A collaborative planning method and device for emergency rescue heterogeneous unmanned equipment

Abstract

The present application belongs to the field of collaborative planning technology, and relates to a collaborative planning method and device for heterogeneous unmanned equipment for emergency rescue. The method includes: obtaining an emergency rescue scene, obtaining multiple types of unmanned equipment, each type of unmanned equipment includes multiple agents, and generating an overall collaborative mechanism for all unmanned equipment; generating a node representation of each agent according to the responsibilities of different agents in the emergency rescue scene; obtaining a homogeneous distribution graph and a heterogeneous collaboration graph for each agent according to the overall collaborative mechanism of all unmanned equipment; obtaining a task graph for each agent according to the responsibilities of different unmanned equipment in the emergency rescue scene; inputting the node representation of the agent into a graph neural network to obtain a node graph embedding vector; obtaining a decision vector of the unmanned equipment according to the node graph embedding vector, the homogeneous distribution graph, the heterogeneous collaboration graph, and the task graph, and using it as a collaborative planning scheme. The present application can improve the collaborative planning effect of heterogeneous unmanned equipment in emergency rescue scenes.

Description

Technical Field

[0001] The present application relates to the field of collaborative planning technology, and in particular to a collaborative planning method and device for heterogeneous unmanned equipment for emergency rescue. Background Art

[0002] With the development of cities in recent years, the scale has been expanding, the structure has become more complex, the population has become more dense, and the level of intelligence has been continuously improved. These changes have brought great safety hazards to emergency rescue. At the same time, due to the influence of various factors such as traffic congestion, dense population, complex building structure, and communication difficulties, the difficulty of emergency rescue has also increased. In the face of emergency rescue situations, the future smart city will provide a new field for the large-scale use of unmanned equipment due to its complex social environment, three-dimensional urban space, and restrictive rescue conditions. It also puts forward extremely high requirements for the reasonable planning and use of unmanned equipment.

[0003] In the field of intelligent emergency rescue, emergency rescue tasks are often complex, diverse, and subject to many constraints. Such multi-task scenarios require heterogeneous multi-agents that can perform different tasks to work together to complete the rescue mission and successfully rescue the trapped people.

[0004] Multi-agent systems are divided into homogeneous multi-agent systems and heterogeneous multi-agent systems. When in a fully cooperative scenario, the research content of homogeneous multi-agent systems focuses more on the communication, navigation, and formation behaviors of a single agent; homogeneous agents have exactly the same action space and can use the same basic behavioral constraints and common goals. However, for heterogeneous multi-agents, it is necessary not only to study the coordination within homogeneous agents, but also to consider the different action spaces or different types of tasks performed between heterogeneous agents; in most cases, the rewards of heterogeneous multi-agents are usually different and interrelated, not independent of each other, so it is difficult to reach the optimal solution at the same time.

[0005] At present, there are two main types of research on heterogeneous multi-task multi-agent systems:

[0006] One is to use a method that combines swarm intelligence algorithms with distributed decision-making and communication, such as traditional mathematical programming algorithms and evolutionary algorithms. When allocating tasks, a distributed task matching method is used to pre-set the allocation rules for various tasks, calculate the benefits of each agent and the whole, and allocate them based on the best. The advantage of this collaborative method is that it is relatively reliable. Under given circumstances, the decision of the agent has a high degree of interpretability, which enables the agent's behavior to be executed in the expected manner. Its disadvantages are also obvious. When the number of agents increases and the types of tasks increase, it becomes very difficult to formulate comprehensive and accurate matching rules, which requires a lot of time and energy, and the final effect may not meet expectations; moreover, when the situation of events changes over time, it may not be appropriate to continue to use the same set of strategies, and may even get results that are completely contrary to the expected effect. Due to the limitations of the complexity of the problem, the number of multiple agents, and the need for flexible behavior strategies, rule-based strategies are difficult to expand their research scale and applicable scenarios, which is a very meaningful direction for real problems. Therefore, in the planning method based on artificially formulated rules, the research on the collaborative method of heterogeneous multi-agents has been developing in difficulties.

[0007] The other is multi-agent reinforcement learning. The goal of reinforcement learning is to minimize the sum of the expected cumulative costs of the multi-agent system (maximize the sum of the expected cumulative rewards). From the perspective of distributed optimization, collaborative multi-agent reinforcement learning can be divided into agent modeling, communication learning and collaborative learning. These three methods also have their own advantages and disadvantages. In the agent modeling method, the agent mainly analyzes and predicts the behavior by modeling other agents. It aims to predict the behavior by agent modeling when the opponent or teammate strategy is unknown. It has strong robustness; but the computational complexity is high and accurate modeling is difficult. The communication learning-based method usually directly assumes that there is information interaction between agents, establishes a display channel, and autonomously learns the optimal strategy during the training process; but there are many parameters and the design architecture of the algorithm is relatively complex. The collaborative learning method does not require communication modeling, but makes its own decisions by directly considering the local observations and strategies of the other agents. It is relatively easy to implement and efficient; however, the application environment of this collaborative mode is relatively single. For situations with complex task logic and heterogeneous subjects, there is a lack of targeted logic combing links, and it is difficult to converge in training.

[0008] Therefore, in the research methods of heterogeneous multi-task multi-agent systems: on the one hand, the planning method based on manually formulated rules is difficult to expand its research scale and applicable scenarios due to limitations such as the complexity of the problem, the number of multiple agents, and the need for flexible behavior strategies; on the other hand, the complexity of the system in the reinforcement learning of the multi-agent system grows exponentially with the increase in the number of agents. For complex tasks and heterogeneous subjects, the algorithm lacks targeted logical combing links, and it is difficult to converge during training.

[0009] In summary, how to achieve better collaborative effects between intelligent agents remains a relatively difficult problem in this research direction. Summary of the invention

[0010] Based on this, it is necessary to provide a collaborative planning method and device for heterogeneous unmanned equipment for emergency rescue in response to the above-mentioned technical problems, which can improve the collaborative planning effect of heterogeneous unmanned equipment in emergency rescue scenarios.

[0011] A collaborative planning method for emergency rescue heterogeneous unmanned equipment, comprising:

[0012] Obtain emergency rescue scenarios, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment;

[0013] Generate node representations for each agent based on their responsibilities in emergency rescue scenarios;

[0014] According to the overall collaboration mechanism of all unmanned equipment, the local collaboration mechanism of each type of unmanned equipment is generated; according to the local collaboration mechanism of each type of unmanned equipment, the isomorphic distribution map and heterogeneous collaboration map of each intelligent agent are obtained; according to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task map of each intelligent agent is obtained;

[0015] Input the node representation of the agent into the graph neural network to obtain the node graph embedding vector;

[0016] According to the node graph embedding vector, isomorphic distribution graph, heterogeneous collaboration graph and task graph, the local value of the current intelligent agent is obtained; according to the local value of the current intelligent agent, the global value of all intelligent agents is obtained; according to the global value of all intelligent agents, the decision vector of the unmanned equipment is obtained and used as a collaborative planning scheme.

[0017] In one embodiment, the local value of the current agent is obtained according to the node graph embedding vector, the homogeneous distribution graph, the heterogeneous collaboration graph, and the task graph, including:

[0018] For the node graph embedding vector, encode an observation period and use the attention mechanism to obtain the time graph embedding vector corresponding to the snapshot of the entire input history;

[0019] According to the time graph embedding vector, homogeneous distribution graph, heterogeneous collaboration graph and task graph, the local value of the current agent is obtained.

[0020] In one embodiment, for a node graph embedding vector, an observation period is encoded, and an attention mechanism is used to obtain a time graph embedding vector corresponding to a snapshot of the entire input history, including:

[0021] For the node graph embedding vector, encode an observation period to obtain the node snapshot vector;

[0022] Different weights are assigned according to the node snapshot vector, and the attention mechanism is used to obtain graph embeddings at different times;

[0023] The graph embeddings at different times are concatenated and integrated to obtain the temporal graph embedding vector corresponding to the snapshot of the entire input history.

[0024] In one embodiment, the local value of the current agent is obtained according to the time graph embedding vector, the homogeneous distribution graph, the heterogeneous collaboration graph, and the task graph, including:

[0025] According to the time graph embedding vector and isomorphic distribution graph, the isomorphic distribution relationship value of the current agent is obtained;

[0026] According to the time graph embedding vector and the heterogeneous collaboration graph, the heterogeneous collaboration relationship value of the current intelligent agent is obtained;

[0027] According to the time graph embedding vector and task graph, the independent value of the current agent is obtained;

[0028] According to the isomorphic distribution relationship value of the current intelligent agent, the heterogeneous collaborative relationship value of the current intelligent agent and the independent value of the current intelligent agent, the local value of the current intelligent agent is obtained.

[0029] In one embodiment, based on the overall cooperation mechanism of all unmanned equipment, a local cooperation mechanism of each type of unmanned equipment is generated, including:

[0030] According to the overall collaboration mechanism of all unmanned equipment, any intelligent agent is taken as the core intelligent agent, and the isomorphic relationships between all intelligent agents in the unmanned equipment where the core intelligent agent is located and the heterogeneous relationships between the core intelligent agent and all intelligent agents in other unmanned equipment are obtained. All heterogeneous relationships are screened with attention to retain multiple heterogeneous relationships that are most closely related to the core intelligent agent.

[0031] Based on the core intelligent agent, isomorphic relationships, and heterogeneous relationships after attention screening, a local collaboration mechanism for the unmanned equipment where the core intelligent agent is located is generated;

[0032] Traverse all unmanned equipment and generate local collaboration mechanisms for various types of unmanned equipment.

[0033] In one embodiment, all heterogeneous relationships are screened for attention to retain multiple heterogeneous relationships that are most closely related to the core agent, including:

[0034] According to the responsibilities of different agents in emergency rescue scenarios, the relationship between agents is judged as emergency or non-emergency;

[0035] Attention is filtered on all heterogeneous relations, ignoring non-urgent relations between agents to retain multiple heterogeneous relations that have urgent relations with the core agent.

[0036] In one embodiment, according to the responsibilities of different agents in emergency rescue scenarios, a node representation of each agent is generated, including:

[0037] According to the responsibilities of different agents in emergency rescue scenarios, determine the node information of each agent. The node information of the agent includes: category, location, status, value and action;

[0038] Generate node representation of each agent based on the agent's node information.

[0039] In one embodiment, according to the responsibilities of different unmanned equipment in emergency rescue scenarios, a task graph of each intelligent agent is obtained, including:

[0040] According to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task layers of various unmanned equipment are obtained;

[0041] According to the task layers of various unmanned equipment, the task graph of each intelligent agent is obtained.

[0042] In one embodiment, the global value of all agents is obtained based on the local value of the current agent, including:

[0043] According to the local value of the current agent, back propagation is used to update each agent in the direction of improving the global reward to obtain the global value of all agents.

[0044] A collaborative planning device for emergency rescue heterogeneous unmanned equipment, comprising:

[0045] The first module is used to obtain emergency rescue scenarios, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment;

[0046] The second module is used to generate node representations of each agent according to the responsibilities of different agents in emergency rescue scenarios;

[0047] The third module is used to generate the local cooperation mechanism of various unmanned equipment according to the overall cooperation mechanism of all unmanned equipment; obtain the homogeneous distribution map and heterogeneous cooperation map of each intelligent agent according to the local cooperation mechanism of various unmanned equipment; obtain the task map of each intelligent agent according to the responsibilities of different unmanned equipment in emergency rescue scenarios;

[0048] The fourth module is used to input the node representation of the agent into the graph neural network to obtain the node graph embedding vector;

[0049] The fifth module is used to obtain the local value of the current intelligent agent based on the node graph embedding vector, isomorphic distribution graph, heterogeneous collaboration graph and task graph; obtain the global value of all intelligent agents based on the local value of the current intelligent agent; and obtain the decision vector of the unmanned equipment based on the global value of all intelligent agents, which is used as a collaborative planning solution.

[0050] The above-mentioned collaborative planning method and device for heterogeneous unmanned equipment for emergency rescue generates the overall collaboration mechanism of all unmanned equipment and the local collaboration mechanism of various types of unmanned equipment according to complex task scenarios (for example: emergency rescue scenarios), and generates the node representation of each agent according to the responsibilities of different agents in emergency rescue scenarios, and obtains the homogeneous distribution graph, heterogeneous collaboration graph and task graph of each agent to build a relationship network based on the graph structure; then, the graph neural network is used to embed the node representation of the agent, and the attention mechanism is integrated to filter key information to generate a dynamic heterogeneous graph neural network agent collaboration mechanism, obtain the global value of all agents, and obtain a collaborative planning scheme. This application can solve the problems of multi-agent collaboration and multi-task joint, effectively improve the multi-agent collaboration effect, and has rich research space and broad development prospects in the fields of artificial intelligence and intelligent systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is an application scenario diagram of a collaborative planning method for heterogeneous unmanned equipment for emergency rescue in one embodiment;

[0052] Figure 2 A flowchart of a collaborative planning method for heterogeneous unmanned equipment for emergency rescue in one embodiment;

[0053] Figure 3 A diagram of the overall cooperation mechanism of all unmanned equipment in one embodiment;

[0054] Figure 4 A diagram of a local cooperation mechanism of an unmanned obstacle removal device in one embodiment;

[0055] Figure 5 A diagram of a local collaboration mechanism for mining unmanned equipment in one embodiment;

[0056] Figure 6 A diagram of a local cooperation mechanism for transporting unmanned equipment in one embodiment;

[0057] Figure 7 A task layer diagram of an obstacle removal unmanned equipment in an embodiment;

[0058] Figure 8 A task layer diagram for mining unmanned equipment and transporting unmanned equipment in one embodiment;

[0059] Fig. 9 A schematic diagram of the architecture of a collaborative planning method for heterogeneous unmanned equipment for emergency rescue in one embodiment;

[0060] Fig.10 It is a structural block diagram of a collaborative planning device for emergency rescue heterogeneous unmanned equipment in one embodiment;

[0061] Fig.11 A schematic diagram of an unmanned obstacle removal device in an embodiment;

[0062] Fig.12 A schematic diagram of an unmanned mining device in one embodiment;

[0063] Fig.13 A schematic diagram of transporting unmanned equipment in one embodiment;

[0064] Fig.14 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0066] In addition, the descriptions of "first", "second", etc. in this application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "multiple groups" means at least two groups, such as two groups, three groups, etc., unless otherwise clearly and specifically defined.

[0067] In this application, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, a physical connection, or a wireless communication connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0068] In addition, the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0069] The collaborative planning method of heterogeneous unmanned emergency rescue equipment provided in this application can be applied to Figure 1 In the application scenario diagram shown, the terminal 102 communicates with the server 104 through the network, and the terminal 102 may include but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 104 may be various portal websites, servers corresponding to the backend of the work system, etc.

[0070] This application provides a collaborative planning method for heterogeneous unmanned emergency rescue equipment, such as Figure 2 The flowchart shown in FIG. 1 shows a method for applying the method to Figure 1 The following is an example of a terminal in the example, including:

[0071] Step 201, acquire an emergency rescue scenario, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment.

[0072] In this step, in common emergency rescue incidents, the rescue forces are mainly the police system, fire system, and medical system.

[0073] For example: in an earthquake disaster, the wounded after the disaster are mainly in collapsed buildings; when taking rescue operations, the police are needed to help clear and evacuate the roads, and then the firefighters are responsible for digging out the buried people from the collapsed buildings, and then the medical staff take over and send the wounded to shelters or hospitals with medical resources; the whole process involves three types of unmanned equipment intelligent bodies, which work together to complete post-disaster rescue.

[0074] In the unmanned equipment of the intelligent emergency rescue environment, the role of the police corresponds to the unmanned equipment for obstacle removal, the role of the firefighter corresponds to the unmanned equipment for excavation, and the role of the medical staff corresponds to the unmanned equipment for transportation. The overall collaboration mechanism of all unmanned equipment generated is shown in the figure below: Figure 3 shown.

[0075] It should be noted that how to generate an overall collaboration mechanism for all unmanned equipment based on emergency rescue scenarios belongs to the existing technology and will not be elaborated here.

[0076] Step 202: Generate a node representation of each intelligent agent according to the responsibilities of different intelligent agents in the emergency rescue scenario.

[0077] Specifically:

[0078] According to the responsibilities of different agents in emergency rescue scenarios, determine the node information of each agent. The node information of the agent includes: category, location, status, value and action;

[0079] Generate node representation of each agent based on the agent's node information.

[0080] More specifically:

[0081] Each agent has different responsibilities, and the information they carry is also different. Each agent, as a node, contains information such as category (wounded or police or fire or medical or mission point), location (coordinates and direction), status (emergency or non-emergency or unknown), value (referring to the value to be developed), and action (the destination to be traveled to).

[0082] The police agent is mainly responsible for clearing the road in post-disaster rescue. Therefore, the main task of the police agent is to assist other agents in clearing the road when needed, and to help find and rescue the injured more quickly. Through analysis, it can be seen that the police agent is always on the move and is the one who takes the initiative to take action instead of waiting to be developed by other agents. Therefore, no value to be developed is set.

[0083] The fire agent is mainly responsible for digging out the wounded buried by the ruins of buildings in post-disaster rescue. In this process, the fire agent may need the police agent to clear the road to reach the disaster site, and the medical agent may also need to transport the rescued wounded to the hospital for treatment. Therefore, the adjacency relationship of the fire agent is relatively complex, so the attribute of untapped value is set to limit the importance of key nodes in the task process. Rescue tasks may occur one after another. When the fire agent requests the police agent to come to the rescue, the current position of the fire agent becomes the new position of the police agent after the police agent completes the task. The nearby road conditions, the wounded that may be found, and the potential collaboration relationship are all new adjacency relationships that the police agent takes over from the fire agent. These relationships determine the benefits of the police agent in the next task cycle or even several subsequent task cycles. Therefore, the untapped value is also an important basis for determining the collaboration between agents.

[0084] Medical agents are mainly responsible for transporting the wounded to hospitals in post-disaster rescue operations. Therefore, medical agents often maintain a collaborative formation with fire-fighting agents. The buried wounded dug out by fire-fighting agents urgently need medical treatment to survive. An effective collaborative formation can greatly shorten the rescue time.

[0085] Therefore, the i Nodes The node representation is expressed in symbolic language as follows:

[0086] ;

[0087] Among them, type corresponds to the node category, and the value can be 0, 1, 2, 3, 4, which respectively represent the wounded (Civil), police (Police), fire (Fireman), medical care (Medical stuff), and task points (Task). The corresponding type of node is represented as , , , , .

[0088] pos corresponds to the node position, and the value represents the coordinates of the current agent's location. If there are two agents, their relative position relationship can be calculated through the pos coordinates.

[0089] state corresponds to the node state, and the value can be 0, 1, or 2, representing emergency, non-emergency, and unknown, respectively. The meaning of emergency state is that the current agent is waiting for other agents to assist in rescue because it has found trapped injured persons, or the current agent is waiting for other agents to rescue it because its own actions are blocked. The meaning of non-emergency state is that the current agent can take normal and independent actions; when the current node is a task point, its state is directly defined as non-emergency state. The meaning of unknown state is that the agent is not discovered or cannot receive and send messages.

[0090] The value corresponds to the value of the node to be developed. The initial value is zero. As the agent continues to act, it completes and supplements the surrounding information and is updated through network calculation. This information mainly focuses on the number of agents in emergency situations, the number of wounded people, the number of mission points, etc.

[0091] Action corresponds to the node action, and the value means the number of the destination that the agent intends to go to. Only the agent will take action, and the value of this item for nodes with the task point type is set to 0.

[0092] In this step, by sorting out the information required for coordination between multiple tasks, a systematic process framework for emergency rescue is constructed, and then the factors involved in the task execution process are further refined and supplemented, the information is screened and integrated, and then the task process nodes are sorted out.

[0093] Step 203, based on the overall collaboration mechanism of all unmanned equipment, generate the local collaboration mechanism of each type of unmanned equipment; based on the local collaboration mechanism of each type of unmanned equipment, obtain the isomorphic distribution map and heterogeneous collaboration map of each intelligent agent; based on the responsibilities of different unmanned equipment in emergency rescue scenarios, obtain the task map of each intelligent agent.

[0094] Specifically:

[0095] According to the overall collaboration mechanism of all unmanned equipment, any intelligent agent is taken as the core intelligent agent, and the isomorphic relationships between all intelligent agents in the unmanned equipment where the core intelligent agent is located and the heterogeneous relationships between the core intelligent agent and all intelligent agents in other unmanned equipment are obtained. All heterogeneous relationships are screened with attention to retain multiple heterogeneous relationships that are most closely related to the core intelligent agent.

[0096] Based on the core intelligent agent, isomorphic relationships, and heterogeneous relationships after attention screening, a local collaboration mechanism for the unmanned equipment where the core intelligent agent is located is generated;

[0097] Traverse all unmanned equipment and generate local cooperation mechanisms for various types of unmanned equipment;

[0098] According to the local cooperation mechanism of various unmanned equipment, the homogeneous distribution map and heterogeneous cooperation map of each intelligent agent are obtained;

[0099] According to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task graph of each intelligent agent is obtained.

[0100] More specifically:

[0101] According to the overall collaboration mechanism of all unmanned equipment, any intelligent agent is taken as the core intelligent agent, and the isomorphic relationship between all intelligent agents in the unmanned equipment where the core intelligent agent is located and the heterogeneous relationship between the core intelligent agent and all intelligent agents in other unmanned equipment are obtained; according to the responsibilities of different intelligent agents in emergency rescue scenarios, the relationship between intelligent agents is judged to be emergency or non-emergency; all heterogeneous relationships are screened for attention, and non-emergency relationships between intelligent agents are ignored to retain multiple heterogeneous relationships that have emergency relationships with the core intelligent agent;

[0102] Based on the core intelligent agent, isomorphic relationships, and heterogeneous relationships after attention screening, a local collaboration mechanism for the unmanned equipment where the core intelligent agent is located is generated;

[0103] Traverse all unmanned equipment and generate local cooperation mechanisms for various types of unmanned equipment;

[0104] According to the local cooperation mechanism of various unmanned equipment, the homogeneous distribution map and heterogeneous cooperation map of each intelligent agent are obtained;

[0105] According to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task layers of various unmanned equipment are obtained; according to the task layers of various unmanned equipment, the task graph of each intelligent agent is obtained.

[0106] In this step, the core agent refers to the agent for which divergence and extraction are performed.

[0107] Attention filtering makes local observations more valuable and less complex.

[0108] It should be noted that for the local collaboration mechanism diagram, various types of unmanned equipment, core intelligent entities, homogeneous relationships and heterogeneous relationships can be marked with different colors, line types, etc. to distinguish different equipment, relationships, etc.

[0109] like Figure 4 The local cooperation mechanism diagram of the obstacle removal unmanned equipment is shown in Figure 5 The local collaboration mechanism diagram of mining unmanned equipment is shown in Figure 6 Diagram of the local collaboration mechanism for transporting unmanned equipment is shown.

[0110] There needs to be a certain distribution relationship between similar agents. When there are many and scattered tasks, the distribution of agents also needs to be dispersed accordingly. When the task is in the later stage and the task points are relatively concentrated, the injuries of the wounded are aggravated and time becomes more urgent, so multiple agents are needed to complete a task together to speed up rescue. The task of the police agent determines that he only pays attention to agents in emergency status. Only when the state of the agent changes to an emergency state will the police agent have an adjacent relationship with it. The adjacency relationship of the fire agent is much more complicated. When the fire agent is in a non-emergency state, it only has an adjacent relationship with the medical agent, maintaining a collaborative formation.

[0111] like Figure 7 The task layer diagram of the unmanned obstacle removal equipment shown in the figure is that for the police agent, its task layer is the road waiting to be cleared, which is the shaded part in the figure.

[0112] like Figure 8 As shown in the task layer diagram of the excavation unmanned equipment and the transportation unmanned equipment, for the fire fighting and medical intelligent agents, their task layer is the building to be searched and rescued, that is, the solid square part in the figure.

[0113] Step 204, input the node representation of the agent into the graph neural network to obtain a node graph embedding vector.

[0114] In this step, at each moment t, the current unmanned equipment has a distribution state, and this distribution state is regarded as a snapshot at the current moment, where the unmanned equipment is regarded as a node. Then, each unmanned equipment can be expressed in a symbolic language based on the node.

[0115] The node The node representation of the input graph neural network first enters the encoder, and the intermediate hidden layer vector is calculated layer by layer through the node and edge information to obtain the node The node graph embedding vector .

[0116] ;

[0117] ;

[0118] in, Is a node The k-th hop hidden state, is a positive integer and , is the neighborhood information aggregation function, are all nodes The set of connected edges, Is a node The set of all adjacent nodes, The last output node The node graph embedding vector of (in the graph neural network, for the graph node information, the initial information of each node is , assuming that each node information can be propagated to adjacent nodes with attenuation Step nodes, information propagation superposition calculation is performed on each node, each node information on the graph is updated, and the node information between each other is fused, so the fused node graph representation is obtained, that is, the node graph embedding vector ), is the hidden layer vector at the current moment (also the integrated node information representation vector), for example: Right now hour, .

[0119] Step 205, based on the node graph embedding vector, the isomorphic distribution graph, the heterogeneous collaboration graph and the task graph, obtain the local value of the current intelligent agent; based on the local value of the current intelligent agent, obtain the global value of all intelligent agents; based on the global value of all intelligent agents, obtain the decision vector of the unmanned equipment and use it as a collaborative planning solution.

[0120] Specifically:

[0121] For the node graph embedding vector, encode an observation period and use the attention mechanism to obtain the time graph embedding vector corresponding to the snapshot of the entire input history; according to the time graph embedding vector, homogeneous distribution graph, heterogeneous collaboration graph and task graph, obtain the local value of the current agent;

[0122] According to the local value of the current agent, the global value of all agents is obtained;

[0123] According to the global value of all intelligent agents, the decision vector of the unmanned equipment is obtained and used as a collaborative planning solution.

[0124] More specifically:

[0125] For the node graph embedding vector, encode an observation period to obtain the node snapshot vector; according to the node snapshot vector, perform different weight assignments and use the attention mechanism to obtain the graph embedding at different times; splice and integrate the graph embeddings at different times to obtain the time graph embedding vector corresponding to the snapshot of the entire input history;

[0126] According to the time graph embedding vector and the isomorphic distribution graph, the isomorphic distribution relationship value of the current agent is obtained; according to the time graph embedding vector and the heterogeneous collaboration graph, the heterogeneous collaboration relationship value of the current agent is obtained; according to the time graph embedding vector and the task graph, the independent value of the current agent is obtained; according to the isomorphic distribution relationship value of the current agent, the heterogeneous collaboration relationship value of the current agent, and the independent value of the current agent, the local value of the current agent is obtained;

[0127] Based on the local value of the current agent, back propagation is used to update each agent in the direction of improving the global reward to obtain the global value of all agents;

[0128] According to the global value of all intelligent agents, the decision vector of the unmanned equipment is obtained and used as a collaborative planning solution.

[0129] In this step, for the node The node graph embedding vector , encoding an observation period , get the node snapshot vector :

[0130] ;

[0131] According to the node snapshot vector, different weights are assigned, and the attention mechanism is used to obtain the graph embedding at different times to input the attention mechanism ( ) to allocate attention, use three trainable parameter matrices to enhance the model’s fitting ability, and enhance the stability of training gradient updates; concatenate and integrate graph embeddings at different times to obtain the time graph embedding vector corresponding to the snapshot of the entire input history:

[0132] ;

[0133] in:

[0134] ;

[0135] In the formula, , , are different weight matrices for weight distribution; , , For the same input matrix The parameter matrix obtained by linear transformation with different weight matrices; Embedding graphs at different times; For the entire input history The temporal graph embedding vector corresponding to the snapshot; softmaxIt is a normalization function, so that the results become values ​​between 0 and 1, and the sum of all values ​​is 1, that is, the result is turned into a vector of assigned weights; yes The dimension size of the attention mechanism is to redistribute the weight of the input matrix, filter out some information, and retain important information. At present, the attention mechanism has a mature encapsulation package, and the user only needs to provide the input node representation vector.

[0136] According to the time graph embedding vector and the isomorphic distribution graph, the isomorphic distribution relationship value of the current agent is obtained; according to the time graph embedding vector and the heterogeneous collaboration graph, the heterogeneous collaboration relationship value of the current agent is obtained; according to the time graph embedding vector and the task graph, the independent value of the current agent is obtained:

[0137] ;

[0138] In the formula, is the isomorphic distribution relationship value of the current agent, is the heterogeneous collaborative relationship value of the current intelligent agent, is the independent value of the current agent, and They are different approximate value functions implemented by neural networks, are the neural network parameters of the agent homogeneous distribution relational value network, is the neural network parameter of the value network of the heterogeneous collaborative relationship of intelligent agents, is the neural network parameter of the agent's independent value network. The input of the network is the agent Action and local observation The multidimensional representation vector ; Represents the representation vector of homogeneous agents in the local collaboration graph, A set of representation vectors representing all heterogeneous agents in a heterogeneous network;

[0139] According to the current agent's isomorphic distribution relationship value, the current agent's heterogeneous collaboration relationship value, and the current agent's independent value, the local value of the current agent is obtained:

[0140] ;

[0141] In the formula, is the local value of the current agent, Representing an Agent For Agents The attention weight, Representing an Agent For Agents The attention weight of the agent With Agent There are heterogeneous relationships, agents With Agent There is an isomorphic relationship;

[0142] According to the local value of the current agent, back propagation is used to allow each agent to update the network in the direction of improving the global reward, completing the implicit credit allocation process to further improve the global cooperation ability of agents in the multi-agent system and obtain the global value of all agents:

[0143] ;

[0144] In the formula, is the global value of all agents, is the number of agents;

[0145] According to the global value of all agents, the decision vector of the unmanned equipment is obtained and used as a collaborative planning scheme:

[0146] ;

[0147] In the formula, is the decision vector of the unmanned equipment.

[0148] It should be noted that the value of the collaboration and distribution relationship between agents is calculated and weighted summed to represent the collaboration between agents in the current state for prediction. The importance of the value. The value (Q value) of the agent describes the value of each action selected by the agent under the current state and observation. The higher the value corresponding to the action, the more rewards the agent can get in the current or even future if it judges that taking this action under the current circumstances.

[0149] In this embodiment, if Fig. 9 The schematic diagram of the collaborative planning method architecture for heterogeneous unmanned emergency rescue equipment is shown in the figure. Taking the emergency rescue scene of smart cities as the background, the multi-agent collaborative planning and decision-making problems of heterogeneous unmanned equipment are studied. Compared with the existing technology, the defect that the multi-agent collaborative planning method is difficult to apply to complex task scenarios is solved. The specific contents include:

[0150] (1) In complex scenarios, graph neural networks are input based on graph structures (graph data structures), multi-agent relationship networks are represented, multi-agent collaboration mechanisms (cooperation mechanisms) are generated, multi-task collaboration relationships are analyzed, and the spatial topological properties of the graph structure are used to describe the characteristics of complex tasks. Important nodes are inferred from multi-agent relationships and multi-agent collaboration mechanisms.

[0151] (2) Based on the traditional task matching method, the chain effect of tasks is considered. The topological characteristics of the relationship network are learned based on the graph neural network. The features of the designed collaboration mechanism are extracted. The time spent on completing the task, resource consumption, success rate, communication cost, etc. are considered. The implicit relationships between related intelligent agents on different task nodes are deeply inferred and mined to predict the implicit collaboration relationship.

[0152] (3) Using dynamic heterogeneous graph neural networks, we integrate the node embedding information of the explicit and implicit relationship networks of multi-agents to improve the algorithm convergence ability and expand the scale of multi-agent training.

[0153] This application improves and optimizes the planning method of heterogeneous unmanned equipment by extracting key nodes from the multi-agent collaborative cooperation mechanism in complex task scenarios using graph neural networks. It focuses on effective information through the attention mechanism, greatly improving the collaborative effect of heterogeneous unmanned equipment in complex task scenarios.

[0154] The above-mentioned collaborative planning method for heterogeneous unmanned equipment for emergency rescue generates the overall collaboration mechanism of all unmanned equipment and the local collaboration mechanism of various types of unmanned equipment according to complex task scenarios (for example, emergency rescue scenarios), and generates node representations of each agent according to the responsibilities of different agents in emergency rescue scenarios, and obtains the homogeneous distribution graph, heterogeneous collaboration graph and task graph of each agent to build a relationship network based on the graph structure; then, the graph neural network is used to embed the node representation of the agent, and the attention mechanism is integrated to filter key information to generate a dynamic heterogeneous graph neural network agent collaboration mechanism, obtain the global value of all agents, and obtain a collaborative planning solution. This application can solve the problems of multi-agent collaboration and multi-task joint, effectively improve the effect of multi-agent collaboration, and has rich research space and broad development prospects in the fields of artificial intelligence and intelligent systems.

[0155] Specifically, the present application has the following beneficial effects:

[0156] 1. Aiming at the collaborative relationship in complex task scenarios, a multi-layer overlay method (overlay of homogeneous distribution graph, heterogeneous collaboration graph and task graph) is proposed to analyze and design a collaborative mechanism based on graph structure, which can comprehensively consider from different dimensions, improve scalability and generalization, and make the collaborative effect better and smarter.

[0157] 2. Most tasks in the field of emergency rescue require relays from various sectors of society. The current task allocation collaboration mechanism may have limitations. Graph neural networks take into account the chain effects of actions, optimize task paths, and can use large-scale data to generate implicit relationships (implicit relationships refer to implicit communication among multiple agents, which is different from explicit communication between agents that directly exchange information through message passing. Agents in implicit communication obtain information by observing the behavior of other agents or changes in the environment, and combine their own observed information and inferred information to infer the relationship between themselves and the surrounding agents), thereby achieving better collaboration between multi-agent systems.

[0158] 3. For heterogeneous multi-agent collaborative networks in complex task scenarios, a collaborative mechanism based on node representation is designed, which can cover the types of agents, various cost factors of tasks, and the impact of action chains.

[0159] 4. The existing mapping methods based on graph neural networks tend to accept more possibilities when building maps, but this situation brings advantages while it cannot avoid generating many connection relationships that do not conform to the real logic, thus greatly reducing the computing efficiency. This application adopts a joint mapping method of rule mapping and attention mechanism subgraph extraction, combining the addition of a rule mapping method with hard constraints with the method of extracting subgraphs using an attention mechanism, which can take into account both hard constraints (hard constraints refer to: constraints that must be met in reality, such as: the accessible roads in the urban road network are certain, and the intelligent body is not 360° movable) and soft constraints (soft constraints refer to: expectations that are met as much as possible, such as: A and B cooperation will bring greater benefits to the whole than A and C cooperation, so it is expected that A chooses to cooperate with B), effectively limiting the generation of edges that violate hard constraints, improving the subsequent computing efficiency of the algorithm, and making the generated graph neural network more interpretable.

[0160] It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as 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. Moreover, Figure 2 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.

[0161] The present application also provides a collaborative planning device for emergency rescue heterogeneous unmanned equipment, such as Fig.10As shown, in one embodiment, it includes: a first module 1001, a second module 1002, a third module 1003, a fourth module 1004 and a fifth module 1005, wherein:

[0162] The first module 1001 is used to obtain emergency rescue scenarios, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment;

[0163] The second module 1002 is used to generate a node representation of each intelligent agent according to the responsibilities of different intelligent agents in the emergency rescue scenario;

[0164] The third module 1003 is used to generate the local cooperation mechanism of each type of unmanned equipment according to the overall cooperation mechanism of all unmanned equipment; obtain the homogeneous distribution map and heterogeneous cooperation map of each intelligent agent according to the local cooperation mechanism of each type of unmanned equipment; obtain the task map of each intelligent agent according to the responsibilities of different unmanned equipment in emergency rescue scenarios;

[0165] The fourth module 1004 is used to input the node representation of the agent into the graph neural network to obtain a node graph embedding vector;

[0166] The fifth module 1005 is used to obtain the local value of the current intelligent agent based on the node graph embedding vector, the isomorphic distribution graph, the heterogeneous collaboration graph and the task graph; obtain the global value of all intelligent agents based on the local value of the current intelligent agent; obtain the decision vector of the unmanned equipment based on the global value of all intelligent agents, and use it as a collaborative planning solution.

[0167] In a specific embodiment, the unmanned equipment includes Fig.11 The unmanned obstacle removal equipment shown, Fig.12 The unmanned excavation equipment shown and Fig.13 In the transport unmanned equipment shown, the role of the police corresponds to the obstacle removal unmanned equipment, the role of the firefighter corresponds to the excavation unmanned equipment, and the role of the medical staff corresponds to the transport unmanned equipment, where:

[0168] 1 is a signal transceiver box carried by the unmanned equipment, which is used to receive information from the main station and signals from nearby intelligent entities, and to make plans and execute decisions.

[0169] 2 is a 3D camera carried by the unmanned equipment, which is used to observe the environment and perform calibration when performing actions.

[0170] 3 is the multi-line laser radar carried by the unmanned equipment, which is used to supplement the information that the 3D camera cannot observe.

[0171] 4 is a road rapid obstacle removal module carried by the unmanned obstacle removal equipment, which is used for rapid road obstacle removal.

[0172] 5 is an unmanned robotic arm, which is used to accurately control the position of the device carried at the end of the robotic arm.

[0173] 6 is an unmanned manipulator for gripping objects.

[0174] 7 is a retractable drill carried by the unmanned excavation equipment, which is used to break through the surface rubble when the wounded are buried, and cooperate with the manipulator of the unmanned excavation equipment to remove obstacles and thus find the wounded.

[0175] 8 is a patient fixation module carried by the unmanned transport equipment, which is used to protect the wounded from secondary injuries during the journey to the medical point.

[0176] It should be noted that the above structures 1 to 8 are all prior art and will not be described in detail here.

[0177] For the specific definition of a collaborative planning device for heterogeneous unmanned equipment for emergency rescue, please refer to the definition of a collaborative planning method for heterogeneous unmanned equipment for emergency rescue mentioned above, which will not be repeated here. Each module in the above device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0178] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.14 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a collaborative planning method for heterogeneous unmanned equipment for emergency rescue is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0179] Those skilled in the art will understand that Fig.14The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0180] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in the above embodiment when executing the computer program.

[0181] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the above embodiment are implemented.

[0182] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and 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 (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0183] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0184] The technical features of the above embodiments may 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.

[0185] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A collaborative planning method for heterogeneous unmanned equipment for emergency rescue, characterized in that: include: Obtain emergency rescue scenarios, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment; Generate node representations for each agent based on their responsibilities in emergency rescue scenarios; According to the overall collaboration mechanism of all unmanned equipment, the local collaboration mechanism of each type of unmanned equipment is generated; according to the local collaboration mechanism of each type of unmanned equipment, the isomorphic distribution map and heterogeneous collaboration map of each intelligent agent are obtained; according to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task map of each intelligent agent is obtained; Input the node representation of the agent into the graph neural network to obtain the node graph embedding vector; According to the node graph embedding vector, isomorphic distribution graph, heterogeneous collaboration graph and task graph, the local value of the current intelligent agent is obtained; according to the local value of the current intelligent agent, the global value of all intelligent agents is obtained; according to the global value of all intelligent agents, the decision vector of the unmanned equipment is obtained and used as a collaborative planning scheme.

2. The collaborative planning method for emergency rescue heterogeneous unmanned equipment according to claim 1 is characterized in that: According to the node graph embedding vector, homogeneous distribution graph, heterogeneous collaboration graph and task graph, the local value of the current agent is obtained, including: For the node graph embedding vector, encode an observation period and use the attention mechanism to obtain the time graph embedding vector corresponding to the snapshot of the entire input history; According to the time graph embedding vector, homogeneous distribution graph, heterogeneous collaboration graph and task graph, the local value of the current agent is obtained.

3. The collaborative planning method for emergency rescue heterogeneous unmanned equipment according to claim 2 is characterized in that: For the node graph embedding vector, encode an observation period and use the attention mechanism to obtain the time graph embedding vector corresponding to the snapshot of the entire input history, including: For the node graph embedding vector, encode an observation period to obtain the node snapshot vector; Different weights are assigned according to the node snapshot vector, and the attention mechanism is used to obtain graph embeddings at different times; The graph embeddings at different times are concatenated and integrated to obtain the temporal graph embedding vector corresponding to the snapshot of the entire input history.

4. The collaborative planning method for emergency rescue heterogeneous unmanned equipment according to claim 3 is characterized in that: According to the time graph embedding vector, homogeneous distribution graph, heterogeneous collaboration graph and task graph, the local value of the current agent is obtained, including: According to the time graph embedding vector and isomorphic distribution graph, the isomorphic distribution relationship value of the current agent is obtained; According to the time graph embedding vector and the heterogeneous collaboration graph, the heterogeneous collaboration relationship value of the current intelligent agent is obtained; According to the time graph embedding vector and task graph, the independent value of the current agent is obtained; According to the isomorphic distribution relationship value of the current intelligent agent, the heterogeneous collaborative relationship value of the current intelligent agent and the independent value of the current intelligent agent, the local value of the current intelligent agent is obtained.

5. A collaborative planning method for emergency rescue heterogeneous unmanned equipment according to any one of claims 1 to 4, characterized in that: Based on the overall collaboration mechanism of all unmanned equipment, the local collaboration mechanism of various types of unmanned equipment is generated, including: According to the overall collaboration mechanism of all unmanned equipment, any intelligent agent is taken as the core intelligent agent, and the isomorphic relationships between all intelligent agents in the unmanned equipment where the core intelligent agent is located and the heterogeneous relationships between the core intelligent agent and all intelligent agents in other unmanned equipment are obtained. All heterogeneous relationships are screened with attention to retain multiple heterogeneous relationships that are most closely related to the core intelligent agent. Based on the core intelligent agent, isomorphic relationships, and heterogeneous relationships after attention screening, a local collaboration mechanism for the unmanned equipment where the core intelligent agent is located is generated; Traverse all unmanned equipment and generate local collaboration mechanisms for various types of unmanned equipment.

6. The collaborative planning method for emergency rescue heterogeneous unmanned equipment according to claim 5 is characterized in that: Attention is filtered on all heterogeneous relations to retain multiple heterogeneous relations that are most closely related to the core agent, including: According to the responsibilities of different agents in emergency rescue scenarios, the relationship between agents is judged as emergency or non-emergency; Attention is filtered on all heterogeneous relations, ignoring non-urgent relations between agents to retain multiple heterogeneous relations that have urgent relations with the core agent.

7. A collaborative planning method for emergency rescue heterogeneous unmanned equipment according to any one of claims 1 to 4, characterized in that: According to the responsibilities of different agents in emergency rescue scenarios, a node representation of each agent is generated, including: According to the responsibilities of different agents in emergency rescue scenarios, determine the node information of each agent. The node information of the agent includes: category, location, status, value and action; Generate node representation of each agent based on the agent's node information.

8. A collaborative planning method for emergency rescue heterogeneous unmanned equipment according to any one of claims 1 to 4, characterized in that: According to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task graph of each agent is obtained, including: According to the responsibilities of different unmanned equipment in emergency rescue scenarios, the task layers of various unmanned equipment are obtained; According to the task layers of various unmanned equipment, the task graph of each intelligent agent is obtained.

9. A collaborative planning method for emergency rescue heterogeneous unmanned equipment according to any one of claims 1 to 4, characterized in that: According to the local value of the current agent, the global value of all agents is obtained, including: According to the local value of the current agent, back propagation is used to update each agent in the direction of improving the global reward to obtain the global value of all agents.

10. A collaborative planning device for heterogeneous unmanned emergency rescue equipment, characterized in that: include: The first module is used to obtain emergency rescue scenarios, obtain multiple types of unmanned equipment, each type of unmanned equipment includes multiple intelligent agents, and generate an overall collaboration mechanism for all unmanned equipment; The second module is used to generate node representations of each agent according to the responsibilities of different agents in emergency rescue scenarios; The third module is used to generate the local cooperation mechanism of various unmanned equipment according to the overall cooperation mechanism of all unmanned equipment; obtain the homogeneous distribution map and heterogeneous cooperation map of each intelligent agent according to the local cooperation mechanism of various unmanned equipment; obtain the task map of each intelligent agent according to the responsibilities of different unmanned equipment in emergency rescue scenarios; The fourth module is used to input the node representation of the agent into the graph neural network to obtain the node graph embedding vector; The fifth module is used to obtain the local value of the current intelligent agent based on the node graph embedding vector, isomorphic distribution graph, heterogeneous collaboration graph and task graph; obtain the global value of all intelligent agents based on the local value of the current intelligent agent; and obtain the decision vector of the unmanned equipment based on the global value of all intelligent agents, which is used as a collaborative planning solution.

Images