Multi-unmanned aerial vehicle cooperative search and rescue intelligent decision-making method and device based on hierarchical intention

Abstract

The invention relates to a hierarchical intention-based intelligent decision-making method and device for cooperative search and rescue of multiple unmanned aerial vehicles. The method comprises the steps of obtaining an environment simulation diagram; constructing a multi-unmanned aerial vehicle cooperative search and rescue model; performing feature extraction on a search and rescue target and an obstacle through a multi-unmanned aerial vehicle cooperative search and rescue model to obtain a target position feature and an obstacle position feature; performing information gain grid division based on the environment simulation graph, and determining a local information gain field of the proxy unmanned aerial vehicle; according to the local information gain field of the agent unmanned aerial vehicle, information of a neighborhood unmanned aerial vehicle is fused for calculation, and local gain information of the agent unmanned aerial vehicle is obtained; performing feature extraction on the local gain information to obtain local gain features; performing attention focusing on the target position feature, the obstacle position feature and the local gain feature, and outputting a multi-source aggregation feature; and processing the multi-source aggregation features through a multi-layer perceptron, and outputting normal distribution of actions of the unmanned aerial vehicle. The search and rescue efficiency of the unmanned aerial vehicle in the search and rescue scene can be improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) collaborative technology, and in particular to a multi-UAV collaborative search and rescue intelligent decision-making method and device based on layered intent. Background Art

[0002] The application of multi-agent reinforcement learning (MARL) to collaborative unmanned aerial vehicle (UAV) swarm tasks, such as tracking moving targets, searching unknown areas, and avoiding obstacles, has become a hot topic of research in recent years. This research focuses on inter-agent collaboration, optimizing division of labor, adapting to partially observable environments, and making decisions in continuous action spaces within these scenarios. In recent years, MARL-based technical solutions, by combining reinforcement learning frameworks with collaborative mechanisms, have continuously improved the collaborative performance of UAV swarms, providing technical support for multi-task execution in complex and dynamic environments.

[0003] The current mainstream technical solutions are mainly divided into three categories: the first is based on the mechanism of graph neural network (GNN) or graph attention (GAT), which regards drones as graph nodes, builds edge relationships through communication range, and uses attention to aggregate neighbor information to promote collaborative decision-making. It is often combined with frameworks such as MAPPO for path planning and information sharing; the second is the role-based multi-agent reinforcement learning (MARL) solution, which solves the problem of homogeneous division of labor through dynamic role allocation, generates role representation with the help of variational inference or attention, and integrates it into the CTDE framework to realize task division; the third is the solution based on exploration mechanism, such as information gain field (IGF) or intrinsic motivation guidance, which uses quantified environmental uncertainty as reward shaping and combines algorithms such as PPO to improve drone exploration coverage and search efficiency.

[0004] However, mechanisms based on graph neural networks (GNNs) or graph attention (GAT) rely on stable communication links. Communication interruptions in dynamic or sparse environments can easily lead to the failure of the attention mechanism. Furthermore, the computational overhead of the graph structure increases quadratically with the number of drones, making it difficult to support large-scale clusters. Role-based multi-agent reinforcement learning suffers from role staticity, making role assignments prone to overfitting to local observations and poor adaptability in dynamic tasks, leading to delayed adjustments in the division of labor. Modules such as the IGF in the exploration mechanism are poorly integrated with the decision network, resulting in insufficient exploration efficiency and slow convergence in sparse reward environments. Furthermore, common flaws such as poor decision robustness in partially observable environments and insufficient scalability for heterogeneous drone swarms further restrict the practical performance of drone swarm collaborative tasks. Summary of the Invention

[0005] Based on this, it is necessary to provide a multi-UAV collaborative search and rescue intelligent decision-making method and device based on hierarchical intent that can solve the problems of collaboration failure caused by communication dependence, insufficient adaptability of role division of labor, and low integration of exploration mechanism and decision-making, in order to address the above technical problems.

[0006] A multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent, the method comprising: Obtaining an environmental simulation map, wherein the environmental simulation map includes drones, search and rescue targets, and obstacles; wherein the drones include proxy drones and neighboring drones; Build a multi-UAV collaborative search and rescue model; Extracting features of the search and rescue target and the obstacle through the multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features; Performing information gain grid division based on the environmental simulation graph, and determining the local information gain field of the proxy UAV based on the grid division; According to the local information gain field of the proxy UAV, the information of the neighboring UAVs is integrated and calculated to obtain the local gain information of the proxy UAV; the local gain information is subjected to feature extraction to obtain the local gain feature; Focusing the target position feature, the obstacle position feature, and the local gain feature, and outputting a multi-source aggregated feature; The multi-source aggregated features are processed by a multi-layer perceptron to output the normal distribution of the drone's movements.

[0007] On the other hand, a multi-UAV collaborative search and rescue intelligent decision-making device based on hierarchical intent is also provided, including: An environment simulation map acquisition module is used to acquire an environment simulation map, wherein the environment simulation map includes drones, search and rescue targets, and obstacles; wherein the drones include proxy drones and neighboring drones; Model building module, used to build a multi-UAV collaborative search and rescue model; A feature extraction module is used to extract features of the search and rescue target and the obstacle through the multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features; A local information gain module is configured to perform information gain grid division based on the environmental simulation map, determine the local information gain field of the proxy UAV based on the grid division, fuse information of neighboring UAVs based on the local information gain field of the proxy UAV, and calculate local gain information of the proxy UAV; perform feature extraction on the local gain information to obtain local gain features; A feature aggregation module, configured to focus on the target position feature, the obstacle position feature, and the local gain feature, and output a multi-source aggregated feature; The motion output module is used to process the multi-source aggregated features through a multi-layer perceptron and output the normal distribution of the UAV's motion.

[0008] Compared with the existing technology, the multi-UAV collaborative search and rescue intelligent decision-making method and device based on hierarchical intent provided by the present invention has the following effects: 1. Without relying on an external communication graph or edge computation, attention is calculated only between the agent's own intention features (exploration / tracking / obstacle avoidance / movement). This ensures stable collaboration even in scenarios with high communication failure rates, with low computational overhead, making it suitable for large-scale drone collaboration. By combining a global role overview to achieve implicit collaboration, the system avoids independent action by each agent, overcoming communication dependency and graph structure fragility, and providing dependency-free collaboration.

[0009] 2. By extracting behavioral features through intent separation and generating dynamic focus intent embeddings through attention aggregation, roles emerge naturally from end-to-end learning without the need for explicit role assignment or clustering. When faced with target evasion or obstacle changes, attention weights can be adjusted in real time to reduce overfitting of local observations, improve training stability and global collaboration, and overcome the static and poor adaptability of role division, thereby achieving dynamic implicit division of labor.

[0010] 3. Integrating the information gain field into exploration intent optimizes the balance between exploration and tracking, reduces manual intervention, and improves exploration efficiency and coverage. For some observable noise and heterogeneity, noise robustness is enhanced through intent extraction, scalability is improved through parameter sharing, and generalization is enhanced by combining the information gain field and gradient rewards, resulting in more stable performance in environments with dense obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required in the embodiments. It should be understood that the following drawings only illustrate some embodiments of the present invention, and those skilled in the art can derive other relevant drawings based on these drawings without creative work.

[0012] Figure 1 This is a flow chart of the intelligent decision-making method for multi-UAV collaborative search and rescue based on hierarchical intent provided in Example 1; Figure 2 This is a diagram of the multi-UAV collaborative search and rescue model and evaluation network structure framework provided in Example 1; Figure 3 Schematic diagram of the test and verification environment provided in Example 1; Figure 4The method of the present invention provided in Example 1 is Figure 3 Schematic diagram of capture rate in the scenario; Figure 5 The method of the present invention provided in Example 1 is Figure 3 Schematic diagram of survival rate under the scenario; Figure 6 This is a structural block diagram of the multi-UAV collaborative search and rescue intelligent decision-making device based on hierarchical intent provided in Example 2.

[0013] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "first," "second," and so on, used in this disclosure are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features being referred to. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this disclosure, "plurality" means at least two, such as two or three, unless otherwise specifically defined.

[0016] It can be understood that the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0017] The following describes the implementation of the present invention in detail with reference to the accompanying drawings in the embodiments of the present invention.

[0018] Example 1 like Figure 1 As shown, this embodiment provides a multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent, including the following steps: Step 201: Obtain an environment simulation map, which includes a drone, a search and rescue target, and obstacles.

[0019] It can be understood that when conducting collaborative search and rescue, each drone is regarded as an agent drone and its adjacent drones are regarded as neighboring drones.

[0020] Step 202: Build a multi-UAV collaborative search and rescue model.

[0021] Step 203 : extracting features of the search and rescue target and the obstacle through a multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features.

[0022] It can be understood that by extracting features from search and rescue targets and obstacles, intention separation and extraction of behavioral features are achieved, which can accurately focus on the core elements in the search and rescue scene and filter out irrelevant information in the environment; at the same time, this method makes up for the limitations of the field of view of a single drone, reduces misjudgments caused by insufficient local information, enhances the comprehensiveness and pertinence of information utilization, and reduces decision-making bias.

[0023] Step 204 : Perform information gain grid division based on the environment simulation graph, and determine the local information gain field of the proxy UAV based on the grid division.

[0024] Step 205 : Based on the local information gain field of the proxy UAV, information of neighboring UAVs is integrated and calculated to obtain the local gain information of the proxy UAV; and feature extraction is performed on the local gain information to obtain local gain features.

[0025] Gain information is incorporated into multi-UAV collaborative search and rescue, enabling UAVs to move preferentially to areas with high information value, optimizing the balance between exploration and tracking, and improving search and rescue efficiency.

[0026] Step 206 , focusing on the target position features, obstacle position features, and local gain features, and outputting multi-source aggregated features.

[0027] It can be understood that through attention focusing and dynamic focus intention embedding, roles emerge naturally from end-to-end learning without the need for explicit allocation or clustering of roles; when facing target escape or obstacle changes, attention weights can be adjusted in real time to reduce overfitting of local observations, improve training stability and global collaboration effects, so as to overcome the static and poor adaptability of role division and realize dynamic implicit division of labor.

[0028] Step 207: Process the multi-source aggregated features through a multi-layer perceptron and output the normal distribution of the drone's motion.

[0029] In one embodiment, the multi-UAV collaborative search and rescue model includes a feature extraction module, an information gain module, a first multi-head attention module, and a first multi-layer perceptron. The feature extraction module extracts features of the search and rescue target and obstacles to obtain target position features and obstacle position features.

[0030] The information gain module is used to perform information gain grid division. Based on the grid division, the local information gain field of the proxy UAV is determined. Then, according to the local information gain field of the proxy UAV, the information of the neighboring UAVs is integrated for calculation to obtain the local gain information of the proxy UAV. Feature extraction is performed on the local gain information to obtain the local gain feature.

[0031] The first multi-head attention module focuses on the target position features, obstacle position features and local gain features, and outputs multi-source aggregated features.

[0032] The multi-source aggregated features are processed by the first multi-layer perceptron to output the normal distribution of the drone's movements.

[0033] Specifically, if Figure 2 As shown in the figure, the feature extraction module includes two parallel multi-layer perceptron networks (MLP) for searching and rescuing targets. With obstacles Perform feature extraction to obtain target position features and obstacle location characteristics , the processing expression is: ; ; Where, Represents the UAV’s local observation of the search and rescue target; Represents the UAV’s local observation of the obstacle; Indicates that the output feature is 64-dimensional.

[0034] The information gain module includes a CNN network and a multi-layer perceptron network (MLP). In the information gain module, the information gain grid is first divided based on the environment simulation map. The resolution of the information gain grid is , covering a world size of Then, the local information gain field of the agent UAV is determined 、 According to the local information gain field of the agent UAV 、 , the information of the neighboring drones is integrated and calculated to obtain the proxy drone Local gain information . The local gain information Input into CNN network and multi-layer perceptron network in turn for processing to obtain local gain features , the processing expression is: .

[0035] Then, the target position features , obstacle location characteristics With local gain characteristics After splicing, the first multi-head attention module is input for attention focusing, and then the average calculation is performed to output the multi-source aggregated features.

[0036] The expression of the attention focusing process through the first multi-head attention module is: ; Where, represents the first multi-head attention module; Represents multi-source aggregate features; Indicates the number of heads for multi-head attention.

[0037] Finally, through the first multi-layer perceptron Process multi-source aggregated features and output the normal distribution of drone actions .

[0038] In one embodiment, the multi-UAV collaborative search and rescue model further includes a context state module; the context state module obtains and processes the state information of the proxy UAV and the neighboring UAVs to obtain the context state information of the proxy UAV.

[0039] After fusing the contextual state information with the multi-source aggregated features, it is processed by a second multi-layer perceptron to output the role representation of the agent drone.

[0040] Specifically, the context state module adopts a multi-layer perceptron structure.

[0041] Will act as a drone , Neighborhood Drones The state information of the agent drone is input into the context state module for processing to obtain the context state information of the agent drone. , the processing expression is: ; Then, the context state information Multi-source aggregation features For splicing, the processing expression is: ; Then, the second multi-layer perceptron (MLP2) is used to process and output the role representation of the agent drone. The processing process is expressed as: ; Where, represents fusion features; Indicates role representation.

[0042] In one embodiment, in the information gain module, information gain grid division is performed based on the environment simulation map, and based on the grid division, the local information gain field of the proxy UAV is determined, including: Information gain grid division is performed based on the environmental simulation graph, and the environmental uncertainty field and search and rescue target uncertainty field of the proxy UAV are initialized.

[0043] The environmental uncertainty field is dynamically updated to obtain the environmental field of the proxy drone; and the target field of the proxy drone is generated according to the target position.

[0044] Specifically, the local information gain field of the agent drone includes the environmental uncertainty field Uncertainty field of search and rescue targets .

[0045] First, initialize the environmental uncertainty field of the agent drone as , and the initialization search and rescue target uncertainty field is ,in, It represents the initial value and is also the maximum value of the uncertainty of the measurement environment.

[0046] Then, the environmental uncertainty field is dynamically updated, including the recovery process and the attenuation process, where the recovery process is expressed as ; The decay process is expressed as .

[0047] Then, according to the target position, a Gaussian distributed target field is generated, and the expression is: ; Where, Indicates the recovery speed of environmental uncertainty; Indicates the decay rate of environmental uncertainty; Indicates map coordinates; represents the observed target position; Represents the variance of the Gaussian distribution field.

[0048] In one embodiment, in the information gain module, the local information gain field of the proxy drone is integrated with the information of the neighboring drones to calculate and obtain the local gain information of the proxy drone. The calculation expression is: ; Where, Represents a proxy drone Local gain information; Indicates the The environmental field of the drone; Indicates the target field for drones; represents the number of drones in the neighborhood; Represents the environmental uncertainty field and the target uncertainty field at all grid points The maximum value after adding up; Represents the environmental uncertainty field and the target uncertainty field at all grid points The minimum value after adding up.

[0049] In one embodiment, when calculating the local gain information of the proxy drone, the following step is further included: calculating the uncertainty of the drone to the environment based on the environmental field and the target field, and the expression is: ; Based on the uncertainty changes before and after the update, the reward of local information gain is determined, and the expression is: ; Where, Reward representing local information gain; represents the reward coefficient; It represents the uncertainty of the drone’s environment before the update; Represents the updated UAV’s uncertainty about the environment, where .

[0050] In the above embodiment, information gain is incorporated into the multi-UAV collaborative search and rescue model. It dynamically updates the environmental uncertainty field and simulates the interaction between the environment and the target through the dynamic changes of the field. At the same time, it optimizes the exploration-tracking balance end-to-end through shaping rewards (uncertainty reduction + gradual neighbor penalty), thereby reducing manual dependence and improving efficiency.

[0051] In one embodiment, an evaluation network is further included, through which the multi-UAV collaborative search and rescue model is trained; the evaluation network includes a third multi-layer perceptron, a second multi-head attention module, and a fourth multi-layer perceptron.

[0052] Obtain the global state based on the environment simulation graph, and obtain the role representation of the agent drone.

[0053] The global state is input into the third multi-layer perceptron for processing, and the query vector is output. The role representation and the query vector are input into the second multi-head attention module for processing, and the role query perception vector is output.

[0054] After the query vector and the role query perception vector are fused, they are input into the fourth multi-layer perceptron for processing and output an evaluation value.

[0055] A value evaluation is performed on the action normal distribution output by the multi-UAV collaborative search and rescue model based on the evaluation value.

[0056] like Figure 2 As shown, first the global state Input the third multi-layer perceptron Process and output the query vector. The processing expression is .

[0057] Character Representation Set With query vector Enter the second multi-head attention module Process and output the role query perception vector. The processing expression is: .

[0058] After the query vector is fused with the role query perception vector, it is input into the fourth multi-layer perceptron for processing and the evaluation value is output. The processing process expression is: .

[0059] In one embodiment, in the evaluation network, the loss function is: ; Where, represents the number of samples; Representation sample With sample The dot product similarity between ; Representation sample With sample The dot product similarity between ; Represents the positive sample set after k-means clustering; Represents the set of negative samples after k-means clustering; The index of the agent representing the process; 、 Represents different positive sample indices.

[0060] In one embodiment, when training a multi-UAV collaborative search and rescue model through an evaluation network, a reward function is also provided, and the reward value of the UAV action is calculated according to the reward function; the reward function expression is:

[0061] in: ; ; Where, represents the total reward; Indicates the base reward; Reward representing local information gain; Indicates distance reward; Indicates time penalty; Represents the distance reward coefficient; Indicates the current step The closest target distance; Indicates the next step The closest target distance; represents the penalty at each time step.

[0062] To overcome communication dependencies and graph structure fragility, this paper shifts to internal self-attention in a multi-UAV collaborative search and rescue model. Attention is calculated only between the agent's own intent features (exploration / tracking / avoidance), without requiring an external communication graph or edge computation. This ensures independent generation of focused intent embeddings even with high communication failure rates, promoting robust decision-making with low computational overhead, making it suitable for large-scale UAVs. The evaluation network implicitly encourages collaboration through a global role overview (average focused intent), preventing degradation to independent agents.

[0063] To overcome the static nature and poor adaptability of role division, intent separation is used to extract behavioral features of drones, search and rescue targets, and obstacles. Multi-head self-attention aggregation is then used to generate dynamic focus intent embeddings. Roles emerge naturally from end-to-end learning, without the need for explicit assignment or clustering. This improves adaptability: when the target evades or obstacles change, attention weights are adjusted in real time, reducing overfitting to local observations. This results in more stable training and better global collaboration. By driving the division of labor through rewards, we avoid clustering and achieve a dynamic implicit division of labor.

[0064] Furthermore, to overcome the problems of low exploration efficiency and sparse rewards, we use the information gain field as input and integrate it into the exploration intent. We also optimize the exploration-tracking balance end-to-end through shaped rewards (uncertainty reduction + gradient neighbor penalty). This reduces manual reliance and improves efficiency.

[0065] Furthermore, the proposed method utilizes intent extraction to ensure robustness against partially observable noise and heterogeneity. Parameter sharing combined with intent attention improves scalability. Generalization is enhanced through information gain and gradient rewards. Overall, performance is more stable in environments with dense obstacles.

[0066] In one embodiment, the method proposed in the present invention is verified. The number of drones is 4 and the number of search and rescue targets is 10. Figure 3 As shown, the blue circle represents the observation range of the drone, the dark red represents obstacles, and the light red represents search and rescue targets.

[0067] like Figure 4 and Figure 5 As described above, the method provided by the present invention can achieve a capture rate of 0.9 and a survival rate of 0.98, which has significant advantages.

[0068] It should be understood that although this embodiment Figure 1The steps in the diagram are shown in the order indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0069] Example 2 Based on the multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent in Example 1, this embodiment discloses a multi-UAV collaborative search and rescue intelligent decision-making device based on hierarchical intent, such as Figure 6 As shown, the multi-UAV collaborative search and rescue intelligent decision-making device based on hierarchical intent includes: an environment simulation map acquisition module 401, a model construction module 402, a feature extraction module 403, a local information gain module 404, a feature aggregation module 405 and an action output module 406, wherein: The environment simulation map acquisition module 401 is used to obtain an environment simulation map, which includes drones, search and rescue targets, and obstacles; wherein the drones include proxy drones and neighboring drones.

[0070] The model building module 402 is used to build a multi-UAV collaborative search and rescue model.

[0071] The feature extraction module 403 is used to extract features of the search and rescue target and the obstacle through the multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features.

[0072] The local information gain module 404 is used to perform information gain grid division based on the environmental simulation map, and determine the local information gain field of the proxy UAV based on the grid division; according to the local information gain field of the proxy UAV, the information of the neighboring UAVs is integrated for calculation to obtain the local gain information of the proxy UAV; and feature extraction is performed on the local gain information to obtain the local gain feature.

[0073] The feature aggregation module 405 is used to focus on the target position features, obstacle position features and local gain features, and output multi-source aggregated features.

[0074] The motion output module 406 is used to process the multi-source aggregated features through a multi-layer perceptron and output the normal distribution of the drone's motion.

[0075] As can be understood, the hierarchical intent-based multi-UAV collaborative search and rescue intelligent decision-making device is installed on the UAV to generate action commands. The UAV is also equipped with a data acquisition module, an execution control module, and a communication module. The data acquisition module collects UAV sensor data and information gain (IGF) input; the execution control module controls the UAV flight to achieve collaborative search and rescue; and the communication module supports neighborhood information fusion.

[0076] In this embodiment, the specific working process and working principle of the environment simulation map acquisition module 401, the model construction module 402, the feature extraction module 403, the local information gain module 404 and the feature aggregation module 405 are the same as those in the method of Example 1, and therefore are not described in detail in this embodiment. Each unit module can be implemented in whole or in part by software, hardware or a combination thereof. Each unit module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above unit modules.

[0077] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0079] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent, characterized by: The method comprises: Obtaining an environmental simulation image, wherein the environmental simulation image includes a drone, a search and rescue target, and obstacles; Build a multi-UAV collaborative search and rescue model; Extracting features of the search and rescue target and the obstacle through the multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features; Performing information gain grid division based on the environmental simulation graph, and determining the local information gain field of the proxy UAV based on the grid division; According to the local information gain field of the proxy UAV, the information of the neighboring UAVs is integrated and calculated to obtain the local gain information of the proxy UAV; the local gain information is subjected to feature extraction to obtain the local gain feature; Focusing the target position feature, the obstacle position feature, and the local gain feature, and outputting a multi-source aggregated feature; The multi-source aggregated features are processed by a multi-layer perceptron to output the normal distribution of the drone's movements.

2. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 1 is characterized in that: The multi-UAV collaborative search and rescue model includes a feature extraction module, an information gain module, a first multi-head attention module and a first multi-layer perceptron; wherein, the feature extraction module extracts features of the search and rescue target and the obstacle to obtain target position features and obstacle position features; The information gain module is used to perform information gain grid division, and based on the grid division, the local information gain field of the proxy UAV is determined; then, based on the local information gain field of the proxy UAV, information of neighboring UAVs is integrated for calculation to obtain local gain information of the proxy UAV; and feature extraction is performed on the local gain information to obtain local gain features; Focusing the target position feature, the obstacle position feature, and the local gain feature through the first multi-head attention module, and outputting a multi-source aggregated feature; The multi-source aggregated features are processed by a first multi-layer perceptron to output a normal distribution of the drone's movements.

3. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 2 is characterized in that: The multi-UAV collaborative search and rescue model also includes a context state module; The context state module obtains and processes the state information of the proxy drone and the neighboring drone to obtain the context state information of the proxy drone; After fusing the context state information with the multi-source aggregated features, the information is processed by a second multi-layer perceptron to output a role representation of the proxy drone.

4. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 2 is characterized in that: In the information gain module, information gain grid division is performed based on the environment simulation map, and based on the grid division, the local information gain field of the proxy UAV is determined, including: Performing information gain grid division based on the environmental simulation graph, and initializing the environmental uncertainty field and the search and rescue target uncertainty field of the proxy UAV; The environmental uncertainty field is dynamically updated to obtain the environmental field of the proxy drone; and the target field of the proxy drone is generated according to the target position.

5. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 4 is characterized in that: In the information gain module, the local information gain field of the proxy UAV is calculated by integrating the information of the neighboring UAVs to obtain the local gain information of the proxy UAV. The calculation expression is: ； Where, Indicates map coordinates; Represents a proxy drone Local gain information; Indicates the The environmental field of the drone; Indicates the target field for drones; represents the number of drones in the neighborhood; Represents the environmental uncertainty field and the target uncertainty field at all grid points The maximum value after adding up; Represents the environmental uncertainty field and the target uncertainty field at all grid points The minimum value after adding up.

6. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 5 is characterized in that: When calculating the local gain information of the proxy UAV, it also includes: calculating the uncertainty of the UAV to the environment based on the environmental field and the target field; Based on the uncertainty changes before and after the update, the reward of local information gain is determined, and the expression is: ； Where, Reward representing local information gain; represents the reward coefficient; It represents the uncertainty of the drone’s environment before the update; Represents the updated UAV’s uncertainty about the environment.

7. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to any one of claims 1 to 6, characterized in that: Also included is an evaluation network, through which the multi-UAV collaborative search and rescue model is trained; The evaluation network includes a third multi-layer perceptron, a second multi-head attention module and a fourth multi-layer perceptron; Obtain the global state based on the environment simulation graph and obtain the role representation of the agent drone; Inputting the global state into the third multi-layer perceptron for processing, and outputting a query vector; Inputting the role representation and the query vector into the second multi-head attention module for processing, and outputting a role query perception vector; After fusing the query vector with the role query perception vector, the resultant vectors are input into a fourth multi-layer perceptron for processing, and an evaluation value is output; A value evaluation is performed on the action normal distribution output by the multi-UAV collaborative search and rescue model based on the evaluation value.

8. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 7 is characterized in that: In the evaluation network, the loss function is: ； Where, represents the number of samples; Representation sample With sample The dot product similarity between ; Representation sample With sample The dot product similarity between ; Represents the positive sample set after k-means clustering; Represents the set of negative samples after k-means clustering; The index of the agent representing the process; 、 Represents different positive sample indices.

9. The multi-UAV collaborative search and rescue intelligent decision-making method based on hierarchical intent according to claim 7 is characterized in that: When the multi-UAV collaborative search and rescue model is trained through the evaluation network, a reward function is also provided, and the reward value of the UAV action is calculated according to the reward function; the reward function expression is: ； Where, represents the total reward; Indicates the base reward; Reward representing local information gain; Indicates distance reward; Indicates time penalty.

10. A multi-UAV collaborative search and rescue intelligent decision-making device based on hierarchical intent, characterized by: The device comprises: An environment simulation map acquisition module is used to acquire an environment simulation map, wherein the environment simulation map includes drones, search and rescue targets, and obstacles; wherein the drones include proxy drones and neighboring drones; Model building module, used to build a multi-UAV collaborative search and rescue model; A feature extraction module is used to extract features of the search and rescue target and the obstacle through the multi-UAV collaborative search and rescue model to obtain target position features and obstacle position features; A local information gain module is configured to perform information gain grid division based on the environmental simulation map, determine the local information gain field of the proxy UAV based on the grid division, fuse information of neighboring UAVs based on the local information gain field of the proxy UAV, and calculate local gain information of the proxy UAV; perform feature extraction on the local gain information to obtain local gain features; A feature aggregation module, configured to focus on the target position feature, the obstacle position feature, and the local gain feature, and output a multi-source aggregated feature; The motion output module is used to process the multi-source aggregated features through a multi-layer perceptron and output the normal distribution of the UAV's motion.

Images