

[0131] In one optional implementation, model training uses historical labeled samples, which include target category, location annotations, and collection condition labels. After training, the model parameters are fixed and written to the task configuration with a version number. During the inference phase, non-maximum suppression can be used to merge overlapping candidates, and the confidence threshold can be a preset value in the range of 0.2-0.6 or adaptively adjusted according to the scenario to maintain a candidate extraction strategy that prioritizes recall.

[0132] Furthermore, the purpose of multi-frame association is to utilize consistency constraints in the temporal dimension to distinguish between persistent potential targets and transient noise or sporadic anomalies. Specifically, multi-frame association can match and aggregate candidates based on their spatial proximity, appearance or feature similarity, and motion continuity in consecutive time frames. For example, candidates whose positional changes are within a reasonable range and whose feature similarity is higher than a threshold in adjacent frames can be grouped into the same suspected target trajectory. Through this process, candidates that appear briefly and cannot form a stable trajectory in the temporal dimension will be naturally eliminated, while candidates that can be continuously observed in multiple frames will be retained, forming a set of suspected targets.

[0133] Those skilled in the art will understand that the time window length, association threshold, and matching strategy used for multi-frame association can be adjusted according to the mother aircraft's flight speed, sensor refresh rate, and target motion characteristics, and this application does not limit these aspects.

[0134] Furthermore, the target center position can be obtained through statistical estimation of candidate positions obtained from multi-frame correlation, such as using weighted averaging or least squares fitting to determine a representative center position. The position uncertainty parameter is used to characterize the credible range of the center position, and its magnitude can be estimated by combining the spatial dispersion of the candidate positions, sensor resolution, and the observation geometry of the mother aircraft, thus reflecting the positioning error of the target under wide-area observation conditions. The target motion hypothesis parameter is inferred based on the displacement trend of the suspected target in multiple frames and can be used to describe the possible motion direction, velocity range, or motion pattern hypothesis of the target. Through the above parameterized modeling, the suspected target clue is no longer just a static position marker, but a structured expression containing spatial uncertainty and temporal evolution characteristics, thus providing a sufficient information basis for subsequent detailed investigation tasks such as coding, differential analysis, and wingman allocation.

[0135] In some optional implementations, multi-frame association employs trajectory management logic based on gating and cost matching:

[0136] For adjacent frame candidates, gating is performed based on the spatial distance threshold under geographic coordinates. After gating, adjacent frame candidates are further constructed with a matching cost matrix and minimum cost matching is performed. Successfully matched adjacent frame candidates are merged into the same trajectory. Trajectories that fail to match the observation for several consecutive frames enter a temporary state and terminate after exceeding the mismatch frame number threshold.

[0137] The spatial distance threshold can be given by the target motion assumption parameters, such as multiplying the maximum speed by the inter-frame time and adding the positioning error margin; the mismatch frame number threshold can be taken in the range of 2-10 frames; the feature similarity can be determined by the cosine similarity of the appearance embedding vector, the difference in echo cluster statistics, or the difference in infrared hot spot shape.

[0138] In one optional implementation, the target center position is obtained by weighted fusion of the observation positions of multiple frames of associated trajectories, with the weights determined by the observation confidence and observation geometric quality; the position uncertainty parameter is estimated by the discreteness of the observation points and the sensor resolution, and can be represented by an elliptical confidence region or variance parameter, and synthesized with the upper bound of the geographic registration error; the target motion assumption parameters are obtained by fitting the trajectory displacement sequence, including at least the velocity range and heading range, and the fitting method can be a constant velocity model, a constant acceleration model, or a piecewise linear model.

[0139] Next, we will further elaborate on the technical content of the method regarding task seeds in this application.

[0140] It should be noted that, in this application, the mission seed can be understood as an intermediate mission expression form between wide-area reconnaissance clues and specific flight control commands. It is different from a static mission marker that only describes the target location, and also different from a complete execution script that is pre-fixed on the mother aircraft side. Instead, it is a structured mission unit that can be deployed on the wingman side in combination with the operational status while maintaining the overall constraints of the mission.

[0141] Understandably, the suspected target cues output during the wide-area reconnaissance phase typically exhibit characteristics such as location uncertainty, incomplete target attribute determination, and difficulty in accurately modeling motion states. The hovering detailed investigation phase, however, requires precise operations with clear timing, attitude, and observation requirements within a local airspace. Directly converting wide-area cues into fixed tracks or rigid control commands often necessitates excessive assumptions about uncertainties at the mothership level, leading to insufficient adaptability to changes in the actual environment. Conversely, issuing only abstract target descriptions to wingmen can easily result in convergent behavior among different wingmen due to a lack of structural constraints, making it difficult to guarantee the complementary nature of the overall evidence collection. The mission seed breaks down the detailed investigation task into two parts: mission skeleton information and rule information. This ensures the task has overall process constraints while retaining space for expansion and adjustment based on actual observation results during the execution phase. This improves the adaptability of the detailed investigation phase to uncertain environments without increasing the complexity of the mothership's decision-making.

[0142] Furthermore, the task skeleton information is used to depict the basic structure and phase logic of the detailed investigation task, such as the sequential relationship of entering the detailed investigation area, performing the search, confirming, and exiting or transferring, so that the wingman always advances around the same task objective during the execution process; the rule information is used to describe how to derive specific actions from the task skeleton under different operating conditions, such as selecting different observation methods or maneuver strategies when the target confidence level changes, the observation conditions change, or the resource constraints change.

[0143] In one example, the detailed investigation task is coded to generate multiple task seeds, including:

[0144] S1.1: Convert the suspected target clues corresponding to the detailed investigation task into a task parameter set, wherein the task parameter set includes detailed investigation area range parameters, detailed investigation time window parameters, target prior category parameters, target motion constraint parameters, and evidence collection requirement parameters;

[0145] Specifically, the suspected target clues output during the wide-area reconnaissance phase are presented in the form of central location, uncertainty, and motion assumptions. This form is convenient for situation presentation, but it is not enough to directly drive wingmen to form a set of executable detailed investigation actions in local airspace. Therefore, it is necessary to reconstruct the clue information into a set of task parameters that can simultaneously cover spatial boundaries, time constraints, motion continuity, and evidence collection requirements, so as to generate a unified skeleton and rules in the future.

[0146] In one embodiment, the task parameter set is organized in the form of a structured data object, used to transform suspected target clues generated during the wide-area reconnaissance phase into input information that can be directly used in task seed generation and wingman allocation. The task parameter set includes at least the following fields, the content of which can be expanded or trimmed according to specific task needs, but must be able to support the organization and execution of the detailed investigation task to a minimum.

[0147] The detailed investigation area range parameter describes the spatial constraints of the detailed investigation task. This parameter includes at least the center position of the detailed investigation area to indicate the representative spatial position of the suspected target; it may also include the area boundary in the form of a radius or polygon to cover the range of uncertainty in target positioning; when there are no-entry zones or high-risk areas within the detailed investigation area, it may further include the no-entry zone clipping results to correct the original detailed investigation area; in addition, the detailed investigation area range parameter may also include a minimum safety boundary to ensure that the wingman maintains the necessary safe distance from obstacles, the ground or other flying objects when performing the detailed investigation task.

[0148] The task parameter set also includes a detailed investigation time window parameter, used to constrain the execution conditions of the detailed investigation task in the time dimension. The detailed investigation time window parameter includes at least the earliest start time and the latest finish time allowed for the detailed investigation task, used to limit the effective execution range of the detailed investigation task; it may also include a time window length, used to indicate the maximum duration for which the detailed investigation task can be continuously executed within the time range; in some embodiments, it may also include an indicator of whether extension is allowed, used to indicate whether a limited extension adjustment of the execution time of the detailed investigation task is allowed when there is a delay in wingman preparation or a change in the deployment order.

[0149] The task parameter set further includes target prior category parameters, used to describe the category attributes exhibited by suspected targets during the wide-area reconnaissance phase. This parameter includes at least a target category ID to distinguish different types of targets; it may also include a corresponding confidence level to reflect the reliability of the target classification results; in some embodiments, it may also include category group information to classify targets into categories such as stationary targets, moving targets, heat source targets, and reflector targets, so that different detailed investigation strategies can be adopted in subsequent task skeleton generation and rule expansion processes.

[0150] The task parameter set also includes target motion constraint parameters, which describe the range of motion behavior that a suspected target may exhibit during the detailed investigation phase. These parameters include at least maximum velocity and maximum acceleration, used to limit the range of displacement the target may make per unit time; they may also include turning constraints, used to constrain possible changes in the target's direction of motion; in some embodiments, the target motion constraint parameters may also include prediction region expansion rules, used to dynamically expand the detailed investigation area based on target motion assumptions, thereby avoiding insufficient detailed investigation range when the target undergoes displacement.

[0151] Furthermore, the task parameter set also includes evidence collection requirement parameters, used to clarify the evidence collection targets of the detailed investigation task. These parameters include at least a required set or range of observation angles, indicating from which relative directions the target should be observed; they may also include a required set or range of height layers, indicating the height layers to be covered during the detailed investigation; and they may include a set of sensor modes, indicating the types or operating modes of sensors to be enabled. In some embodiments, they may further include parameters such as minimum dwell time and minimum coverage angle span, used to constrain the continuity and completeness of the detailed investigation actions. In addition, the evidence collection requirement parameters may also include a data transmission strategy, indicating whether the detailed investigation data is transmitted in full, in summary, or via keyframe transmission.

[0152] In one example, the suspected target clues corresponding to the detailed investigation task are converted into a set of task parameters, including:

[0153] Elliptic thresholding is performed on the target center position and position uncertainty parameters in the suspected target clues to determine the range parameters of the detailed investigation area;

[0154] Extrapolate the motion assumption parameters of the suspected target clues in the time domain to determine the detailed investigation time window parameters;

[0155] The confidence level of the prior category parameters of the suspected target clues is normalized to determine the parameters required for evidence collection.

[0156] S1.2: Based on the task parameter set, a task skeleton information describing the detailed investigation task is generated by a task state machine generation algorithm. The task skeleton information includes at least one task node and the transition relationship between task nodes. The task node includes at least one of an entry node, a search node, a confirmation node, a tracking node, and an exit node. Each task node is associated with corresponding spatial constraint parameters, height constraint parameters, and speed constraint parameters.

[0157] Specifically, detailed investigation tasks often exhibit phased characteristics at the execution level:

[0158] From exiting the cabin and entering the detailed investigation area, to forming a search path within the area, to confirming and tracking the target after it is discovered, and finally exiting and transmitting the results back.

[0159] If these stages are fixed with a single script, the script is prone to interruption due to uncertainty of the target or changes in the local environment. If there is no stage division, the wingman may experience strategy drift during execution, making it difficult for the master machine to interpret and verify the detailed investigation process. Therefore, a task state machine is used to form the task skeleton, so that the detailed investigation task has clear stage boundaries and controllable transfer logic.

[0160] In this embodiment, the task state machine generation algorithm can be executed according to the process of node template selection - node parameter instantiation - transition relationship construction. The node template selection is determined based on the target prior category parameters and evidence collection requirement parameters. For example, when the target prior category is a static facility, the skeleton can include confirmation nodes that focus on multi-angle confirmation and detailed collection; when the target prior category is a moving target, the skeleton can include tracking nodes and set tracking priority.

[0161] Furthermore, during node parameter instantiation, the detailed survey area range parameters are mapped to spatial constraint parameters for the entry node and the search node. For example, the entry node is associated with the allowed passage or flight segment constraints from the deployment point to the boundary of the detailed survey area; the search node is associated with the allowed sub-areas and coverage density requirements within the area. Altitude constraint parameters are jointly determined by evidence collection requirements and environmental constraints. For example, a lower altitude layer can be set when higher resolution is required, but this is subject to safety altitude and obstacle limitations. Speed ​​constraint parameters can be determined based on sensor sampling requirements and target motion constraint parameters. For example, the maximum speed is limited when image clarity is required, while a higher speed limit is allowed but accompanied by a stricter energy budget when rapid interception is required. Transfer relationships are established based on the detailed survey time window parameters, target motion constraint parameters, and evidence collection requirements. For example, the transfer from the search node to the confirmation node can be triggered when the target candidate confidence reaches a threshold; the transfer from the confirmation node to the tracking node can be triggered when the target shows a movement trend; and the transfer from any node to the exit node can be triggered when the remaining energy is insufficient to guarantee a return or the time window is about to expire.

[0162] S1.3: Generate rule information based on the task parameter set to expand the task skeleton information into detailed investigation actions, wherein the rule information includes a mapping relationship between triggering conditions and action templates;

[0163] Specifically, the mission skeleton addresses what phased tasks to perform, but in actual flight, specific actions such as when to hover, when to circle, when to use sector scanning, and when to switch sensor modes need to be dynamically determined based on local observations and operational status. If only the skeleton nodes are divided without rules, wingmen may adopt inconsistent strategies within the same node, leading to unstable evidence collection. If the actions are hardcoded, they cannot adapt to changes in local wind fields, obstructions, or communication status. Therefore, action selection needs to be extracted from the skeleton into rule information and described by a mapping relationship between trigger conditions and action templates.

[0164] In this embodiment, the rule information may include at least three parts: a set of triggering conditions, a set of action templates, and a mapping table.

[0165] The set of trigger conditions is used to describe the threshold or interval conditions of observable state quantities. These conditions may include observation confidence conditions, target loss duration conditions, communication availability conditions, remaining energy conditions, local occlusion rate conditions, and security risk conditions. Among these, observation confidence can be obtained from the target detection output of the wingman payload on the real-time observation data, communication availability can be obtained from link quality indicators or backhaul throughput statistics, and remaining energy can be obtained from battery management data or energy consumption estimation.

[0166] Those skilled in the art will understand that the acquisition of the aforementioned state variables can be configured according to the actual platform conditions, as long as they can minimally characterize the wingman's operating state.

[0167] The action template set is used to describe the basic detailed investigation action units that can be invoked. For example, the hovering fixed-point observation template can include the hovering duration range, lens pointing strategy, and sampling frequency range; the fan-shaped scanning template can include the scanning angle range, scanning step and rotation strategy; the orbiting and hovering template can include the radius range, height layer switching strategy, and entry and exit point rules; the multi-height layer verification template can include the height layer sequence and the dwell conditions at each height layer; and the predictive interception template can include the interception point update rules generated based on the target motion constraint parameters.

[0168] The mapping table is used to bind a trigger condition or a combination of trigger conditions to the corresponding action template, and can set priorities to resolve selection conflicts when multiple triggers are met simultaneously.

[0169] S1.4: Divide the task parameter set according to the preset task differentiation strategy, and allocate corresponding task skeleton information and rule information according to the division result, wherein the task differentiation strategy includes observation angle differentiation, height layer differentiation, sensor mode differentiation and time window differentiation.

[0170] Specifically, wide-area clues often cannot be used to make high-confidence judgments in a single observation. The detailed investigation phase requires obtaining sufficient multi-dimensional discriminative evidence within limited resources. If all evidence collection requirements are superimposed on a single wingman, it can easily lead to excessively long mission durations, insufficient energy, or data transmission overload. If multiple wingmen perform the same mission, they will highly overlap in terms of angle, time, or mode, resulting in redundant transmission data and crowding out critical evidence collection windows.

[0171] In this embodiment, the execution of the task differentiation strategy can be manifested as splitting the evidence collection requirement parameters and assigning them to different sub-tasks. For example, the set of observation angles that must be covered can be split into several angle subsets and bound to different task seeds; the set of required altitude layers can be split into high, medium, low or several altitude bands and bound to them respectively; the sensor mode combination can be split into task seeds based on priority, such as those based on visible light and those based on infrared or radar; the detailed investigation time window can be split into early window, middle window, late window or sub-windows layered by arrival delay and bound to them respectively, so that even if there are differences in the wingman preparation delay, some evidence can still be obtained on time. After the division, corresponding task skeleton information and rule information are generated for each sub-task parameter set. During generation, the skeleton node set can be kept consistent but the node parameters can be different. For example, both are confirmation nodes, but one task seed performs hovering fixed-point observation at a lower altitude layer, while another task seed performs orbiting and circling at a higher altitude layer to obtain different geometric perspectives; differences can also be formed at the rule level, such as one task seed prioritizing the triggering of the multi-altitude layer verification template, while another task seed prioritizes the triggering of the prediction interception template.

[0172] Next, we will further elaborate on the technical aspects of the method in this application regarding the mission wingman.

[0173] It should be noted that the complementary codebook in this application can be understood as a structured evidence organization carrier used to constrain and guide the detailed investigation behavior of multiple wingmen. It is not a simple set of task tags, but a priori expression of which types of evidence should be collected during the detailed investigation phase and which evidence should not be repeatedly collected by the same wingman or the same execution window.

[0174] By introducing complementary codebooks, the selection of wingmen no longer depends solely on the individual capabilities or spatial relationships of the wingmen, but rather forms a direct mapping with the organization of evidence in the detailed investigation task, thereby enabling controllable complementarity in the collaborative execution of multiple wingmen.

[0175] Understandably, the core issue in selecting a mission wingman lies in:

[0176] Detailed investigation tasks often require acquiring multi-dimensional evidence within a limited timeframe. However, wingmen naturally differ in payload configuration, flight performance, preparation delays, and available time windows. Ignoring these differences and relying solely on the ability to execute the task can easily lead to a concentration of observation angles, altitudes, or times during the execution phase, resulting in insufficient coverage of key dimensions. The complementary codebook discretizes the detailed investigation evidence collection requirements into several codewords, allowing different evidence combinations to exist in a comparable and filterable form. Each codeword corresponds to a set of explicit observation constraints, such as limiting the observation angle range, altitude layer, sensor mode, and execution time window. Therefore, the selection process for a wingman can be transformed into determining whether the wingman can support a specific codeword under given constraints, transforming the matching relationship between wingman capabilities and detailed investigation requirements from empirical judgment to structured matching.

[0177] In one example, the complementary codebook is constructed in the following ways:

[0178] Based on historical data, a discrete set of evidence dimensions is determined to characterize the combination of evidence collected in the detailed investigation. The discrete set of evidence dimensions includes a discrete set of observation angles, a discrete set of height layers, a discrete set of sensor modes, and a discrete set of time windows.

[0179] A candidate codeword set is generated based on the discrete set of evidence dimensions, wherein each candidate codeword is used to represent at least one discrete value selected from the discrete set of observation angles, the discrete set of height layers, the discrete set of sensor modes, and the discrete set of time windows.

[0180] The similarity of the candidate codewords in the candidate codeword set is calculated for each pair of candidate codewords. The similarity of the codewords is used to characterize the degree of overlap of the detailed evidence collection combination corresponding to the two candidate codewords in terms of observation angle, height layer, sensor mode and time window.

[0181] Codeword similarity is obtained by weighted summation of the overlap of four dimensions. Among them, observation angle overlap can be calculated as the proportion of the intersection of angle intervals to the union, height layer overlap can be calculated as the proportion of the intersection of height bands to the union, sensor mode overlap can be calculated as the proportion of the intersection of mode sets to the union, and time window overlap can be calculated as the proportion of the intersection of time intervals to the union. The weights can be set according to the key dimension weights in the evidence collection requirements parameters.

[0182] A complementary codebook is obtained by filtering from the candidate codeword set according to a preset similarity threshold. The filtering includes: determining candidate codewords to be selected one by one according to a preset codeword generation order; discarding the candidate codeword when the codeword similarity between the candidate codeword and any selected codeword in the complementary codebook is greater than the similarity threshold; and adding the candidate codeword to the complementary codebook when the codeword similarity between the candidate codeword and each selected codeword in the complementary codebook is less than or equal to the similarity threshold.

[0183] In some optional implementations, the construction of the complementary codebook is not a simple enumeration of detailed investigation tasks, but a structured expression formed by summarizing and abstracting which combinations of detailed investigation evidence have distinguishing value in engineering practice and which combinations are prone to redundancy based on past task operation data and execution results.

[0184] Historical data can include the execution trajectory of wingmen, the distribution of observation angles, the actual altitude layers used, the working modes of sensors, and the distribution of investigation activities over time in each detailed investigation mission. It can also be combined with indicators such as the validity of the investigation results, the data transmission load, and the mission completion time for correlation analysis.

[0185] Those skilled in the art will understand that the aforementioned historical data can originate from real flight missions, as well as simulation or semi-physical verification environments, as long as it reflects the performance of different evidence collection methods in actual operation. After statistical analysis of the historical data, several core dimensions for characterizing the combination of evidence collection in a detailed investigation can be extracted, and these dimensions can be discretized into a finite set, enabling the detailed investigation behavior to be described and compared in a unified discrete space.

[0186] Furthermore, each candidate codeword does not aim to cover all possible execution methods, but rather characterizes an evidence-gathering combination that can independently form discriminative information in engineering practice. For example, it may be limited to a certain angle range and a certain height layer, and a detailed investigation is performed using a specific sensor mode within a specified time window. The generation of candidate codewords can be rule-driven or data-driven. For example, it can prioritize generating combinations that have appeared frequently in historical tasks or have made significant contributions to the judgment, or it can restrict the combination of values ​​in different dimensions to avoid generating a large number of non-executable or low-value codewords.

[0187] Furthermore, codeword similarity calculation reflects whether two codewords overlap significantly in dimensions such as observation angle, height layer, sensor mode, and time window. For example, if two codewords differ only slightly in height layer while being identical in other dimensions, their similarity is high; if they differ in multiple dimensions, their similarity is low. By setting a similarity threshold and filtering candidate codewords according to a preset codeword generation order, a complementary codebook can be gradually constructed, resulting in a limited overlap and overall complementary combination of codewords at the evidence collection level. This filtering process is not completed all at once but is dynamically compared as codewords are gradually added to the complementary codebook, thus avoiding the introduction of evidence collection requirements that are highly similar at the execution level into the codebook.

[0188] In yet another example, based on the multiple task seeds, the selection of the corresponding task wingman includes:

[0189] S2.1: Obtain a complementary codebook, wherein the complementary codebook includes multiple codewords, each codeword representing a set of detailed evidence collection combinations, the detailed evidence collection combinations including the combination requirements of observation angle, height layer, sensor mode and time window;

[0190] Specifically, when multiple wingmen conduct parallel detailed investigations, if allocation is based solely on whether the location can be reached and photographed, multiple wingmen may repeatedly collect evidence from the same angle, altitude, or time period, while neglecting to cover another crucial angle or time window, leading to an imbalance in the dimensionality of the detailed investigation evidence. The complementary codebook abstracts the detailed investigation evidence collection requirements into a searchable and comparable set of codewords, making the differences in evidence combinations describable and constrainable from the allocation stage, thus providing a unified reference for subsequent task seed mapping and wingman matching.

[0191] In this embodiment, the complementary codebook can be pre-configured as mission configuration data on the mothership side before flight, or it can be updated during mission execution based on mission area, payload configuration, or historical execution results. Each codeword in the complementary codebook contains at least four types of fields:

[0192] The observation angle field is used to describe the observation azimuth range or azimuth sequence relative to the target, for example, specifying to enter from the south-east side of the target and complete the lateral observation;

[0193] The height layer field is used to describe the height band or set of height layers when performing detailed analysis. For example, it specifies whether to perform hover detail acquisition at a low height layer or perform wrap-around acquisition of structural relationships at a high height layer.

[0194] The Sensor Mode field describes the type of sensor or combination of operating modes that need to be enabled, such as visible light fixed focus, visible light zoom, infrared, radar, or a combination thereof.

[0195] The time window field is used to describe the evidence collection period or trigger sequence corresponding to the codeword. For example, low-altitude layer detailed evidence collection is completed within the first few seconds after deployment, or tracking confirmation is completed within the subsequent time window after the target may move.

[0196] Those skilled in the art will understand that the above fields can be configured according to the task type, as long as they can minimally characterize the combination of evidence examined.

[0197] S2.2: For each task seed, the corresponding target codeword is determined in the complementary codebook based on the task difference information corresponding to the task seed. The target codewords corresponding to different task seeds are complementary to each other in the dimensions corresponding to the observation angle, height layer, sensor mode and time window.

[0198] Specifically, task seeds already carry differential information such as observation angle, altitude layer, sensor mode, and time window during generation. However, this differential information itself remains an internal constraint of the task. Without establishing a mapping relationship with complementary codebooks, it is difficult to directly determine the degree of complementarity between multiple task seeds during the allocation phase, and it is also difficult to form a unified scale for subsequent wingman capability matching. Mapping task seeds to target codewords is equivalent to projecting the requirements of task seeds in the evidence collection dimension into the codebook space, thereby allowing the differences between different task seeds to be explicitly presented and constrained by differences in codeword fields.

[0199] In one example, determining the corresponding target codeword in the complementary codebook based on the task difference information corresponding to the task seed includes:

[0200] Obtain the set of task parameters associated with the task seed, and extract task differential labels from the set of task parameters. The task differential labels include observation angle differential labels, height layer differential labels, sensor mode differential labels, and time window differential labels.

[0201] The dimension value combination of the target codeword is determined based on the task differential label, wherein the dimension value combination includes: the target observation angle value determined by the observation angle differential label, the target height layer value determined by the height layer differential label, the target sensor mode value determined by the sensor mode differential label, and the target time window value determined by the time window differential label.

[0202] Search the complementary codebook for codewords that match the combination of the dimension values ​​to determine the corresponding target codeword.

[0203] S2.3: Obtain the set of support codes corresponding to each wingman, wherein the support codes are used to characterize the combination of detailed evidence collection corresponding to the codes that the wingman can execute under the condition that the load capacity, flight performance and mission preparation delay meet the preset constraints.

[0204] Specifically, the evidence collection combinations corresponding to the same codeword may not be feasible on different wingmen. This is because wingmen differ in payload configuration, achievable altitude, wind resistance, remaining range, communication capabilities, and the preparation time required to enter a stable state after takeoff. Without pre-modeling which codewords the wingmen can support, the allocation phase may result in logically complementary but infeasible pairings, leading to issues such as incorrect deployment windows, missing evidence, or premature withdrawal. Mapping wingman capabilities to a set of supported codewords transforms the differences between wingmen into relationships satisfying codeword fields, thus ensuring consistency between complementary and executable constraints in the allocation algorithm.

[0205] In this embodiment, the determination of the support codeword set can be calculated based on wingman operational data and platform parameters. Payload capacity is used to determine whether the sensor mode field is satisfied, such as whether the wingman is equipped with an infrared payload, whether it supports zoom, whether it supports rapid switching between multiple modes, and whether the switching latency falls within the allowed range of the codeword time window. Flight performance is used to determine whether the altitude and observation angle fields are satisfied, such as whether the wingman can ascend to the specified altitude and maintain it for the required duration under current remaining energy conditions, whether it can complete the specified angular span with the orbital radius required by the codeword, and whether it can maintain hovering stability under wind conditions. Mission preparation latency is used to determine whether the time window field is satisfied, such as whether the time required for entering a controllable state after exiting the cockpit, payload activation, time synchronization, and mission structure activation is less than the earliest evidence window specified by the codeword. The determination of support relationships can adopt a combination of hard and soft constraints: hard constraints are used to exclude unsatisfactory codewords, such as those lacking a corresponding payload mode or insufficient ceiling; soft constraints are used to give priority to executable codewords with different costs, such as high-altitude codewords that can still be executed when energy reserves are low, but at a higher cost.

[0206] S2.4: Based on the target codeword and the support codeword set, determine the feasible pairing relationship between the task seed and the wingman through a preset wingman allocation algorithm, and obtain the task wingman corresponding to each task seed;

[0207] Specifically, given that the target codewords are determined and the codeword sets for each wingman are obtained, the allocation problem is no longer an abstract selection of several wingmen. Instead, it requires forming a pairing relationship within the combination space of multiple task seeds and multiple wingmen, ensuring that each task seed can find an executable wingman while maintaining the complementarity of codewords among different task seeds overall. If allocation is done sequentially based on local optima, it is easy to occupy key wingmen early on, leaving subsequent task seeds without feasible matches, or causing codeword overlap in the final pairings, thus destroying complementarity. Therefore, a pre-defined wingman allocation algorithm is used to select wingmen under global constraints within the candidate pairing space.

[0208] In one example, the step of determining the feasible pairing relationship between task seeds and wingmen through a preset wingman allocation algorithm to obtain the task wingman corresponding to each task seed includes:

[0209] Based on the target codeword and the support codeword set, a candidate pairing set between the task seed and the wingman is established. When the support codeword set of the wingman includes the target codeword of the task seed, the corresponding wingman is determined as a candidate wingman of the task seed, and the corresponding candidate pairing relationship is generated.

[0210] The pairing cost is calculated based on the candidate pairing relationship, wherein the pairing cost is used to characterize the resource consumption and task risk when the corresponding wingman executes the corresponding task seed, and the pairing cost includes energy cost, time cost and communication cost;

[0211] Based on the candidate pairing relationships and the corresponding pairing costs, the initial pairing results are determined under the condition of satisfying the wingman resource constraints, wherein the wingman resource constraints include the wingman available energy threshold, the available time window threshold, and the available communication bandwidth threshold.

[0212] The initial pairing results are subjected to complementarity verification, and the initial pairing results are subjected to conflict resolution to obtain the final pairing results. The conflict resolution includes: replacing the conflicting task seed with another candidate wingman until the final pairing results satisfy the complementarity verification.

[0213] Output the final pairing result as the wingman for each task seed.

[0214] It's important to note that the wingman allocation algorithm isn't simply about randomly or proximity when selecting feasible wingmen. Instead, it revolves around the engineering goal of ensuring the stable, timely, and compliant collection of detailed evidence corresponding to complementary codewords under limited resource conditions. By decomposing the matching relationship between task seeds and wingmen into stages such as candidate pair generation, pairing cost evaluation, initial pair determination, and complementarity verification and conflict resolution, the allocation process comprehensively considers wingman capability differences, resource consumption, and the overall evidence collection layout before execution. This avoids the complexity of compensating for allocation imbalances through temporary adjustments after deployment.

[0215] When establishing the candidate pairing set, based on the inclusion relationship between the target codewords of the task seed and the set of supporting codewords for wingmen, wingmen that can satisfy the codeword constraints at the execution level are selected one by one, and candidate pairing relationships are generated for each task seed-wingman combination. This candidate pairing relationship not only represents logical executability but also serves as the basic unit for subsequent cost evaluation. For each candidate pairing relationship, the resource consumption required to execute the task seed is estimated by combining the location of the detailed investigation area, the altitude layer and observation method required by the codeword, the current operating status of the wingman, and the expected deployment location of the mother aircraft. The energy cost can be comprehensively evaluated based on the wingman's flight distance from the deployment point to the detailed investigation area, the altitude of climb or descent, the expected hovering or orbiting duration, and the sensor power consumption; the time cost can be estimated from the time required for the wingman to complete task preparation, enter the detailed investigation area, perform the evidence collection actions specified by the codeword, and reserve a safety margin; the communication cost can be estimated based on the sensor mode, sampling frequency, data compression strategy, and backhaul link conditions corresponding to the codeword.

[0216] Those skilled in the art will understand that the above cost assessment does not require absolute precision; it only needs to provide a relatively comparable quantitative basis between different candidate pairings. Therefore, the corresponding cost calculation can be achieved through rule-based estimation based on operational state parameters, lookup table-based empirical models, or lightweight prediction models. For example, energy costs can be estimated by combining flight distance, altitude changes, and payload operating status with preset energy consumption estimation rules; time costs can be estimated by the relationship between mission preparation delay, estimated arrival time, and codeword corresponding time windows; and communication costs can be estimated by sensor operating modes, sampling frequencies, and link state parameters. This application does not further limit the specific form of the cost calculation model, as long as it can provide consistent and repeatable relative cost ranking results among candidate pairings to support subsequent pairing selection and conflict resolution.

[0217] Furthermore, wingman resource constraints are used to limit the workload that a single wingman can undertake within the same time period, avoiding execution failures or mid-process exits due to over-allocation.

[0218] For example, when the energy cost of a candidate pairing exceeds the wingman's current available energy threshold, the pairing can be directly excluded; when the time cost conflicts with the time window corresponding to the codeword, the pairing can be marked as low priority or eliminated; when the communication cost may cause backhaul link congestion, its participation priority can be reduced. By selecting candidate pairings with lower costs while satisfying the above resource constraints, an initial pairing result covering all or most task seeds can be formed, ensuring that each task seed is assigned to a wingman with execution capabilities. The initial pairing result guarantees the availability of each task in engineering, but the overall complementary layout has not yet been finalized.

[0219] Furthermore, complementarity verification checks whether the target codewords corresponding to each task seed in the initial pairing results overlap beyond the preset allowable range in dimensions such as observation angle, height layer, sensor mode, or time window. If some task seeds are found to be highly overlapping in multiple dimensions, it indicates that the current allocation may lead to multiple wingmen collecting highly relevant evidence at the execution level. In this case, conflict resolution processing can be performed on the conflicting task seeds. For example, another candidate wingman supporting its target codeword can be selected for one of the task seeds, or, without violating the task differential label constraints, the task seed can be remapped to a neighboring codeword in the complementary codebook, and a wingman supporting that codeword can be re-matched. Through the gradual resolution of conflicts, the final pairing results can maintain the complementary relationship between different task seeds in the evidence collection dimension while satisfying the wingman resource constraints, thus providing a stable and controllable foundation for subsequent task loading, pre-compilation, and deployment execution.

[0220] Next, a specific and complete example will be used to illustrate the entire process. The example is only to illustrate the feasibility at the computational level and does not represent the actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments.

[0221] In one optional implementation, for each candidate pairing relationship, the energy cost, time cost, and communication cost are calculated using interpretable component estimates, and the component estimates are normalized to form a comparable comprehensive pairing cost.

[0222] For example, energy costs can be broken down into three parts: arrival energy consumption, operational energy consumption, and payload energy consumption, where arrival energy consumption consists of horizontal range energy consumption and climb energy consumption.

[0223] Horizontal range energy consumption can be estimated by multiplying the expected range by the energy consumption per unit range:

[0224] Energy consumption per unit distance can be obtained from calibration data or historical flight statistics. For example, under windless or weak wind conditions, the range of 0.35 to 0.60 amp-hours per kilometer can be taken. The expected distance is determined by the geographical distance from the mother aircraft drop point to the entrance point of the survey area. For example, if it is 1.8 kilometers, the horizontal distance energy consumption can be estimated to be 0.63 to 1.08 amp-hours.

[0225] Climbing energy consumption can be estimated by multiplying the climbing height by the energy consumption per unit of climbing:

[0226] The energy consumption per unit climb can be obtained from the platform calibration, for example, 0.08 to 0.15 amp-hours per 100 meters; when the code requires climbing from an altitude of 250 meters to 350 meters, the climb height is 100 meters, and the energy consumption per climb is estimated to be 0.08 to 0.15 amp-hours.

[0227] Energy consumption for operation can be estimated by multiplying the operation duration by the unit power consumption for operation:

[0228] The unit power consumption can be looked up in the table according to the hovering or surround mode. For example, the hovering power consumption is 220 to 320 watts and the surround power consumption is 260 to 380 watts. When the code requires 120 seconds of hovering for evidence collection and 180 seconds of surround for evidence collection, the corresponding ampere-hour consumption can be converted by combining the nominal value of the battery voltage, so as to obtain the estimated value of the operating energy consumption.

[0229] Load energy consumption can be estimated by multiplying load power consumption by load operating time:

[0230] For example, a visible light fixed-focus payload consumes 3 to 6 watts, a visible light zoom payload consumes 5 to 12 watts, an infrared payload consumes 8 to 15 watts, and a miniature radar payload consumes 15 to 35 watts. When the code requires infrared operation for 240 seconds, the payload energy consumption can be calculated based on the above power consumption range.

[0231] The energy cost of the candidate pairing is obtained by summing the above items and comparing it with the current available energy threshold of the wingman. For example, if the wingman's battery capacity is 12 amp-hours and a 40% margin is reserved for return and safety, the available energy threshold is 7.2 amp-hours. If the estimated energy cost exceeds 7.2 amp-hours, the candidate pairing is directly eliminated. If the energy cost is lower than the threshold but close to it, such as exceeding 6.0 amp-hours, the candidate pairing is marked as high energy consumption level so that its priority is reduced in subsequent sorting.

[0232] In the same implementation, the time cost is estimated using three factors: preparation delay, arrival time, and operation time.

[0233] The preparation delay can be composed of the time required for the wingman to enter stable control after exiting the cockpit, the payload initiation time, and the mission execution structure activation time. The exit stabilization time can be obtained by acceleration stabilization determination after exit detection, for example, stability can be determined if the attitude angular velocity is below the threshold for 1.5 consecutive seconds. The payload initiation time can be obtained from the payload self-test return, for example, 0.5 to 1.5 seconds for visible light payloads, 1.0 to 3.0 seconds for infrared payloads, and 2.0 to 6.0 seconds for radar payloads. The mission execution structure activation time can be determined by the pre-compiled output initiation parameters, for example, 0.2 to 0.8 seconds.

[0234] Arrival time can be estimated by dividing the expected distance by the available cruising speed, which can be set within safe speed constraints, such as 8 to 14 meters per second; when the expected distance is 1.8 kilometers, the arrival time is approximately 129 to 225 seconds.

[0235] The operation time is given directly by the code word field or determined by the action template after the rule information is expanded. For example, if hovering is 120 seconds and circling is 180 seconds, then the operation time is 300 seconds.

[0236] The time cost is the sum of three factors and compared with the time window threshold. For example, if the codeword time window requires the main evidence collection to be completed within 600 seconds after deployment, then it will be removed if the estimated total time exceeds 600 seconds; if the estimated total time falls within the range of 480 to 600 seconds, it will be marked as a tight constraint level, so that in subsequent allocation, a wingman with a shorter preparation time will be given priority to undertake the task corresponding to the codeword.

[0237] In the same implementation, the communication cost is estimated using the data generation rate, compression / decompression strategy, and effective link throughput.

[0238] The data generation rate is determined by the sensor mode field and sampling requirements. For example, 1080p visible light video can take 2 to 6 megabits per second under common encoding, infrared video can take 1 to 4 megabits per second, and key frame backhaul can be estimated at 0.2 to 1.5 megabytes per frame and 1 to 5 frames per second.

[0239] If the codeword requires 240 seconds of infrared video capture and adopts a summary return strategy, only keyframes are returned, at 2 frames per second, with each frame containing 0.8 megabytes, then the total data volume is approximately 384 megabytes.

[0240] The effective throughput of the link can be estimated by the real-time link quality. For example, if the current effective throughput is 1.5 megabytes per second, the time required to complete the backhaul is about 256 seconds. If simultaneous sampling and transmission is required and the time window is tight, the candidate pair can be marked as a high communication pressure level or a load reduction strategy can be triggered.

[0241] Communication costs can be represented by the estimated data volume and the bandwidth duration, and compared with the available communication bandwidth threshold. For example, when multiple wingmen are transmitting back simultaneously, in order to avoid congestion, the average available throughput of each wingman can be limited to no more than one-third of the total throughput. When the estimated throughput requirement of a candidate pair exceeds the quota, its priority is reduced or it is eliminated during the initial pairing determination stage.

[0242] Furthermore, the energy cost, time cost, and communication cost are each normalized to a score between 0 and 1:

[0243] Energy score can be obtained by the proportion of energy cost to available energy threshold, time score can be obtained by the proportion of time cost to time window length, and communication score can be obtained by the proportion of expected throughput demand to communication quota. The comprehensive pairing cost can be obtained by weighted superposition, where the weight can be given by the evidence collection requirement parameter. For example, time weight is increased for tasks with high timeliness requirements, communication weight is increased for bandwidth-constrained scenarios, and energy weight is increased for long-endurance scenarios.

[0244] Furthermore, complementarity verification and conflict resolution can also employ computable rule-based processes.

[0245] For example, for any two selected pairings in the initial pairing results, compare the degree of overlap of their target codewords in the four fields:

[0246] If the observation angle field is represented by azimuth sectors, then overlap can be determined by whether the sectors are the same or adjacent. For example, the same sectors are recorded as completely overlapping, adjacent sectors are recorded as partially overlapping, and non-adjacent sectors are recorded as not overlapping.

[0247] If the height layer field uses discrete height bands, then the same height band is recorded as completely overlapping, and adjacent height bands are recorded as partially overlapping;

[0248] If the sensor mode fields are the same, they are considered to overlap; otherwise, they are not considered to overlap.

[0249] If the time intervals of the time window field overlap, it is considered to be overlapping.

[0250] Based on the above determination, conflict criteria can be set. For example, if "any two pairing relationships are judged to be in conflict if three or more of the four fields are judged to be overlapping", then it is judged to be in conflict. Alternatively, different conflict weights can be assigned to complete overlap and partial overlap to form conflict levels.

[0251] When a conflict occurs, the priority is to resolve the conflict by replacing the candidate wingman.

[0252] In the candidate wingman set of the task seed, the wingman with the second lowest overall pairing cost and without introducing new conflicts is selected for replacement. If all candidate wingmen would introduce conflicts, conflict resolution is performed by codeword neighborhood replacement. That is, codewords adjacent to the original codewords under differential label constraints are selected from the complementary codebook. For example, the sensor mode is kept unchanged but the observation angle sector is switched to the adjacent sector, or the observation angle is kept unchanged but the altitude layer is switched to the adjacent altitude band, or the time window is switched to the next time segment when the allowable continuation flag is true. The new energy, time and communication costs are recalculated and then the feasibility and complementarity are checked until the final pairing result that meets the conflict criteria is obtained.

[0253] In yet another example, after the mission wingman is deployed, the method further includes:

[0254] The task seed is activated when the deployment trigger condition is detected to be met, wherein the deployment trigger condition includes off-board detection and the distance to the mother machine reaching a preset distance threshold.

[0255] It should be noted that existing UAV platforms typically possess multi-source state perception capabilities for detecting deployment trigger conditions and activating the mission. For example, they can determine whether deployment has been completed using door status sensors, acceleration change detection, air pressure change detection, or attitude change detection. Simultaneously, they can continuously calculate the spatial distance to the mother aircraft using inertial navigation, satellite positioning, or relative positioning to determine if a preset safe distance threshold has been reached. In engineering implementation, these conditions can be used as simple logical combinations to trigger mission activation. For instance, a pre-compiled mission execution structure can be activated when deployment is confirmed and the distance to the mother aircraft exceeds a threshold. This type of trigger logic is widely used in existing deployment UAVs, catapult-launched aircraft, or separate payload control scenarios, and those skilled in the art can directly implement it based on existing flight control and mission management modules.

[0256] After activating the task seed, the task wingman expands the task skeleton information according to the rule information to generate candidate detailed investigation plans.

[0257] Understandably, the task skeleton information essentially corresponds to a set of state nodes and their transition conditions, while the rule information corresponds to the mapping relationship between state trigger conditions and action templates. After the task is activated, the wingman can enter the initial node according to the predetermined state machine logic, and determine whether the trigger conditions defined in the rules are met based on real-time observation data and internal state data, thereby selecting the corresponding action template and generating candidate detailed search plans. For example, after entering the search node, different scanning or hovering actions are triggered based on changes in target confidence. This type of rule-based action scheduling has been widely used in existing autonomous UAV mission systems, so no additional modifications to the underlying control mechanism are required.

[0258] Based on the local environmental information and wingman status information acquired by the sensors, the candidate detailed inspection schemes are constrained and screened to determine the target detailed inspection scheme, and the hovering detailed inspection task is executed according to the target detailed inspection scheme.

[0259] Furthermore, existing UAV autonomous planning technologies can also meet the requirements for constraint solving and screening of candidate detailed investigation schemes. During execution, the wingman continuously acquires local environmental information, such as obstacle distribution, terrain height, and wind field estimation results, while combining this with its own state information, such as remaining energy, attitude stability, and communication status, to assess the feasibility of candidate detailed investigation schemes. The constraint solving process can employ existing techniques such as path feasibility checks, flight envelope constraint verification, and energy reachability analysis to eliminate schemes that do not meet safety or resource constraints, and select the target detailed investigation scheme that meets the rule priority or has the lowest cost from the remaining schemes for execution. The above constraint screening and execution logic already has mature implementation paths in existing autonomous inspection, automatic obstacle avoidance, and online replanning UAV systems, and those skilled in the art can implement it based on existing algorithm frameworks.

[0260] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for coordinated wide-area reconnaissance and hovering precision reconnaissance of fixed-wing aircraft deploying multi-rotor drones, applied to an airborne homeport system including a mother aircraft and multiple wingmen operating around the mother aircraft, characterized in that... The mother aircraft corresponds to a fixed-wing UAV, and the wingmen correspond to multi-rotor UAVs. The mother aircraft is used to perform wide-area reconnaissance missions and loads detailed reconnaissance missions to be performed onto multiple wingmen. The method includes: The detailed investigation task is encoded to generate multiple task seeds. Each task seed includes task skeleton information describing the detailed investigation task and rule information for expanding the task skeleton information into detailed investigation actions, including: The suspected target clues corresponding to the detailed investigation task are converted into a task parameter set, wherein the task parameter set includes detailed investigation area range parameters, detailed investigation time window parameters, target prior category parameters, target motion constraint parameters, and evidence collection requirement parameters; Based on the set of task parameters, a task skeleton information describing the detailed investigation task is generated by a task state machine generation algorithm. The task skeleton information includes at least one task node and the transition relationship between task nodes. The task node includes at least one of an entry node, a search node, a confirmation node, a tracking node, and an exit node. Each task node is associated with corresponding spatial constraint parameters, height constraint parameters, and speed constraint parameters. Based on the task parameter set, rule information is generated to expand the task skeleton information into detailed investigation actions, wherein the rule information includes a mapping relationship between trigger conditions and action templates; The task parameter set is divided according to a preset task differentiation strategy, and corresponding task skeleton information and rule information are assigned according to the division results. The task differentiation strategy includes observation angle differentiation, height layer differentiation, sensor mode differentiation and time window differentiation. Based on the multiple task seeds, select the corresponding task wingman and output a preprocessing completion indication to the master machine, including: Obtain a complementary codebook, wherein the complementary codebook includes multiple codewords, each codeword being used to represent a set of detailed evidence collection combinations, the detailed evidence collection combinations including the combination requirements of observation angle, height layer, sensor mode and time window; For each task seed, the corresponding target codeword is determined in the complementary codebook based on the task difference information corresponding to the task seed. The target codewords corresponding to different task seeds are complementary to each other in the dimensions corresponding to the observation angle, height layer, sensor mode and time window. Obtain the set of support codes for each wingman, wherein the support codes are used to characterize the combination of detailed evidence collection corresponding to the codes that the wingman can execute under the preset constraints of load capacity, flight performance and mission preparation delay. Based on the target codeword and the set of supporting codewords, a feasible pairing relationship between task seeds and wingmen is determined by a preset wingman allocation algorithm, and the task wingman corresponding to each task seed is obtained. Upon receiving the preprocessing completion instruction, the mission wingman is deployed.

2. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 1, characterized in that, The aforementioned wide-area reconnaissance mission includes: Acquire wide-area reconnaissance data, wherein the wide-area reconnaissance data includes at least one of the image data, video data, radar data, and infrared data acquired by the sensors mounted on the mothership; The wide-area reconnaissance data is preprocessed to obtain corresponding wide-area clue data, wherein the preprocessing includes time synchronization, geographic registration and target area segmentation; Based on the wide-area clue data, at least one suspected target clue is identified, and a detailed investigation task to be executed is generated for the suspected target clue. The identification of at least one suspected target clue includes: extracting suspected target candidates from the wide-area clue data through a preset target classification model; performing multi-frame association on the suspected target candidates to obtain a suspected target set; generating a corresponding suspected target clue for each suspected target in the suspected target set. The suspected target clue includes the target center position, position uncertainty parameters, and target motion hypothesis parameters. The target classification model is constructed using historical training samples.

3. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 2, characterized in that, The suspected target clues corresponding to the detailed investigation task are converted into a set of task parameters, including: Elliptic thresholding is performed on the target center position and position uncertainty parameters in the suspected target clues to determine the range parameters of the detailed investigation area; Extrapolate the motion assumption parameters of the suspected target clues in the time domain to determine the detailed investigation time window parameters; The confidence level of the prior category parameters of the suspected target clues is normalized to determine the parameters required for evidence collection.

4. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 1, characterized in that, The complementary codebook is constructed in the following ways: Based on historical data, a discrete set of evidence dimensions is determined to characterize the combination of evidence collected in the detailed investigation. The discrete set of evidence dimensions includes a discrete set of observation angles, a discrete set of height layers, a discrete set of sensor modes, and a discrete set of time windows. A candidate codeword set is generated based on the discrete set of evidence dimensions, wherein each candidate codeword is used to represent at least one discrete value selected from the discrete set of observation angles, the discrete set of height layers, the discrete set of sensor modes, and the discrete set of time windows. The similarity of the candidate codewords in the candidate codeword set is calculated for each pair of candidate codewords. The similarity of the codewords is used to characterize the degree of overlap of the detailed evidence collection combination corresponding to the two candidate codewords in terms of observation angle, height layer, sensor mode and time window. A complementary codebook is obtained by filtering from the candidate codeword set according to a preset similarity threshold. The filtering includes: determining candidate codewords to be selected one by one according to a preset codeword generation order; discarding the candidate codeword when the codeword similarity between the candidate codeword and any selected codeword in the complementary codebook is greater than the similarity threshold; and adding the candidate codeword to the complementary codebook when the codeword similarity between the candidate codeword and each selected codeword in the complementary codebook is less than or equal to the similarity threshold.

5. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 1, characterized in that, The step of determining the corresponding target codeword in the complementary codebook based on the task difference information corresponding to the task seed includes: Obtain the set of task parameters associated with the task seed, and extract task differential labels from the set of task parameters. The task differential labels include observation angle differential labels, height layer differential labels, sensor mode differential labels, and time window differential labels. The dimension value combination of the target codeword is determined based on the task differential label, wherein the dimension value combination includes: the target observation angle value determined by the observation angle differential label, the target height layer value determined by the height layer differential label, the target sensor mode value determined by the sensor mode differential label, and the target time window value determined by the time window differential label. Search the complementary codebook for codewords that match the combination of the dimension values ​​to determine the corresponding target codeword.

6. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 1, characterized in that, The step of determining feasible pairings between task seeds and wingmen through a preset wingman allocation algorithm to obtain the task wingman corresponding to each task seed includes: Based on the target codeword and the support codeword set, a candidate pairing set between the task seed and the wingman is established. When the support codeword set of the wingman includes the target codeword of the task seed, the corresponding wingman is determined as a candidate wingman of the task seed, and the corresponding candidate pairing relationship is generated. The pairing cost is calculated based on the candidate pairing relationship, wherein the pairing cost is used to characterize the resource consumption and task risk when the corresponding wingman executes the corresponding task seed, and the pairing cost includes energy cost, time cost and communication cost; Based on the candidate pairing relationships and the corresponding pairing costs, the initial pairing results are determined under the condition of satisfying the wingman resource constraints, wherein the wingman resource constraints include the wingman available energy threshold, the available time window threshold, and the available communication bandwidth threshold. The initial pairing results are subjected to complementarity verification, and the initial pairing results are subjected to conflict resolution to obtain the final pairing results. The conflict resolution includes: replacing the conflicting task seed with another candidate wingman until the final pairing results satisfy the complementarity verification. Output the final pairing result as the wingman for each task seed.

7. The method for wide-area reconnaissance and hovering precision reconnaissance collaboration of fixed-wing multi-rotor deployment according to claim 1, characterized in that, After the mission wingman is deployed, the method further includes: The task seed is activated when the deployment trigger condition is detected to be met, wherein the deployment trigger condition includes off-board detection and the distance to the mother machine reaching a preset distance threshold. After activating the task seed, the task wingman expands the task skeleton information according to the rule information to generate candidate detailed investigation plans. Based on the local environmental information and wingman status information acquired by the sensors, the candidate detailed inspection schemes are constrained and screened to determine the target detailed inspection scheme, and the hovering detailed inspection task is executed according to the target detailed inspection scheme.

8. A wide-area reconnaissance and hovering precision reconnaissance collaborative system for fixed-wing multirotor deployment, used to implement the wide-area reconnaissance and hovering precision reconnaissance collaborative method for fixed-wing multirotor deployment as described in any one of claims 1-7, characterized in that, The system includes: The task generation module is used to execute wide-area reconnaissance tasks and acquire wide-area reconnaissance data, determine suspected target clues based on the wide-area reconnaissance data and generate detailed investigation tasks, and encode the detailed investigation tasks to generate multiple task seeds. The complementary allocation module is used to obtain the complementary codebook and, based on the target codewords corresponding to each task seed and the support codeword set corresponding to each wingman, execute the wingman allocation algorithm to determine the feasible pairing relationship between the task seed and the wingman. The task execution module is used to load the corresponding task seed into the selected task wingman and deploy the task wingman; the task wingman is used to pre-compile the task seed before deployment to generate a task execution structure to be activated, and to activate the task execution structure after deployment to execute the hovering detailed investigation task and send the detailed investigation results back to the host machine.

Images