Multi-agent-based unmanned aerial vehicle autonomous flight control method and server

Abstract

The invention provides a multi-agent-based unmanned aerial vehicle autonomous flight control method and a server, and relates to the technical field of unmanned aerial vehicle control. The method comprises the following steps: receiving an original task instruction for the unmanned aerial vehicle input by a user by adopting a central scheduling agent; analyzing the original task instruction by adopting a central scheduling agent to generate a task execution scheme; a central scheduling agent is adopted, and according to the time sequence dependency relationship, instructions of the multiple sub-tasks are sequentially sent to the corresponding sub-agents respectively; a plurality of sub-agents are adopted, the sub-tasks are executed according to the instructions of the respective sub-tasks, execution results of the respective sub-tasks are obtained, and the execution results of the respective sub-tasks are fed back to the central scheduling agent; and adopting the central scheduling agent to generate a target task execution result according to the execution results of the plurality of sub-tasks, and feeding back the target task execution result to the user. The sensing and understanding capability of the unmanned aerial vehicle to a complex environment is remarkably enhanced, and the autonomous control level of the unmanned aerial vehicle is further improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle control, in particular to an unmanned aerial vehicle autonomous flight control method based on multiple agents and a server. BACKGROUND

[0002] As one of the high-techs developing most rapidly in recent years, the unmanned aerial vehicle technology has realized large-scale application in civil and special fields by virtue of its advantages such as maneuverability, convenient deployment and controllable cost.

[0003] However, the current mainstream unmanned aerial vehicle control system still has significant technical bottlenecks, which is difficult to meet the autonomous operation requirements in complex scenarios. The current control system is usually constructed based on a preset flight path and a fixed task execution logic, and its core control mode relies on a pre-planned route. When facing dynamic changes in the actual environment (such as sudden air flow, temporary no-fly zone, unknown obstacles, complex terrain, etc.), it lacks autonomous decision-making and adaptive adjustment capabilities, and often needs real-time intervention by staff.

[0004] This control mode relying on manual operation not only greatly increases the operation cost and personnel burden, but also faces problems such as response delay and low coordination efficiency in complex task scenarios, which seriously restricts the application of unmanned aerial vehicle technology in high-complexity scenarios. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application provides an unmanned aerial vehicle autonomous flight control method based on multiple agents and a server to solve the problems in the prior art.

[0006] The technical solutions adopted by the embodiments of the present application are as follows: In a first aspect, the embodiments of the present application provide an unmanned aerial vehicle autonomous flight control method based on multiple agents, which is applied to a collaborative work system pre-constructed on a server. The collaborative work system integrates a central scheduling agent and multiple sub-agents. The method comprises the following steps: The central scheduling agent receives original task instructions for unmanned aerial vehicles input by a user; The central scheduling agent analyzes the original task instructions to generate a task execution scheme, which includes instructions of multiple sub-tasks and time sequence dependency relationships between the multiple sub-tasks; The central scheduling agent sends the instructions of the multiple sub-tasks to the corresponding sub-agents in sequence according to the time sequence dependency relationships; The multiple sub-agents execute the sub-tasks according to the instructions of the respective sub-tasks to obtain the execution results of the respective sub-tasks, and feed back the execution results of the respective sub-tasks to the central scheduling agent. The central scheduling agent is used to generate a target task execution result according to execution results of the plurality of sub-tasks, and the target task execution result is fed back to the user.

[0007] In an embodiment, the central scheduling agent is used to analyze the original task instruction, and generate a task execution scheme, including: The central scheduling agent is used to analyze the original task instruction according to a preset central scheduling strategy library and an agent capability index library, and generate the task execution scheme; the preset central scheduling strategy library stores task disassembling logic and execution steps corresponding to the original task instruction, and the agent capability index library stores capability information of the plurality of sub-agents.

[0008] In an embodiment, the plurality of sub-agents include a route planning agent, and the instructions of the plurality of sub-tasks include route planning instructions. The plurality of sub-agents are used to execute sub-tasks according to respective instructions of the sub-tasks, respectively, and obtain execution results of the respective sub-tasks, including: The route planning agent is used to execute a route planning sub-task according to a route planning strategy library and the route planning instructions, and obtain an execution result of the route planning sub-task; the route planning strategy library stores route adjustment strategy information, and the execution result of the route planning sub-task includes determining a planned route of the unmanned aerial vehicle according to the route adjustment strategy information.

[0009] In an embodiment, the plurality of sub-agents further include a flight control agent, and the instructions of the plurality of sub-tasks include flight control instructions. The plurality of sub-agents are used to execute sub-tasks according to respective instructions of the sub-tasks, respectively, and obtain execution results of the respective sub-tasks, including: The flight control agent is used to execute a flight control sub-task according to a flight control instruction set, the planned route of the unmanned aerial vehicle, and the flight control instructions, and obtain an execution result of the flight control sub-task; the flight control instruction set stores flight control strategy information, and the execution result of the flight control sub-task includes performing flight action control on the unmanned aerial vehicle according to the planned route of the unmanned aerial vehicle and the flight control strategy information.

[0010] In an embodiment, the plurality of sub-agents further include a task execution agent, and the instructions of the plurality of sub-tasks include task execution instructions. The plurality of sub-agents are used to execute sub-tasks according to respective instructions of the sub-tasks, respectively, and obtain execution results of the respective sub-tasks, including: The task execution agent is adopted to execute a work subtask of the UAV according to a task execution strategy library and the task execution instruction, and an execution result of the work subtask of the UAV is obtained; the task execution strategy library stores task execution step information of the UAV, and the execution result of the work subtask of the UAV includes determining a task to be executed of the UAV according to the task execution step information of the UAV.

[0011] In an embodiment, the plurality of sub-agents further include an emergency disposal agent and a question and answer query agent; the instructions of the plurality of sub-tasks include emergency disposal instructions and question and answer query instructions. The plurality of sub-agents are adopted to execute sub-tasks according to respective sub-task instructions, and execution results of the respective sub-tasks are obtained, including: The emergency disposal agent is adopted to execute an emergency disposal subtask according to an emergency disposal strategy library and the emergency disposal instruction, and an execution result of the emergency disposal subtask is obtained; the emergency disposal strategy library stores emergency disposal strategy information of the UAV, and the execution result of the emergency disposal subtask includes performing emergency disposal on the UAV according to the emergency disposal strategy information. And / or, the question and answer query agent is adopted to execute a question and answer query subtask according to a question and answer query template library and the question and answer query instruction, and an execution result of the question and answer query subtask is obtained; the question and answer query template library stores a question and answer query index, and the execution result of the question and answer query subtask includes querying a reply script corresponding to a question input by a user according to the question and answer query index.

[0012] In an embodiment, before the central scheduling agent receives the original task instruction input by the user for the UAV, the method further includes: A preset central scheduling strategy library and an agent capability index library are adopted to train a preset large language model, and the central scheduling agent is obtained.

[0013] In an embodiment, the central scheduling agent is adopted to analyze the original task instruction according to a preset central scheduling strategy library and an agent capability index library, and the task execution scheme is generated, including: The central scheduling agent is adopted to analyze the original task instruction and environment data of the UAV according to a preset central scheduling strategy library and an agent capability index library, and the task execution scheme is generated; the preset central scheduling strategy library stores task disassembly logic and execution steps corresponding to the original task instruction, and task environment data of the UAV.

[0014] In an embodiment, after the central scheduling agent is employed to generate a target task execution result according to the execution results of the plurality of sub-tasks and feedback to the user, the method further comprises: receiving evaluation information returned by the user; updating the preset central scheduling strategy library according to the evaluation information.

[0015] In a second aspect, the embodiments of the present application further provide a server, comprising a processor, a storage medium and a bus, the storage medium stores program instructions executable by the processor, when the server is running, the processor and the storage medium communicate through the bus, and the processor executes the program instructions to implement the unmanned aerial vehicle autonomous flight control method based on multiple agents according to any of the above embodiments.

[0016] The beneficial effects of the present application are: the present application provides an unmanned aerial vehicle autonomous flight control method based on multiple agents, by constructing a multi-agent collaborative working system deeply integrating multi-modal data, the perception and understanding ability of the unmanned aerial vehicle to complex environment is significantly enhanced, and the autonomous control level is improved, and the dependence on manual control intervention is greatly reduced. Moreover, the artificial learning cost can be greatly reduced, and the unmanned aerial vehicle control efficiency and flexibility are further improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0018] Figure 1 a structural schematic diagram of the collaborative working system provided by the embodiments of the present application; Figure 2 one of the flowcharts of the unmanned aerial vehicle autonomous flight control method based on multiple agents provided by the embodiments of the present application; Figure 3 the second flowchart of the unmanned aerial vehicle autonomous flight control method based on multiple agents provided by the embodiments of the present application; Figure 4 a structural schematic diagram of the unmanned aerial vehicle autonomous flight control device based on multiple agents provided by the embodiments of the present application; Figure 5 a structural schematic diagram of the server provided by the embodiments of the present application.

[0019] Label: 10 - receiving module, 20 - parsing module, 30 - sending module, 40 - executing module, 50 - feedback module, 100 - processor, 200 - storage medium, 300 - bus. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application.

[0021] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art without making creative efforts based on the embodiments in the present application belong to the scope of protection of the present application.

[0022] In addition, the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0023] It should be noted that the features in the embodiments of the present application can be combined with each other without conflict.

[0024] The current unmanned aerial vehicle control system faces the problem of low degree of autonomy of traditional methods, and it is urgent to propose an unmanned aerial vehicle autonomous flight control method based on large model application. Through multi-agent collaborative control, the technical bottleneck of the prior art is broken through, and the intelligent level, collaborative efficiency and environmental adaptability of the unmanned aerial vehicle autonomous flight control are further improved to meet the application requirements in complex scenarios.

[0025] Therefore, the embodiments of the present application provide a multi-agent based unmanned aerial vehicle autonomous flight control method, which is applied to a collaborative work system pre-constructed on a server. The multi-agent based unmanned aerial vehicle autonomous flight control method provided by the present application is specifically illustrated by a plurality of examples in combination with the drawings.

[0026] Firstly, the pre-constructed collaborative work system on the server is described as shown in the figure Figure 1 The collaborative work system integrates a central scheduling agent and multiple sub-agents, including a flight path planning agent, a flight control agent, a task execution agent, an emergency disposal agent, and a question and answer query agent.

[0027] The collaboration mechanism between each agent adopts a centralized interaction architecture, that is, the central scheduling agent is responsible for coordinating and managing the interaction between all sub-agents, and only the central scheduling agent interacts with the user. Specifically, the responsibilities of each agent role include: Central scheduling agent: as a central agent, it undertakes the overall scheduling responsibility, is responsible for receiving original task instructions directly from the user or triggered by the external environment, understands the instruction intent, analyzes the task target and performs sub-task splitting and sequencing, calls the corresponding function agent according to the content of each sub-task, monitors the execution of the task and dynamically adjusts the task execution scheme, evaluates the task execution result and feeds back to the user; Flight path planning agent: according to the target take-off and landing site and task type issued by the central agent, it integrates the internal and external constraint conditions such as the performance data of the unmanned aerial vehicle itself, the task target, real-time weather data, airspace control rules, etc., to generate the optimal global flight path, and dynamically adjust the flight path according to the real-time situation changes in the journey; Flight control agent: according to the planned flight path, or according to the action instructions issued by the central agent, it follows the corresponding unmanned aerial vehicle operation specification to output control signals to control the unmanned aerial vehicle flight; Task execution agent: according to the sub-task target issued by the central agent, it combines real-time sensor data to generate an execution strategy suitable for the actual internal and external environment, combines and schedules the unmanned aerial vehicle payload and intelligent algorithm to complete the sub-task target, and feeds back the task result to the central agent for evaluation; Emergency disposal agent: it monitors the health indicators and fault information of the unmanned aerial vehicle in real time, and when it detects alarm or abnormal conditions that endanger flight safety or affect the completion of the flight task, it formulates a disposal scheme according to the corresponding unmanned aerial vehicle operation specification, dynamically consults with the central scheduling agent, starts the corresponding emergency disposal program, and takes over the central scheduling responsibility and authority of the central scheduling agent when necessary; Question and answer query agent: responsible for processing user intentions that are not directly related to the task itself, which does not have the means and authority to control the unmanned aerial vehicle.

[0028] It should be further explained that the task execution agent is constructed based on a vision language model (VLM), and the other agents are constructed based on a large language model (LLM), for example, a preset central scheduling strategy library and an agent capability index library are used to train a preset large language model to obtain a central scheduling agent.

[0029] Each agent is equipped with a retrieval-augmented generation (RAG) multi-modal database of customized content, and the construction process of the database includes: full data acquisition and preprocessing, and construction of a multi-modal database.

[0030] The full data acquisition and preprocessing includes: acquiring data related to unmanned aerial vehicle flight control and task execution full link, for training and strengthening the understanding and decision-making ability of each agent role in the scene, specifically including flight operation documents, unmanned aerial vehicle performance parameters, historical flight control records, historical task logs, historical task acquisition data, emergency decision-making cases, product technical documents and maintenance records, and industry general knowledge documents.

[0031] The flight operation documents include unmanned aerial vehicle operation manuals, flight control protocols, emergency operation guidelines, etc.; the unmanned aerial vehicle performance parameters include maximum range, maximum speed, load capacity, etc.; the historical flight control records include control instructions, sensor data, environmental data, etc. during flight; the historical task logs include task targets, waypoint records, load and intelligent algorithm call records, task result reports, etc.; the historical task acquisition data includes multi-modal data such as aerial photography images, infrared thermal imaging, and laser radar point clouds; the emergency decision-making cases include unmanned aerial vehicle abnormal situation diagnosis records, emergency disposal reports, etc.; the product technical documents and maintenance records include the model, capacity, precision of each airborne device, and maintenance records, etc.

[0032] Further, the collected data is cleaned, integrated and labeled: for structured data, basic duplicate removal, outlier removal, missing value filling, format verification and business logic verification are performed; for text data, special character cleaning, document segmentation, topic classification, content expansion and enhancement are performed; for image data, format conversion, resolution unification, data enhancement, and target object labeling are performed; all historical task related data are labeled according to the unmanned aerial vehicle model, task type and load configuration to improve the subsequent knowledge retrieval efficiency.

[0033] Then, a multi-modal knowledge base is constructed: the pre-processed multi-dimensional data is integrated into the knowledge base, and the retrieval and content generation capabilities of the knowledge base are formed by using RAG technology; a multi-modal large model is used to generate a text summary in combination with the original data and labeled labels, and the summary is converted into an embedding and stored in a vector database.

[0034] Further, the original data is divided into three types of scene strategy library, interface tool library and history database to establish a knowledge base, wherein the scene strategy library contains scene decision templates with annotated spatio-temporal attributes, in the form of task environment data, historical operations and corresponding timestamps, which are direct references for the agent to analyze the current task; the interface tool library contains standardized interaction protocols, in the form of function tool interface protocols that each agent can call, or capability descriptions of other agents that can be dispatched within its permission range; the history database contains multi-modal data collected and labeled during the historical task or algorithm model training process, which is an auxiliary reference for the agent to make decisions.

[0035] For example, the database corresponding to each agent specifically contains the following data: Central scheduling agent: (1) integrated central scheduling strategy library (scene strategy library), containing customer original instructions, task environment data, task decomposition logic and execution steps, execution result evaluation, and labeling of unmanned aerial vehicle model, task type and payload configuration for each historical task; (2) integrated agent capability index library (interface tool library), labeling the capability boundaries of each agent and the central scheduling permission transfer rules; (3) integrated flight operation document library (history database), used to evaluate the legitimacy and necessity of requests from sub-agents.

[0036] Route planning agent: (1) integrated route planning strategy library (scene strategy library), containing historical task waypoint records, weather influence or regulation rules for each segment, and corresponding route adjustment strategies; (2) integrated route planning tool library (interface tool library), containing route dynamic planning algorithms and obstacle avoidance algorithms; (3) integrated unmanned aerial vehicle performance parameter library (history tool library), used to assist in querying the rationality of generated routes.

[0037] Flight control agent: (1) integrated flight control instruction set (scene strategy library), containing "instruction-action" pairs in historical flight control records, forming a mapping from language instructions to flight control interfaces; (2) integrated flight control tool library (interface strategy library), containing flight control instruction protocols for each type of unmanned aerial vehicle; (3) integrated flight operation document library (history strategy library), used to assist in querying attitude correction operation procedures.

[0038] Task execution agent: (1) integrate task execution strategy library (scene strategy library), including subtask instructions, subtask environment data, subtask disassembly logic and execution steps, payload and algorithm enable and switching logic, flight path and flight control adjustment application record, and mark each historical task of unmanned aerial vehicle model, task type and payload configuration; (2) integrate payload algorithm tool library (interface strategy library), including different configuration of task payload control instructions and adjustable intelligent algorithm; (3) integrate multi-modal data case library (history strategy library), which is used to assist understanding of current task environment and calibrate decision results.

[0039] Emergency disposal agent: (1) integrate emergency disposal strategy library (scene strategy library), including abnormal condition diagnosis record of each type of unmanned aerial vehicle and emergency disposal decision case; (2) integrate agent capability index library (interface strategy library), mark the capability boundary of each agent and the central dispatch authority takeover rule; (3) integrate flight operation document library (history strategy library), which is used to assist the query of alarm or abnormal condition judgment method and emergency disposal operation program.

[0040] Question and answer query agent, (1) integrate question and answer query template library (scene strategy library), including answer tactics and query index formed according to intention type; (2) integrate flight operation document library (interface strategy library), which is used to assist the query of unmanned aerial vehicle operation specification; (3) integrate general knowledge base (history strategy library), including product technical documents and maintenance records of unmanned aerial vehicles and airborne equipment, and industry general knowledge documents.

[0041] Figure 2 One of the flowcharts of the unmanned aerial vehicle autonomous flight control method based on multi-agent provided by the embodiments of the present application is shown in Figure 2 As shown in the figure, the method comprises: S101, adopting the central dispatch agent to receive the original task instruction of the unmanned aerial vehicle input by the user.

[0042] The original task instruction is a complete task request issued by the user according to the actual demand without disassembly, which can include text instructions, voice instructions or preset format instruction codes, and the content can cover task target (such as aerial photography area, inspection route, distribution site, etc.), task time requirement, task precision requirement and other key information. The central dispatch agent establishes connection with the user terminal through the preset communication interface (such as wireless communication module, network interface, etc.), receives the original task instruction sent by the user in real time, and preliminarily checks the integrity of the instruction. If it is found that the instruction information is missing, the request for supplementing the instruction will be fed back to the user.

[0043] S102, adopting the central dispatch agent to analyze the original task instruction and generate a task execution scheme.

[0044] The central scheduling agent parses the original task instruction to generate a task execution scheme, specifically including: the central scheduling agent calls a preset central scheduling strategy library and an agent capability index library, based on the content characteristics of the original task instruction, the original task instruction is disassembled and planned according to the preset analysis logic, and a task execution scheme is generated.

[0045] The preset central scheduling strategy library is a database pre-constructed and stored in the central scheduling agent locally or in an associated server, which internally stores task disassembly logic and execution step templates corresponding to different types of original task instructions. For example, for a "regional aerial photography task", the strategy library stores the disassembly logic "determine the aerial photography area boundary-plan the aerial photography route-set the shooting parameters-allocate the aerial photography unmanned aerial vehicle" and the corresponding execution steps; for a "power inspection task", the disassembly logic "determine the inspection line segment-plan the inspection path-set the inspection sensor parameters-allocate the inspection unmanned aerial vehicle and data processing agent" and the execution steps are stored.

[0046] The agent capability index library is also pre-constructed and maintained, for storing the capability information of a plurality of sub-agents, the capability information including the types of sub-agents (such as route planning agent, flight control agent, task execution agent, emergency disposal agent, question and answer query agent, etc.). The central scheduling agent can obtain the capability matching of each sub-agent by querying the agent capability index library.

[0047] The task execution scheme includes the instructions of a plurality of sub-tasks and the time sequence dependency relationship between the plurality of sub-tasks. The sub-task instruction is a specific operation instruction for a single sub-agent, and the time sequence dependency relationship is used to define the execution order of each sub-task, for example, sub-task 1 (route planning) is executed, then sub-task 2 (flight control) is executed, and sub-task 2 (flight control) and sub-task 3 (task data collection) are executed synchronously.

[0048] S103, using the central scheduling agent, according to the time sequence dependency relationship, sequentially sending the instructions of the plurality of sub-tasks to the corresponding sub-agents.

[0049] The central scheduling agent constructs a sub-task execution time sequence queue based on the time sequence dependency relationship in the task execution scheme, and sequentially judges whether the execution conditions of each sub-task are met (such as whether the pre-sub-task has been completed, whether the corresponding sub-agent is in an idle state, etc.) according to the queue order. When the execution condition of a sub-task is met, the central scheduling agent accurately sends the instruction of the sub-task to the corresponding sub-agent through an internal communication bus or a wireless communication network. At the same time, the central scheduling agent records the sending time of the sub-task instruction, the receiving sub-agent identifier and other information, which is used for tracking and management of the subsequent task execution process.

[0050] S104, a plurality of sub-agents are adopted to respectively execute sub-tasks according to instructions of respective sub-tasks, obtain execution results of respective sub-tasks, and feed back the execution results of respective sub-tasks to the central scheduling agent.

[0051] After receiving the sub-task instructions sent by the central scheduling agent, each sub-agent first analyzes the instructions to determine the operation requirements of itself, and then calls the function modules of itself to execute the sub-tasks. For example, the flight control agent adjusts the flight attitude, speed and route of the unmanned aerial vehicle according to the instructions to complete the flight task; the task execution agent collects task data through the sensors (such as cameras, infrared detectors, etc.) carried by the agent, and performs preliminary preprocessing.

[0052] During the execution of the sub-tasks, the sub-agents feed back the execution states (such as "executing", "completed", "execution exception") to the central scheduling agent in real time; when the sub-tasks are completed, the sub-agents package and send the execution results containing execution result data, execution logs and other information to the central scheduling agent.

[0053] S105, the central scheduling agent is adopted to generate a target task execution result according to the execution results of a plurality of sub-tasks, and feed back to the user.

[0054] After receiving the execution results of all sub-tasks, the central scheduling agent first checks the integrity and validity of the execution results of respective sub-tasks to determine whether the execution results of the sub-tasks meet the requirements of the task execution scheme. If there is a sub-task execution result that is incomplete or invalid, the central scheduling agent processes it according to the preset fault-tolerant mechanism (such as reassigning sub-tasks, calling backup sub-agents to execute tasks, etc.) until valid sub-task execution results are obtained.

[0055] Subsequently, the central scheduling agent integrates and analyzes the valid sub-task execution results according to the target requirements of the original task instructions. For example, for a regional aerial photography task, the central scheduling agent splices and fuses the images collected by each data collection sub-agent to generate a complete regional aerial photography image, and combines the flight log of the flight control sub-agent to generate a target task execution result containing information such as aerial photography image, flight parameter, shooting time, etc.; for a power inspection task, the central scheduling agent combines the analysis results of the data processing sub-agent with the inspection route information to generate a target task execution result containing information such as fault point position, fault type, inspection report, etc.

[0056] Finally, the central scheduling agent sends the target task execution result to the user terminal through a preset feedback mode (such as text, image, voice or report form), and completes the entire execution process of the unmanned aerial vehicle task. For example, "3 fire sources are found, coordinates …, 2 survivors, coordinates …, and a rescue package has been dropped", and fire scene image data. If the user approves the sub-task execution result, the central scheduling agent issues a return instruction. If the user does not approve the sub-task execution result, the central scheduling agent adjusts the task execution scheme according to the user feedback.

[0057] To sum up, the embodiment of the application provides an unmanned aerial vehicle autonomous flight control method based on multiple agents. By constructing a multi-agent collaborative working system deeply integrating multi-modal data, the perception and understanding ability of the unmanned aerial vehicle to the complex environment is significantly enhanced, and the autonomous control level is improved, and the dependence on manual control intervention is greatly reduced. Moreover, the artificial learning cost is greatly reduced, and the unmanned aerial vehicle control efficiency and flexibility are further improved.

[0058] In an embodiment, the sub-agent includes a route planning agent, and the instruction of the sub-task includes a route planning instruction. The step of adopting multiple sub-agents to execute the sub-tasks according to the instructions of the respective sub-tasks to obtain the execution results of the respective sub-tasks can include: adopting the route planning agent to execute a route planning sub-task according to a route planning strategy library and the route planning instruction to obtain an execution result of the route planning sub-task.

[0059] Specifically, after completing the analysis of the original task instruction, the central scheduling agent generates a sub-task instruction including a route planning instruction, and according to the time sequence dependency relationship, when the execution condition (such as the pre-task "determine the task area" has been completed and the route planning agent is in an idle state) is met, the route planning instruction is sent to the corresponding route planning agent.

[0060] The route planning instruction includes key parameters such as task area boundary coordinates, flight height limit, no-fly zone information, and task time requirement, for example, "plan a flight route in the area of east longitude 116.3°-116.5°, north latitude 39.9°-40.1°, the flight height is not more than 100 meters, avoid the two no-fly zones A and B in the area, and the route planning needs to be completed within 1 hour".

[0061] After receiving the path planning instruction, the path planning agent first parses the key parameters in the instruction, and then calls the pre-constructed path planning strategy library to execute the path planning subtask in combination with the parameter information. The path planning strategy library stores a variety of path adjustment strategy information, including "shortest path priority strategy", "obstacle avoidance priority strategy", "energy optimization strategy", "task efficiency priority strategy", etc. For example, when the instruction contains no-fly zone information, the path planning agent calls the "obstacle avoidance priority strategy" to bypass the no-fly zone through a path algorithm (such as the RRT algorithm); when the instruction contains strict time requirements, the "task efficiency priority strategy" is called to plan a path with higher straightness and fewer turns to shorten the flight time.

[0062] During the planning process, the path planning agent will also verify the preliminary planned path in combination with the performance parameters of the UAV (such as maximum flight speed, turning radius, etc., obtained from the agent capability index library) to ensure that the path meets the actual flight capabilities of the UAV.

[0063] After the path planning agent completes the planning, it generates the execution result of the path planning subtask, which includes information such as the specific coordinate point sequence of the planned path, the flight speed suggestion of each segment, the estimated flight time, and the obstacle avoidance point identifier. For example, the planned path is "start point P1 (116.3°E, 39.9°N) → P2 (116.4°E, 39.9°N) → P3 (116.4°E, 40.0°N) → P4 (116.5°E, 40.0°N) → end point P5 (116.5°E, 40.1°N), segment P2-P3 bypasses no-fly zone A, estimated flight time 45 minutes".

[0064] Subsequently, the path planning agent packages the execution result and feeds it back to the central scheduling agent, which verifies the result and confirms that it meets the requirements of the path planning instruction before using it as the basis for subsequent flight control subtasks.

[0065] In an embodiment, the sub-agents include a flight control agent, and the instructions of the sub-tasks include flight control instructions. S104 can include: using the flight control agent, executing the flight control subtask according to the flight control instruction set, the planned path of the UAV, and the flight control instruction, to obtain the execution result of the flight control subtask.

[0066] Specifically, after receiving the planned path fed back by the path planning agent, the central scheduling agent generates flight control instructions and sends the flight control instructions and the planned path to the flight control agent according to the time sequence dependency relationship.

[0067] The flight control instructions include take-off instructions, cruise instructions, landing instructions, attitude adjustment instructions, etc., such as "take off from the take-off point P0, fly according to the planned route P1-P5, cruise at a speed of 15 m / s, hover for 30 seconds after reaching the end point P5, and then return to the take-off point P0 for landing"; the flight control instruction set is a database pre-stored in the flight control intelligent agent, and contains flight control strategy information, such as "take-off phase: after leaving the ground, climb to a preset height at an acceleration of 5 m / s 2 ", "cruise phase: when the side wind speed exceeds 5 m / s, adjust the attitude of the fuselage to keep the route stable", "landing phase: use gradual descent, and slow down to 2 m / s when leaving the ground by 10 meters", etc.

[0068] After receiving the flight control instructions and the planned route, the flight control intelligent agent matches the two with the strategy information in the flight control instruction set, generates a specific flight action control scheme, and sends control signals through an interface (such as a PWM interface or a CAN bus) with the flight control system of the unmanned aerial vehicle to realize flight action control of the unmanned aerial vehicle.

[0069] In the take-off phase, the flight control intelligent agent sends motor speed control signals according to the "take-off phase" control strategy, so that the unmanned aerial vehicle leaves the ground and climbs to the starting height (such as 50 meters) of the planned route; in the cruise phase, the GPS positioning information and the gyroscope attitude information of the unmanned aerial vehicle are received in real time, and are compared with the planned route; if there is a route deviation (such as a deviation of more than 2 meters), the rudder control signal is sent according to the "cruise phase" obstacle avoidance and correction strategy to adjust the heading of the unmanned aerial vehicle, and the attitude of the fuselage is adjusted in combination with the wind speed sensor data; when approaching the end point, the unmanned aerial vehicle is controlled to hover according to the instruction requirements, and after the hovering ends, the unmanned aerial vehicle is controlled to land stably according to the "landing phase" strategy.

[0070] During the execution of the flight control sub-task, the flight control intelligent agent feeds back the flight state (such as the current position, flight speed, attitude of the fuselage, remaining power, etc.) of the unmanned aerial vehicle to the central scheduling intelligent agent in real time; after completing all flight control actions, the execution result of the flight control sub-task is generated, including complete flight logs (take-off time, cruise trajectory, flight parameters in each phase, landing time, etc.) and flight state statistical data (such as average flight speed, maximum deviation, power consumption, etc.), and the result is fed back to the central scheduling intelligent agent.

[0071] In an embodiment, the sub-intelligent agent further includes a task execution intelligent agent, and the instructions of the sub-task include task execution instructions. S104 can include: using the task execution intelligent agent, executing the work sub-task of the unmanned aerial vehicle according to the task execution strategy library and the task execution instructions, to obtain the execution result of the work sub-task of the unmanned aerial vehicle.

[0072] The task execution agent does not directly intervene in flight control and has no right to directly modify the flight path. If the task execution agent needs to adjust the flight path or flight attitude due to load scheduling requirements, it should apply for adjustment to the central agent, which judges the rationality of the demand and the impact range, and then dispatches the flight path planning agent and the flight control agent to respond cooperatively.

[0073] Specifically, after receiving the task execution instruction, the task execution agent calls the corresponding step information in the task execution strategy library to execute the work subtask of the UAV. Specifically, the task execution agent starts and runs the load equipment (such as an infrared temperature measurement sensor) carried by the UAV according to the preset parameters, completes data collection at the specified flight segment, and performs preliminary preprocessing (such as filtering and denoising) on the collected data to generate the execution result of the work subtask, which includes the collected raw data, the preprocessed data, and the data collection log (collection time, corresponding flight path coordinates, etc.).

[0074] If the task execution agent needs to adjust the flight path or flight attitude (such as "the infrared temperature measurement sensor detects that the temperature of a certain section of the line is abnormal, and needs to be collected again within 5 meters") during the execution of the work subtask due to load scheduling requirements, it applies for adjustment to the central dispatch agent according to the preset procedure.

[0075] The adjustment application contains the reason for the application, the expected adjustment content (such as "deviating from the planned flight path by 5 meters at coordinate point P2.5, hovering for 5 seconds"), and the time required for adjustment. After receiving the application, the central dispatch agent judges the rationality of the demand (such as whether it meets the task objective) and the impact range (such as whether it will cause the flight time to exceed the deadline or enter a no-fly zone). If the judgment is passed, the flight path planning agent is dispatched to re-plan the local flight path, and the flight control agent is sent an adjusted flight control instruction to cooperatively respond to the adjustment demand. If the judgment is not passed, the task execution agent is fed back the rejection reason and an alternative solution (such as "it is not possible to deviate from the flight path, it is suggested to increase the sampling frequency of the sensor").

[0076] After the task execution agent completes the work subtask, it feeds back the execution result to the central dispatch agent, which integrates it into the overall task execution result.

[0077] In an embodiment, the plurality of sub-agents further includes an emergency handling agent and a question and answer query agent; and the instructions of the plurality of sub-tasks include emergency handling instructions and question and answer query instructions.

[0078] In S104, a plurality of sub-agents are adopted to execute sub-tasks according to the instructions of the respective sub-tasks, and execution results of the respective sub-tasks are obtained. Specifically, an emergency handling agent is adopted to execute an emergency handling sub-task according to an emergency handling strategy library and emergency handling instructions, and an execution result of the emergency handling sub-task is obtained.

[0079] Specifically, the central scheduling agent sends an emergency handling instruction to the emergency handling agent at the beginning of task execution. The content of the instruction is "real-time monitoring of the flight state and task execution state of the unmanned aerial vehicle, and performing handling operations according to the emergency handling strategy library when an abnormality is found". The emergency handling strategy library stores the emergency handling strategy information of the unmanned aerial vehicle, which is divided into "equipment failure type" (such as motor failure, sensor failure), "environmental abnormality type" (such as sudden strong wind, thunderstorm weather), "task abnormality type" (such as data acquisition failure) according to the abnormality type, and "first level (endangering flight safety)" "second level (affecting task execution but not endangering safety)" according to the abnormality level.

[0080] The emergency handling agent obtains the flight state data (such as motor speed, battery remaining capacity, current wind speed) of the unmanned aerial vehicle and the task execution state data of the sub-agents through the communication interface with the state monitoring module (such as motor speed sensor, battery voltage monitoring module, meteorological sensor) of the unmanned aerial vehicle.

[0081] When an abnormality is detected (such as "battery voltage drops to below 3.0V"), the emergency handling agent determines the abnormality type ("equipment failure type - battery failure") and the abnormality level ("first level - endangering flight safety") according to the unmanned aerial vehicle operation guide.

[0082] If it is determined to be a first-level abnormality (such as the above-mentioned battery failure): the emergency handling agent immediately reports the abnormality and level to the central scheduling agent, and after confirmation by the central scheduling agent, directly takes over the central scheduling responsibilities and authority, sends an instruction to terminate the non-critical operation of the task execution agent (such as stopping infrared temperature measurement data acquisition), and calls the "battery failure emergency strategy" in the emergency handling strategy library (such as "immediately start the return program, plan the shortest return route, and reduce the flight speed to save power"), sends an emergency flight control instruction to the flight control agent, and controls the unmanned aerial vehicle to safely return.

[0083] If it is determined to be a second-level abnormality (such as "data acquisition sensor temporary failure, automatically recovered after 10 seconds"): the emergency handling agent does not directly take over the authority, but submits a handling suggestion (such as "suggest that the flight control agent pause the flight for 10 seconds after the sensor is recovered, and supplement the data") to the central scheduling agent, which decides to dispatch the task execution agent and the flight control agent to respond cooperatively, and adjusts the task execution scheme (such as extending the flight time by 10 seconds) if necessary.

[0084] After the emergency handling is completed, the emergency handling agent generates an execution result containing the abnormality details, the handling process, and the handling effect (such as "the unmanned aerial vehicle successfully returns, with 5% remaining power"), and feeds back to the central scheduling agent.

[0085] When the central scheduling agent receives a query question raised by the user during the task execution process (such as "What is the current flight height of the UAV?", "How is the data collection progress?"), a question and answer query instruction is generated, and the user question and the corresponding task identifier are sent to the question and answer query agent.

[0086] After the question and answer query agent receives the instruction, a pre-constructed question and answer query template library is called, and the template library stores a question and answer query index, which is classified according to common user question types (such as "flight state query", "task progress query", "abnormal situation query") and is associated with corresponding reply rhetoric templates and data acquisition interfaces.

[0087] For example, for the question "What is the current flight height of the UAV?", the question and answer query agent matches the "flight state query" category through indexing, calls the corresponding reply rhetoric template ("The current flight height of the UAV is X meters"), and acquires real-time flight height data through the data interface with the flight control agent, fills in the template to generate reply content. The question and answer query agent feeds the generated reply content as an execution result to the central scheduling agent, which sends the reply content to the user terminal to complete the question and answer interaction.

[0088] S102 described employs a central scheduling agent to parse the original task instruction according to a preset central scheduling strategy library and an agent capability index library to generate a task execution scheme, which can include: employing a central scheduling agent to parse the original task instruction and the environmental data of the UAV according to a preset central scheduling strategy library and an agent capability index library to generate a task execution scheme; wherein the preset central scheduling strategy library stores task decomposition logic and execution steps corresponding to the original task instruction, and task environmental data of the UAV.

[0089] An environmental data collection module is deployed in the UAV system, which includes various sensors and data receiving units, as follows: A meteorological sensor is used to collect real-time meteorological data of the task area, including wind speed, wind direction, rainfall, visibility, air humidity, temperature, and other parameters; a geographic environment sensor is used to collect terrain data (such as altitude, slope, terrain type) and obstacle data (such as building height, tree distribution, power tower position) of the task area; an airspace data receiving unit is used to communicate with the air traffic control system or the UAV supervision platform to acquire real-time airspace control information (such as temporary no-fly zones, flight height limits, and real-time positions of other flying objects) of the task area; and a ground environment monitoring unit is used to collect ground environment data (such as ground wind speed, site flatness, and obstacle distance) of the takeoff point, landing point, and task critical nodes.

[0090] The environmental data collection module collects data at a preset frequency (e.g., 1 time / 30 seconds) and transmits the data in real time to an environmental data buffer of the central scheduling agent through a wireless communication link.

[0091] For example, before a certain execution task, the central scheduling agent receives the original task instruction input by the user through the user interaction interface, such as "execute fire search and rescue in region A and identify the survivor position", at the same time, the central scheduling agent extracts the real-time environmental data of the task region and the associated region from the environmental data buffer, and determines that it is a night environment, then the central scheduling agent needs to screen the unmanned aerial vehicle that can execute the night task.

[0092] Then, the central scheduling agent refers to the central scheduling strategy library to generate a task execution scheme, determines the subtask type one by one, and dispatches to the corresponding agent role according to the step and time sequence dependency, for example, the task is divided into (1) a route planning agent: generate a global route combined with real-time weather data, and plan a cruising route in region A, (2) a flight control agent: control flight action according to the planned route, (3) a task execution agent: execute fire search and rescue after arriving at region A, identify and return the survivor coordinates, (4) an emergency disposal agent: monitor the health status throughout the process, and (5) a question and answer query agent: on standby.

[0093] Each sub-agent analyzes the subtask target issued by the central scheduling agent, combines the real-time internal and external environmental data and the query result of the multi-modal knowledge base to generate a subtask execution scheme, for example, the task execution agent divides its subtask into (1) enabling an infrared camera after arriving at region A, calling a target recognition algorithm to scan the fire point, (2) calling a target recognition algorithm to analyze the infrared image in real time, locating the survivor position and returning, and (3) adjusting the position of the unmanned aerial vehicle after identifying the survivor, and calling a delivery device to deliver a rescue package.

[0094] Figure 3 Fig. 2 is a flowchart of a method for autonomous flight control of an unmanned aerial vehicle based on multiple agents provided by an embodiment of the present application, as shown in Figure 3 After the central scheduling agent generates the target task execution result according to the execution results of the multiple subtasks and feeds back to the user, the method of the present application further includes: S201, receiving the evaluation information returned by the user.

[0095] After the unmanned aerial vehicle returns to the preset point and stops, the task is marked as completed, the central scheduling agent generates a task summary report and submits it to the user, requests the user to evaluate the task execution process and result, and mobilizes each agent to store the task data into the knowledge base.

[0096] S202, updating the preset central scheduling strategy library according to the evaluation information.

[0097] According to the evaluation information of the user, the preset central scheduling strategy library is updated. Optionally, the route planning strategy library, the task execution strategy library, the emergency disposal strategy library, and the multi-modal data case library can also be updated. For example, according to the user feedback, it is identified that the user expresses positive evaluation on the performance of the central scheduling agent, the route planning agent, and the task execution agent in this task, and the knowledge base updating instruction is issued by the central scheduling agent.

[0098] Specifically, the central scheduling agent adds the instruction analysis and splitting logic of this task to the central scheduling strategy library, the route planning agent adds the global route planning of this task and the route adjustment strategy of the night fire search and rescue to the route planning strategy library, the task execution agent adds the load and algorithm deployment logic of this task to the task execution strategy library, and the thermal imaging data and identification annotation of this task are added to the multi-modal data case library.

[0099] The following continues to explain the device, equipment and storage medium for executing the multi-agent based unmanned aerial vehicle autonomous flight control method provided by any of the above embodiments of the present application, the specific implementation process and the technical effects produced are the same as those of the corresponding method embodiments described above. For brief description, the parts not mentioned in the following embodiments can be referred to the corresponding contents in the method embodiments.

[0100] Figure 4 The structure diagram of the multi-agent based unmanned aerial vehicle autonomous flight control device provided by the embodiments of the present application is shown in Figure 4 The present application also provides a multi-agent based unmanned aerial vehicle autonomous flight control device, which is applied to a pre-constructed collaborative working system of a server, the collaborative working system is integrated with a central scheduling agent and a plurality of sub-agents, and the device comprises: A receiving module 10 is configured to receive an original task instruction for an unmanned aerial vehicle input by a user by using the central scheduling agent.

[0101] An analysis module 20 is configured to analyze the original task instruction by using the central scheduling agent to generate a task execution scheme, the task execution scheme comprising instructions of a plurality of sub-tasks and a time sequence dependency relationship between the plurality of sub-tasks.

[0102] A sending module 30 is configured to sequentially send the instructions of the plurality of sub-tasks to the corresponding sub-agents according to the time sequence dependency relationship by using the central scheduling agent.

[0103] The execution module 40 is configured to execute the sub-tasks according to the instructions of the respective sub-tasks by using the plurality of sub-agents, to obtain the execution results of the respective sub-tasks, and to feed back the execution results of the respective sub-tasks to the central scheduling agent.

[0104] The feedback module 50 is configured to generate a target task execution result according to the execution results of the plurality of sub-tasks by using the central scheduling agent, and to feed back the target task execution result to the user.

[0105] Optionally, the analysis module 20 is further configured to analyze the original task instructions according to a preset central scheduling strategy library and an agent capability index library, to generate the task execution scheme; the preset central scheduling strategy library stores task disassembling logic and execution steps corresponding to the original task instructions, and the agent capability index library stores capability information of the plurality of sub-agents.

[0106] Optionally, the plurality of sub-agents include a route planning agent, and the instructions of the plurality of sub-tasks include route planning instructions.

[0107] The execution module 40 is further configured to execute a route planning sub-task according to a route planning strategy library and the route planning instructions by using the route planning agent, to obtain an execution result of the route planning sub-task; the route planning strategy library stores route adjustment strategy information, and the execution result of the route planning sub-task includes a planned route of the unmanned aerial vehicle determined according to the route adjustment strategy information.

[0108] Optionally, the plurality of sub-agents further include a flight control agent, and the instructions of the plurality of sub-tasks include flight control instructions.

[0109] The execution module 40 is further configured to execute a flight control sub-task according to a flight control instruction set, the planned route of the unmanned aerial vehicle and the flight control instructions by using the flight control agent, to obtain an execution result of the flight control sub-task; the flight control instruction set stores flight control strategy information, and the execution result of the flight control sub-task includes flight action control of the unmanned aerial vehicle according to the planned route of the unmanned aerial vehicle and the flight control strategy information.

[0110] Optionally, the plurality of sub-agents further comprises a task execution agent; and the instructions of the plurality of sub-tasks comprise task execution instructions. The execution module 40 is further configured to execute, by using the task execution agent, a work sub-task of the UAV according to a task execution strategy library and the task execution instructions, to obtain an execution result of the work sub-task of the UAV; wherein the task execution strategy library stores task execution step information of the UAV, and the execution result of the work sub-task of the UAV comprises determining a task to be executed of the UAV according to the task execution step information of the UAV.

[0111] Optionally, the plurality of sub-agents further comprises an emergency disposal agent and a question and answer query agent; and the instructions of the plurality of sub-tasks comprise emergency disposal instructions and question and answer query instructions. The execution module 40 is further configured to execute, by using the emergency disposal agent, an emergency disposal sub-task according to an emergency disposal strategy library and the emergency disposal instructions, to obtain an execution result of the emergency disposal sub-task; wherein the emergency disposal strategy library stores emergency disposal strategy information of the UAV, and the execution result of the emergency disposal sub-task comprises performing emergency disposal on the UAV according to the emergency disposal strategy information. The execution module 40 is further configured to execute, by using the question and answer query agent, a question and answer query sub-task according to a question and answer query template library and the question and answer query instructions, to obtain an execution result of the question and answer query sub-task; wherein the question and answer query template library stores a question and answer query index, and the execution result of the question and answer query sub-task comprises querying a reply script corresponding to a question input by the user according to the question and answer query index.

[0112] Optionally, the apparatus further comprises a training module configured to train a preset large language model by using a preset central scheduling strategy library and an agent capability index library, to obtain the central scheduling agent.

[0113] Optionally, the parsing module 20 is further configured to parse, by using the central scheduling agent, the original task instructions and environment data of the UAV according to a preset central scheduling strategy library and an agent capability index library, to generate the task execution scheme; wherein the preset central scheduling strategy library stores task decomposition logic and execution steps corresponding to the original task instructions, and task environment data of the UAV.

[0114] Optionally, the apparatus further comprises an updating module configured to receive evaluation information returned by the user; and update the preset central scheduling strategy library according to the evaluation information.

[0115] The apparatus described above is used to execute the method provided in the foregoing embodiments, and has similar implementation principles and technical effects, which will not be described here in detail.

[0116] The above modules can be one or more integrated circuits configured to implement the above methods, for example, one or more application specific integrated circuits (ASICs), or one or more microprocessors, or one or more field programmable gate arrays (FPGAs), etc. For another example, when a certain module above is implemented in the form of a processing element scheduling code, the processing element can be a general purpose processor, such as a central processing unit (CPU) or other processor capable of invoking program code. For another example, the modules can be integrated together to implement a system-on-a-chip (SOC).

[0117] As shown in Figure 5 The application further provides a server, including a processor 100, a storage medium 200, and a bus 300, the storage medium stores program instructions executable by the processor, when the server is running, the processor and the storage medium communicate through the bus, the processor executes the program instructions to implement the multi-agent based unmanned aerial vehicle autonomous flight control method in any of the above embodiments.

[0118] Optionally, the application further provides a readable storage medium, the readable storage medium stores program instructions, when the program instructions are run by a processor, the multi-agent based unmanned aerial vehicle autonomous flight control method in any of the above embodiments is implemented.

[0119] In several embodiments provided in the application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiment described above is only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interface, apparatus or unit, and can be electrical, mechanical or other forms.

[0120] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0121] In addition, each of the function units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software function unit.

[0122] The integrated unit realized in the form of software function unit can be stored in a computer readable storage medium. The software function unit stored in a storage medium includes a plurality of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to execute part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (English: Read-Only Memory, for short: ROM), a random access memory (English: Random Access Memory, for short: RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0123] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for autonomous flight control of unmanned aerial vehicles based on multi-agent systems, characterized in that, A collaborative work system pre-built on a server, the collaborative work system integrating a central scheduling agent and multiple sub-agents, the method comprising: The central scheduling agent receives the user's original mission instructions for the UAV. The central scheduling agent is used to parse the original task instructions and generate a task execution plan. The task execution plan includes instructions for multiple sub-tasks and the temporal dependencies between the multiple sub-tasks. The central scheduling agent is used to send the instructions of multiple sub-tasks to the corresponding sub-agents in sequence according to the time-series dependency relationship; The multiple sub-agents are used to execute sub-tasks according to the instructions of their respective sub-tasks, obtain the execution results of their respective sub-tasks, and feed back the execution results of their respective sub-tasks to the central scheduling agent; The central scheduling agent generates the target task execution result based on the execution results of the multiple sub-tasks and feeds it back to the user.

2. The method according to claim 1, characterized in that, The process of parsing the original task instructions using the central scheduling agent to generate a task execution plan includes: The central scheduling agent parses the original task instruction according to the preset central scheduling strategy library and the agent capability index library to generate the task execution plan; wherein, the preset central scheduling strategy library stores the task decomposition logic and execution steps corresponding to the original task instruction, and the agent capability index library stores the capability information of multiple sub-agents.

3. The method according to claim 1, characterized in that, The plurality of sub-agents include: a route planning agent; the instructions for the plurality of sub-tasks include: route planning instructions. The process of employing the multiple sub-agents, each executing a sub-task according to its own sub-task instructions, and obtaining the execution result of each sub-task, includes: The route planning agent executes route planning sub-tasks based on the route planning strategy library and the route planning instructions, and obtains the execution results of the route planning sub-tasks. The route planning strategy library stores route adjustment strategy information, and the execution results of the route planning sub-tasks include: determining the planned route of the UAV based on the route adjustment strategy information.

4. The method according to claim 3, characterized in that, The plurality of sub-agents further includes: a flight control agent; the instructions for the plurality of sub-tasks include: flight control instructions; The process of employing the multiple sub-agents, each executing a sub-task according to its own sub-task instructions, and obtaining the execution result of each sub-task, includes: The flight control agent executes flight control sub-tasks based on the flight control instruction set, the UAV's planned route, and the flight control instructions, and obtains the execution results of the flight control sub-tasks. The flight control instruction set stores flight control strategy information, and the execution results of the flight control sub-tasks include: performing flight maneuver control on the UAV based on the UAV's planned route and the flight control strategy information.

5. The method according to claim 3, characterized in that, The plurality of sub-agents further includes: a task execution agent; the instructions for the plurality of sub-tasks include: task execution instructions; The process of employing the multiple sub-agents, each executing a sub-task according to its own sub-task instructions, and obtaining the execution result of each sub-task, includes: The task execution agent is used to execute the drone's sub-tasks according to the task execution strategy library and the task execution instructions, and obtain the execution results of the drone's sub-tasks; wherein, the task execution strategy library stores the drone's task execution step information, and the execution results of the drone's sub-tasks include: determining the drone's task to be executed according to the drone's task execution step information.

6. The method according to claim 1, characterized in that, The plurality of sub-intelligent agents also include: an emergency response intelligent agent and a question-and-answer query intelligent agent; the instructions for the plurality of sub-tasks include: emergency response instructions and question-and-answer query instructions; The process of employing the multiple sub-agents, each executing a sub-task according to its own sub-task instructions, and obtaining the execution result of each sub-task, includes: The emergency response agent executes emergency response sub-tasks based on the emergency response strategy library and emergency response instructions, and obtains the execution results of the emergency response sub-tasks; wherein, the emergency response strategy library stores emergency response strategy information for the UAV, and the execution results of the emergency response sub-tasks include: performing emergency response on the UAV based on the emergency response strategy information; And / or, using the question-and-answer query agent, according to the question-and-answer query template library and the question-and-answer query instruction, execute the question-and-answer query sub-task and obtain the execution result of the question-and-answer query sub-task; wherein, the question-and-answer query template library stores the question-and-answer query index, and the execution result of the question-and-answer query sub-task includes: according to the question-and-answer query index, querying the reply script corresponding to the user's input question.

7. The method according to claim 1, characterized in that, Before the central scheduling agent receives the user's original mission instructions for the UAV, the method further includes: The central scheduling agent is obtained by training a pre-set large language model using a pre-set central scheduling strategy library and an agent capability index library.

8. The method according to claim 2, characterized in that, The process involves using the central scheduling agent to parse the original task instructions based on a preset central scheduling strategy library and an agent capability index library, generating the task execution plan, including: The central scheduling agent is used to parse the original task instructions and the environmental data of the UAV according to the preset central scheduling strategy library and the agent capability index library to generate the task execution plan; wherein, the preset central scheduling strategy library stores the task decomposition logic and execution steps corresponding to the original task instructions, as well as the task environment data of the UAV.

9. The method according to claim 2, characterized in that, After the central scheduling agent generates the target task execution result based on the execution results of the multiple sub-tasks and feeds it back to the user, the method further includes: Receive the evaluation information returned by the user; The preset central scheduling strategy library is updated based on the evaluation information.

10. A server, characterized in that, include: The server comprises a processor (100), a storage medium (200), and a bus (300), wherein the storage medium (200) stores program instructions executable by the processor (100), and when the server is running, the processor (100) communicates with the storage medium (200) via the bus (300), and the processor (100) executes the program instructions to implement the method described in any one of claims 1 to 9.

Images