Methods, devices, equipment, media, and software products for selecting communication nodes for unmanned aerial vehicles (UAVs).

Abstract

This application discloses a method, apparatus, device, medium, and program product for selecting communication nodes for unmanned aerial vehicles (UAVs), relating to the field of UAV communication technology. The method includes: for each of a plurality of UAVs, extracting multiple semantic features from a currently acquired visual image of the UAV, and generating a semantic map based on the multiple semantic features; obtaining prior communication information based on the semantic map, the current location information of the UAV, and a large-scale communication model; obtaining the input state of the agent corresponding to the UAV based on a user task description, the semantic map, and the prior communication information, and obtaining the action intent output by the agent based on the input state, the action intent including relay selection information for instructing the UAV's next-hop relay node; wherein one UAV corresponds to one agent. The solution of this application solves the problem of unstable UAV communication links.

Description

Technical Field

[0001] This application belongs to the field of unmanned aerial vehicle (UAV) communication technology, specifically relating to a method, apparatus, equipment, medium, and program product for selecting UAV communication nodes. Background Technology

[0002] Unmanned Aerial Vehicle (UAV) systems are widely used in scenarios such as military reconnaissance, disaster monitoring, environmental patrol, and emergency communications. With the development of 5G evolution (5G-Advanced, 5G-A) and cutting-edge 6G communication technologies, UAVs can achieve wide-area rapid deployment and information relay through multi-hop networks. These networks typically rely on multi-agent reinforcement learning (MARL) algorithms to optimize dynamic communication links between multiple UAVs.

[0003] Currently, the MARL algorithm enables multiple UAV agents to make autonomous decisions based on environmental conditions (such as signal strength, distance, and energy), thereby achieving adaptive network topology maintenance. However, due to their relatively limited environmental perception dimensions, these algorithms struggle to effectively identify and model dynamic obstacles, fine terrain, or specific target features in complex airspace. Furthermore, they fail to fully consider key factors in real-world communication scenarios such as 5G-A or 6G (e.g., signal attenuation in non-line-of-sight transmission, co-channel interference, dynamic spectrum sharing, and high mobility), leading to unstable UAV communication links. Summary of the Invention

[0004] This application provides a method, apparatus, device, medium, and program product for selecting communication nodes for unmanned aerial vehicles (UAVs) to solve the problem of unstable communication links in existing technologies.

[0005] In a first aspect, embodiments of this application provide a method for selecting a UAV communication node, including:

[0006] For each of the multiple drones, multiple semantic features are extracted from the visual images currently acquired by the drone, and a semantic map is generated based on the multiple semantic features;

[0007] Based on the semantic map, the current location information of the UAV, and the large-scale communication model, prior communication information is obtained;

[0008] Based on the user task description, the semantic map, and the prior communication information, the input state of the agent corresponding to the UAV is obtained, and the action intention output by the agent based on the input state is obtained. The action intention includes relay selection information for instructing the UAV to use the next hop relay node; wherein, one UAV corresponds to one agent.

[0009] Optionally, the UAV communication node selection method includes extracting multiple semantic features from the currently acquired visual image of the UAV and generating a semantic map based on the multiple semantic features, comprising:

[0010] Using a large visual language model, multiple semantic features are extracted from the visual images currently collected by the UAV. These semantic features include terrain structure semantic features, obstacle distribution semantic features, and target point semantic features.

[0011] A semantic map is generated based on multiple semantic features and a feature fusion network.

[0012] Optionally, the UAV communication node selection method, wherein obtaining the input state of the agent corresponding to the UAV based on the user task description, the semantic map, and the communication prior information includes:

[0013] Based on the user task description, the semantic map, and the large language model, task planning information is obtained;

[0014] Based on the task planning information, the semantic map, the communication prior information, and at least one of the current position information, pose information, dynamic state, and energy state of the UAV, the input state of the intelligent agent corresponding to the UAV is obtained.

[0015] Optionally, the UAV communication node selection method further includes:

[0016] Translate the stated action intent into flight control commands;

[0017] When the flight control command is issued to the corresponding UAV, the distance between any two UAVs among the plurality of UAVs is monitored;

[0018] If the distance is less than the safe distance, the intention of the action will be modified according to the avoidance control rules.

[0019] Optionally, in the UAV communication node selection method, the prior communication information includes at least one of the following:

[0020] Feasible link prediction information; candidate relay node set; power and spectrum allocation suggestions; communication risk mask.

[0021] Optionally, in the UAV communication node selection method, the action intent further includes flight control information and / or power allocation information.

[0022] Secondly, embodiments of this application also provide a UAV communication node selection device, comprising:

[0023] A generation module is used to extract multiple semantic features from the visual image currently acquired by each of the multiple drones, and generate a semantic map based on the multiple semantic features.

[0024] The acquisition module is used to obtain prior communication information based on the semantic map, the current location information of the UAV, and the large communication model;

[0025] The selection module is used to obtain the input state of the agent corresponding to the UAV based on the user task description, the semantic map, and the communication prior information, and to obtain the action intention output by the agent based on the input state. The action intention includes relay selection information for instructing the UAV to select the next hop relay node. Each UAV corresponds to one agent.

[0026] Thirdly, embodiments of this application also provide a drone communication node selection device, including: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor executes the program or instructions to implement the drone communication node selection method as described in the first aspect.

[0027] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the UAV communication node selection method as described in the first aspect.

[0028] Fifthly, embodiments of this application also provide a computer program product, including computer instructions, which, when executed by a processor, implement the UAV communication node selection method as described in the first aspect.

[0029] Compared with existing technologies, this application provides a method, apparatus, device, medium, and program product for selecting communication nodes for unmanned aerial vehicles (UAVs). For each of a plurality of UAVs, multiple semantic features are extracted from the currently acquired visual image, and a semantic map is generated based on these features. Communication prior information is obtained based on the semantic map, the current location information of the UAV, and a large-scale communication model. The input state of the agent corresponding to the UAV is obtained based on the user task description, the semantic map, and the communication prior information. The action intention output by the agent based on the input state is also obtained, including relay selection information for instructing the UAV to select its next-hop relay node. Each UAV corresponds to one agent. This enables multi-hop communication decision-making for UAVs and solves the problem of unstable UAV communication links. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the UAV communication node selection method described in an embodiment of this application;

[0031] Figure 2 This is a flowchart illustrating one embodiment of the UAV communication node selection method described in this application.

[0032] Figure 3 This is a schematic diagram of the architecture of the application system for the UAV communication node selection method described in the embodiments of this application;

[0033] Figure 4 This is a schematic diagram of the module of the UAV communication node selection device described in the embodiments of this application;

[0034] Figure 5 This is a hardware block diagram of the UAV communication node selection device described in the embodiments of this application. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not used to describe a specified order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first object can be one or more. Furthermore, in the specification and claims, "and" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0037] Please refer to Figure 1 This application provides a method for selecting a drone communication node, including the following steps:

[0038] S101, for each of the multiple UAVs, extract multiple semantic features from the visual image currently acquired by the UAV, and generate a semantic map based on the multiple semantic features;

[0039] In one implementation, optionally, multiple semantic features are extracted from the visual image currently acquired by the UAV, and a semantic map is generated based on the multiple semantic features, including:

[0040] The Vision Language Mode (VLM) is used to extract multiple semantic features from the visual images currently acquired by the UAV. These semantic features include terrain structure semantic features, obstacle distribution semantic features, and target point semantic features.

[0041] A semantic map is generated based on multiple semantic features and a feature fusion network.

[0042] It is understood that the application scenario of this application embodiment is a multi-UAV scenario. In a multi-UAV scenario, the environment is complex and occlusions are dynamically changing. In order to improve the environmental understanding capability of communication planning, each UAV can be equipped with a lightweight visual perception unit, including a camera for real-time acquisition of visual images. First, using VLM, multiple semantic features are extracted from the visual images currently acquired by the UAV's camera, including terrain structure semantic features, obstacle distribution semantic features, and target point semantic features. Then, the extracted multiple semantic features are processed by a feature fusion network to generate a semantic map. As shown in the following formula (1):

[0043] (1);

[0044] in, Indicates the first The visual images currently being acquired by the UAV.

[0045] It should be noted that the semantic map can be used as part of the input state of the agent in the subsequent S103 and participate in the agent's decision-making. It not only provides spatial information for path planning, but also provides occlusion and reflection environment estimation for communication channel selection.

[0046] S102, based on the semantic map, the current location information of the UAV, and the large communication model, obtain prior communication information;

[0047] In this embodiment, the large-scale communication model is an industry-level model in the field of communications, possessing in-depth knowledge of 5G-A / 6G networks, link adaptation, and interference modeling. Optionally, the large-scale communication model is the Jiutian LLM (Jiutian Large Model).

[0048] Optionally, the prior communication information includes at least one of the following:

[0049] Link Feasibility Map;

[0050] Set of candidate relay nodes;

[0051] Power and spectrum allocation recommendations;

[0052] Communication risk mask.

[0053] Here, semantic maps and the current location information of the UAV Input the Nine Heavens Model, output prior communication information As shown in the following formula (2):

[0054] (2).

[0055] It should be noted that the prior communication information is used to provide communication domain knowledge guidance for the agent in S103, reduce the cost of policy exploration, and thus improve the reliability of UAV multi-hop communication.

[0056] S103, based on the user task description, the semantic map, and the communication prior information, obtain the input state of the agent corresponding to the UAV, and obtain the action intention output by the agent according to the input state. The action intention includes relay selection information for instructing the next hop relay node of the UAV; wherein, one UAV corresponds to one agent.

[0057] It should be noted that the application scenario of this application embodiment is a multi-UAV scenario. Since one UAV corresponds to one intelligent agent, this application embodiment actually adopts the MARL algorithm.

[0058] The embodiments of this application adopt a centralized training with decentralized execution (CTDE) structure to learn the relay selection strategy of UAV in three-dimensional space.

[0059] In one implementation, optionally, the input state of the agent corresponding to the UAV is obtained based on the user task description, the semantic map, and the communication prior information, including:

[0060] Based on the user task description, the semantic map, and the Large Language Model (LLM), task planning information is obtained;

[0061] The input state of the agent corresponding to the UAV is obtained based on the task planning information, the semantic map, the communication prior information, and at least one of the current position information, pose information, dynamic state and energy state of the UAV.

[0062] In this embodiment of the application, firstly, the user task description and semantic map are received via LLM. To obtain task planning information, the user task description and semantic map are input into the LLM to obtain task planning information, as shown in the following formula (3):

[0063] (3);

[0064] in, This indicates a user task description, including but not limited to: target area reconnaissance, coverage communication, and emergency relay.

[0065] It represents task planning information in sequence form, used to guide the agent in generating action intentions. It can transform abstract tasks into specific reinforcement learning objectives, bridging language to action, including but not limited to: stage goals, role division and collaboration constraints.

[0066] Then, based on the task planning information... The semantic map The prior communication information The input state of the agent corresponding to each UAV is obtained by taking at least one of the current position information, pose information, dynamic state and energy state of the UAV, as shown in the following formula (4):

[0067] (4);

[0068] in, This represents the current position and pose information of the UAV, which serves as the basic position input for trajectory planning and collision detection, and is used to calculate the relative distance to neighbors and topological relationships;

[0069] It represents the dynamic state or velocity, and is used to predict short-term trajectories, estimate energy consumption, and determine safety constraints;

[0070] It indicates the energy state or electrical quantity, and is used for task allocation, recycling strategies, and energy consumption penalties.

[0071] In one embodiment, optionally, the intention of action may also include flight control information and / or power allocation information.

[0072] In this embodiment, in addition to learning the relay selection strategy of the UAV in three-dimensional space, the flight and power distribution strategies of the UAV in three-dimensional space can also be learned. Therefore, this embodiment can learn a joint strategy for UAV flight, power distribution, and relay selection in three-dimensional space.

[0073] The intended action is shown in the following formula (5):

[0074] (5);

[0075] in, This indicates relay selection information, specifically the next-hop relay node for the UAV;

[0076] This indicates power allocation information, specifically the transmit power of the UAV;

[0077] It represents flight control information, specifically the three-dimensional spatial position of the UAV.

[0078] Specifically, in S103 above, obtaining the action intention output by the agent based on the input state includes:

[0079] Obtain the action intent output by the agent based on the input state and the reward function;

[0080] The reward function comprehensively considers communication quality, task completion, energy efficiency, and security constraints, as shown in the following formula (6):

[0081] (6);

[0082] in, For communication rewards, used to measure the large-scale communication model;

[0083] As a reward for completing the task;

[0084] As an energy reward;

[0085] As a safety reward;

[0086] Distillation rewards are used to measure the consistency between the agent's policy and the communication big model and LLM guidance signals.

[0087] It should be noted that the above model is obtained through joint training in this embodiment of the application. The training objective is to minimize the joint loss shown in the following formula (7), that is, to combine the reinforcement learning loss and the behavior cloning loss based on guided demonstration to achieve knowledge distillation-style guided learning:

[0088] (7);

[0089] in, This indicates the loss in reinforcement learning;

[0090] This represents the loss of a large-scale communication model, such as the loss of a nine-day large-scale model.

[0091] Indicates the loss of LLM;

[0092] , Indicates the weight.

[0093] In one embodiment, optionally, after step 103, the method further includes:

[0094] Translate the stated action intent into flight control commands;

[0095] When the flight control command is issued to the corresponding UAV, the distance between any two UAVs among the plurality of UAVs is monitored;

[0096] If the distance is less than the safe distance, the intention of the action will be modified according to the avoidance control rules.

[0097] In this embodiment, the action intent output by the intelligent agent can be converted into executable flight control commands, and UAV monitoring can be performed by combining feedback such as visual perception, neighboring UAV information, and security monitoring to achieve adaptive control.

[0098] Specifically, real-time constraints are applied to UAV flight control and relay selection to ensure a minimum safe distance between any two UAVs and adherence to no-fly zone constraints. If the agent's intended actions violate safety boundaries, corrections are made. This is achieved by calculating the distances between UAVs. ,when When an avoidance behavior is triggered, the safety monitor corrects the action output based on the rules. The avoidance control adopts the artificial potential field method, as shown in the following formula (8):

[0099] (8);

[0100] in, This indicates that the distance is controlled when it approaches the preset safe distance. The intensity of the repulsive force at that time, the potential field method generates a repulsive force. Achieve smooth obstacle avoidance;

[0101] The repulsive force between UAVs was calculated using the artificial potential method. Meanwhile, considering the influence of all neighboring UAVs, multiple repulsive forces are vector-superimposed to obtain the total avoidance control quantity Fi. Then, Fi is combined with the action intention output by the agent to obtain the corrected action intention.

[0102] Therefore, this embodiment monitors parameters such as distance, speed, and altitude between UAVs in real time. When a potential collision risk is detected, it triggers local obstacle avoidance behavior according to preset rules. Compared with existing optimization control methods, it requires less computation, responds faster, is easier to interpret and deploy, and can be seamlessly integrated with the action intentions output by the intelligent agent, improving the system's real-time performance and scalability while ensuring flight safety.

[0103] Figure 2 This is a flowchart illustrating one embodiment of the UAV communication node selection method described in this application. Figure 2 As shown in the embodiments of this application, the UAV communication node selection method aims to solve the problems of unstable communication, insufficient environmental awareness, weak task coordination, and poor security in existing UAV multi-hop networks. Essentially, it is an intelligent optimization method that integrates visual perception and the Jiutian industry big data model to achieve highly reliable communication and autonomous task execution for UAV swarms. Specifically, the method includes the following steps:

[0104] S201, Input, includes the visual image currently acquired by the UAV;

[0105] S202, Visual perception, extracting multiple semantic features from visual images and generating a semantic map based on these features;

[0106] S203, Communication Priors: Based on the semantic map, the current location information of the UAV, and the large-scale communication model, communication prior information is obtained.

[0107] S204, Task Planning: Obtain task planning information based on user task description, semantic map, and large language model;

[0108] S205, Multi-agent reinforcement learning control, obtains the input state of the agent corresponding to the UAV based on task planning information, semantic map, communication prior information, and at least one of the current position information, pose information, dynamic state and energy state of the UAV, and obtains the action intention output by the agent based on the input state;

[0109] S206, Execute, convert the action intention into flight control command, and when the flight control command is issued to the corresponding UAV, monitor the distance between any two UAVs among multiple UAVs. If the distance is less than the safe distance, the action intention is modified according to the avoidance control rules.

[0110] Figure 3 This is a schematic diagram of the architecture of the application system for the UAV communication node selection method described in this application embodiment. The application system for the UAV communication node selection method is the UAV communication node selection system, which can be simply referred to as the system in this application embodiment.

[0111] like Figure 3 As shown, the system includes a MARL control module, for the first... The UAV, the edge / local convergence module in the MARL control module, and the UAV's camera The connection is used to obtain the visual images currently acquired by the UAV, extract multiple semantic features from the visual images through a large visual language model, and generate a semantic map based on the multiple semantic features. The edge / local aggregation module is also used to obtain prior communication information. The edge / local aggregation module aggregates neighbor link measurements, including but not limited to Signal-to-Noise Ratio (SINR) and Received Signal Strength Indication (RSSI); the edge / local aggregation module also merges prior communication information. and task planning information ;

[0112] Obtain the input state for each agent, including at least one of the UAV's current position, pose, dynamics, and energy states, as well as a semantic map. Information on neighboring drones, prior communication information, and mission planning information.

[0113] The MARL control module adopts a CTDE structure to learn the joint strategies of UAV in three-dimensional space for flight, power distribution, and relay selection. The MARL control module interacts directly with the UAV entity, responsible for converting the agent's output action intentions into executable flight control commands, and combining feedback from visual perception, neighbor information, and safety monitoring to achieve adaptive control. The MARL control module uses a rule-based safety monitor, collision detection, and a simple avoidance logic safety layer to impose real-time constraints on flight control and relay selection.

[0114] In summary, the UAV communication node selection method described in this application is essentially a multi-agent reinforcement learning method for UAV multi-hop network optimization that integrates visual perception and the Jiutian big model. It integrates the intelligent decision-making capabilities of industry-level big models in the communications field and introduces the Jiutian big model into the UAV multi-hop communication decision-making process for the first time. By utilizing its built-in channel propagation characteristics, interference prediction model, and spectrum scheduling knowledge, it provides industry-level prior information for reinforcement learning. This mechanism can achieve accurate link quality prediction and adaptive power scheduling in dynamic channel environments, significantly improving network stability and data transmission reliability. Compared with the existing technology that uses idealized channel models or empirical parameter training, this application has the advantages of strong interpretability, fast convergence speed, and high generalization ability.

[0115] This application's embodiments introduce a visual perception unit on each UAV, using deep learning methods to identify obstacles, terrain, and target areas in the environment, generating a semantic map. This semantic map, along with prior communication information, is input into the agent, enabling the UAV to achieve autonomous path planning and task recognition in unknown and complex environments. Compared to existing technologies that only use low-dimensional state variables such as position and speed, this application's embodiments, based on high-dimensional environment modeling using visual semantic perception, possess stronger environmental understanding capabilities and robustness.

[0116] This application presents a multimodal reinforcement learning architecture that integrates vision, communication, and control layers to uniformly process multi-source information from visual sensing, communication prediction, and task semantics. This framework enables end-to-end joint decision-making for communication link optimization and task scheduling under dynamic network topology changes. This fusion mechanism effectively avoids the problems of unstable decision-making and convergence difficulties in the existing MARL algorithm under multi-source conditions, constructs a multimodal fusion cross-layer reinforcement learning framework, and improves the overall coordination of the system.

[0117] Please refer to Figure 4 This application also provides a drone communication node selection device, including:

[0118] The generation module 401 is used to extract multiple semantic features from the visual image currently acquired by each of the multiple drones, and generate a semantic map based on the multiple semantic features.

[0119] The acquisition module 402 is used to obtain prior communication information based on the semantic map, the current location information of the UAV, and the large communication model;

[0120] Selection module 403 is used to obtain the input state of the agent corresponding to the UAV based on the user task description, the semantic map and the communication prior information, and to obtain the action intention output by the agent based on the input state. The action intention includes relay selection information for instructing the UAV to the next hop relay node; wherein, one UAV corresponds to one agent.

[0121] Optionally, in the aforementioned UAV communication node selection device, the generation module 401 is specifically used for:

[0122] Using a large visual language model, multiple semantic features are extracted from the visual images currently collected by the UAV. These semantic features include terrain structure semantic features, obstacle distribution semantic features, and target point semantic features.

[0123] A semantic map is generated based on multiple semantic features and a feature fusion network.

[0124] Optionally, in the aforementioned UAV communication node selection device, the selection module 403 is specifically used for:

[0125] Based on the user task description, the semantic map, and the large language model, task planning information is obtained;

[0126] Based on the task planning information, the semantic map, the communication prior information, and at least one of the current position information, pose information, dynamic state, and energy state of the UAV, the input state of the intelligent agent corresponding to the UAV is obtained.

[0127] Optionally, the UAV communication node selection device further includes:

[0128] The conversion module is used to convert the action intent into flight control commands;

[0129] The monitoring module is used to monitor the distance between any two of the multiple drones when the flight control command is issued to the corresponding drone;

[0130] The correction module is used to correct the intention of the action according to the avoidance control rules if the distance is less than the safe distance.

[0131] Optionally, in the aforementioned UAV communication node selection device, the prior communication information includes at least one of the following:

[0132] Feasible link prediction information; candidate relay node set; power and spectrum allocation suggestions; communication risk mask.

[0133] Optionally, in the aforementioned UAV communication node selection device, the action intent further includes flight control information and / or power allocation information.

[0134] It should be noted that the UAV communication node selection device provided in this application embodiment can implement all the method steps implemented in the above UAV communication node selection method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0135] This application also provides a drone communication node selection device, such as... Figure 5 As shown, it includes:

[0136] The processor 501, memory 502, transceiver 503, and programs or instructions stored in the memory 502 and executable on the processor 501; when the processor 501 executes the programs or instructions, it implements the various processes of the above-described UAV communication node selection method embodiments and achieves the same technical effect. To avoid repetition, these will not be described again here.

[0137] The transceiver 503 is used to receive and send data under the control of the processor 501.

[0138] Among them, Figure 5 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically connecting various circuits of one or more processors represented by processor 501 and memory represented by memory 502. The bus architecture can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 503 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 504 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0139] The processor 501 is responsible for managing the bus architecture and general processing, while the memory 502 can store the data used by the processor 501 when performing operations.

[0140] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described UAV communication node selection method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0141] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described UAV communication node selection method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0142] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0144] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for selecting communication nodes for unmanned aerial vehicles (UAVs), characterized in that, include: For each of the multiple drones, multiple semantic features are extracted from the visual images currently acquired by the drone, and a semantic map is generated based on the multiple semantic features; Based on the semantic map, the current location information of the UAV, and the large-scale communication model, prior communication information is obtained; Based on the user task description, the semantic map, and the prior communication information, the input state of the agent corresponding to the UAV is obtained, and the action intention output by the agent based on the input state is obtained. The action intention includes relay selection information for instructing the UAV to use the next hop relay node; wherein, one UAV corresponds to one agent.

2. The method according to claim 1, characterized in that, Extracting multiple semantic features from the visual images currently acquired by the drone, and generating a semantic map based on the multiple semantic features, including: Using a large visual language model, multiple semantic features are extracted from the visual images currently collected by the UAV. These semantic features include terrain structure semantic features, obstacle distribution semantic features, and target point semantic features. A semantic map is generated based on multiple semantic features and a feature fusion network.

3. The method according to claim 1, characterized in that, The input state of the agent corresponding to the UAV is obtained based on the user task description, the semantic map, and the prior communication information, including: Based on the user task description, the semantic map, and the large language model, task planning information is obtained; Based on the task planning information, the semantic map, the communication prior information, and at least one of the current position information, pose information, dynamic state, and energy state of the UAV, the input state of the intelligent agent corresponding to the UAV is obtained.

4. The method according to claim 1, characterized in that, The method further includes: Translate the stated action intent into flight control commands; When the flight control command is issued to the corresponding UAV, the distance between any two UAVs among the plurality of UAVs is monitored; If the distance is less than the safe distance, the intention of the action will be modified according to the avoidance control rules.

5. The method according to claim 1, characterized in that, The prior communication information includes at least one of the following: Feasible link prediction information; candidate relay node set; power and spectrum allocation suggestions; communication risk mask.

6. The method according to claim 1, characterized in that, The intent of action also includes flight control information and / or power distribution information.

7. A UAV communication node selection device, characterized in that, include: A generation module is used to extract multiple semantic features from the visual image currently acquired by each of the multiple drones, and generate a semantic map based on the multiple semantic features. The acquisition module is used to obtain prior communication information based on the semantic map, the current location information of the UAV, and the large communication model; The selection module is used to obtain the input state of the agent corresponding to the UAV based on the user task description, the semantic map, and the communication prior information, and to obtain the action intention output by the agent based on the input state. The action intention includes relay selection information for instructing the UAV to select the next hop relay node. Each UAV corresponds to one agent.

8. A UAV communication node selection device, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor, when executing the program or instructions, implements the UAV communication node selection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the UAV communication node selection method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the UAV communication node selection method as described in any one of claims 1 to 6.

Images