A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations

Abstract

This invention discloses a self-organizing and cooperative method for heterogeneous multi-agent systems in lunar surface operations, relating to the field of intelligent control for spacecraft and lunar operations. The method includes: constructing cooperative units and their corresponding member attribute groups for specific tasks based on the current lunar operation mission and the capability information of each heterogeneous agent; obtaining the cooperative topology, sub-task allocation results, and global waypoint sequences for each node by constructing a cooperative graph structure and generating a joint cooperative matrix; dynamically generating local reference states and performing cooperative control by evaluating and adjusting consistency convergence errors to obtain updated node states and cooperative deviation values; and performing anomaly detection to trigger and execute local corrections or cooperative reorganization, thereby achieving continuous execution of the cooperative task. This method enables self-organizing and cooperative construction, precise consistency control, and adaptive anomaly reorganization of heterogeneous multi-agent systems on the lunar surface, improving the quality of cooperative execution and mission continuity.

Description

Technical Field

[0001] This invention relates to the field of intelligent control for spacecraft and lunar operations, and more specifically to a heterogeneous multi-agent self-organizing and collaborative method for lunar operations. Background Technology

[0002] As lunar exploration missions evolve from single-point roving and single-point sampling to long-term stays, multi-task concurrency, and multi-equipment collaborative operations, heterogeneous intelligent agents such as lunar transport vehicles, robotic dogs, and rovers are gradually becoming important execution units in the lunar operation system. Lunar operation scenarios cover tasks such as material transportation, terrain mapping and environmental perception, equipment deployment and assembly, emergency rescue and fault handling, etc. These tasks are usually characterized by strong disturbances in unstructured environments, multi-task concurrency and multi-resource coupling, significant differences in the capabilities of heterogeneous intelligent agents, limited computing power on the edge side, and intermittent communication links.

[0003] Among existing multi-agent collaboration methods, one type of approach mainly adopts rule-based or script-based collaboration, relying on manually preset behavioral rules, task flows, and exception handling logic. While this type of approach has a certain degree of controllability in known environments and limited scenarios, it is insufficient to adapt to the large number of unknown disturbances and complex emergencies in long-term lunar missions. It is difficult to cover the collaboration needs in variable environments, especially when task requirements and agent capabilities need to be dynamically matched. Rule-based methods cannot automatically determine whether collaboration is needed, how to construct collaboration units, and how to select members based on real-time status. This results in a lack of optimization in the composition of collaboration units, which can easily lead to redundancy of capabilities or missing key capabilities.

[0004] Another approach employs a strongly centralized scheduling method, where a central control node handles task decomposition, resource allocation, and path planning. While this method can achieve superior global performance under conditions of stable communication and sufficient computing power, it is highly dependent on link continuity and the reliability of the central node. In environments where lunar communication is intermittent and end-side autonomy is required, centralized scheduling is prone to failure due to single-point failures or link interruptions. In particular, when it is necessary to dynamically generate collaborative topologies and role assignments based on the complementarity of agent capabilities, communication link quality, and spatial distance, the centralized method struggles to autonomously generate collaborative relationships under weak or no-centralized conditions, and it also cannot maintain the consistency of collaborative states during task execution.

[0005] In addition, there is another type of approach that focuses on local optimization of a single agent, such as single-platform path planning, single-robot adaptive control, or local obstacle avoidance. These methods usually lack a holistic consideration of the collaborative relationships, role evolution, and overall system benefits among multiple agents, making it difficult to achieve a synergistic gain of 1+1>2 at the system level. Especially under conditions of multi-task concurrency and resource coupling, single-agent optimization methods are prone to problems such as repetitive work, resource waste, and inconsistency between local optima and global objectives. At the same time, existing methods generally lack the ability to detect and adaptively correct and reorganize collaborative anomalies online. When environmental changes, node failures, or link deterioration occur, it is difficult to quickly restore collaborative execution without relying on centralized control.

[0006] Therefore, how to design a heterogeneous multi-agent self-organizing and collaborative method for lunar surface operations, give full play to the complementary advantages of heterogeneous multi-agent capabilities, and improve the collaborative efficiency and robustness of lunar surface operations is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a heterogeneous multi-agent self-organizing and cooperative method for lunar surface operation scenarios. It aims to address the problems of insufficient cooperative stability and low cooperative efficiency in the absence of a central or weak central mode in lunar heterogeneous multi-agent systems under conditions of communication interruption and strong environmental disturbances. The method aims to achieve stable cooperation, rapid adaptation, efficient execution and continuous evolution under conditions of multi-task concurrency, give full play to the complementary advantages of different agents, and reduce repetitive operations and resource waste.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations includes the following steps: S1. Based on the current lunar surface operation task and the capability information of each heterogeneous intelligent agent, identify the collaborative requirements and construct collaborative units for specific tasks and their corresponding member attribute groups. S2. Based on the cooperative unit and its corresponding member attribute group, a cooperative graph structure is constructed and a joint cooperative matrix is ​​generated to generate self-organizing cooperative relationships, thereby obtaining the cooperative topology, sub-task allocation results and global waypoint sequence of each node. S3. During the collaborative execution process, based on the collaborative topology, subtask allocation results and global waypoint sequence of each node, combined with the real-time status of each node, a local reference state is dynamically generated, and collaborative control is performed by evaluating and adjusting the consistency convergence error to obtain the updated node status and collaborative deviation value. S4. Based on the updated node status and coordination deviation value, perform anomaly detection. When it is determined that there is a risk of coordination mismatch, trigger and execute local correction or coordination reorganization to achieve continuous execution of the coordination task.

[0010] Preferably, S1 includes: The individual agent fit is calculated based on the task requirement vector and the agent capability vector, and the collaborative requirement judgment function is used to determine whether the current task requires multi-agent collaborative execution. After determining that collaboration is needed, agents that meet the fitness threshold, remaining energy, communication quality and spatial reachability constraints are selected to form a candidate collaboration member set; By minimizing the comprehensive cost function, members are optimally selected from the set of candidate collaborative members, collaborative units are constructed, and the member attribute group corresponding to the collaborative unit is output.

[0011] Preferably, the collaborative requirement determination function Represented as:

[0012] in, , , , , These are weighting coefficients. , , , The trigger threshold for the corresponding factor. For indicator functions, For all candidate agents on the task Maximum single-unit adaptability, , , , These are respectively: load requirements, space range, risk level, and communication assurance. The maximum carrying capacity of all intelligent agents; when When it is determined that collaborative execution is required, This represents the total threshold for collaborative needs.

[0013] Preferably, the comprehensive cost function Represented as:

[0014] in, For the purpose of the task Constructed collaborative units For the cost of maneuver, For the cost of energy, For the cost of communication, For the size of the members of the collaborative unit, For the benefit of overall capabilities, , , , , These are the corresponding weighting coefficients; The collaborative unit Capability coverage constraints must be met:

[0015] in, Let u be the capability value of the agent in the u-th dimension. For the task The demand value in the u-th dimension.

[0016] Preferably, S2 includes: The collaborative unit is abstracted as a graph structure, where the agent is the graph node; the collaborative edge weight is calculated based on the complementarity of capabilities between nodes, communication link quality, spatial distance, collaboration credibility and task relevance, and an adjacency matrix is ​​constructed to generate the collaborative topology; The main coordinating node is determined based on the comprehensive role score of the nodes, and the roles of the remaining nodes are organized accordingly. The current task is decomposed into a set of subtasks, a subtask dependency matrix and an information interaction matrix are constructed, a subtask allocation matrix is ​​generated based on the matching degree between nodes and subtasks, and a joint coordination matrix is ​​constructed. Based on the subtask allocation matrix and subtask dependency matrix, a task sequence for each node is generated, and a global waypoint sequence for each node is generated by combining role correction terms and cooperation correction terms.

[0017] Preferably, the matching degree between the node and the subtask Represented as:

[0018] in, For subtasks The matching weight on the u-th capability dimension, For nodes The ability value in the u-th ability dimension. For subtasks The required value in the corresponding dimension, It is a small positive quantity; The global waypoint sequence includes the guidance points. Represented as:

[0019] in, The basic target location for the subtask. This refers to the positional correction amount caused by the node role. This represents the positional correction caused by the cooperative topology.

[0020] Preferably, S3 includes: Based on the node's current task execution status, the current active guidance point is selected from the global guidance waypoint sequence. Combined with the node's current position, neighborhood cooperation relationship, and local environmental constraints, a local reference position, reference progress status, reference energy status, and reference link status are constructed to form a local reference status. Define the consistency convergence error of the node, and generate precise cooperative control input based on the consistency convergence error to drive the node state update, thereby obtaining the updated basic state vector and cooperative execution state. Calculate the overall consistency cost function and the collaborative deviation function, and output the collaborative deviation value.

[0021] Preferably, the consistency convergence error Represented as:

[0022] in, and This is the weight matrix. For nodes The collaborative execution status, For nodes Local reference state, For the set of neighboring nodes, For adjacency matrix elements, and They are nodes and neighboring nodes Current location For nodes with neighboring nodes The ideal relative positional relationship; The precise collaborative control input Represented as:

[0023]

[0024] in, For nodes Consistency control items To ensure consistent control gain, The gain is tracked for the reference state.

[0025] Preferably, S4 includes: Based on the updated node status and coordination deviation value, a coordination anomaly detection index is calculated. When the index exceeds the anomaly detection threshold, it is determined that the current coordination unit has a risk of coordination mismatch. Based on the integrity criterion of the collaborative structure, distinguish between locally correctable anomalies and overall anomalies that require reorganization; For locally correctable anomalies, correction operations, including control parameter adjustment, adjacency relationship fine-tuning, local subtask reallocation, or local update of waypoints, are performed by solving the problem of minimizing the local correction cost function. For cases where overall reorganization is required, based on the current set of available agents and the set of remaining tasks, new collaborative units are determined by maximizing the reorganization objective function, and steps S1 to S3 are re-executed.

[0026] Preferably, the local correction cost function Represented as:

[0027] in, To correct the remaining coordination bias of the system after local correction, At the cost of partial adjustments, The time required for recovery For the affected nodes and task scope, , , , These are the corresponding weighting coefficients; The recombination objective function Represented as:

[0028] in, For the newly defined collaborative unit. Let be the set of available agents at time t. For the remaining task set, Candidate collaborative unit The ability to cover the remaining task set. The connectivity of candidate coordinating units. The energy margin of the candidate coordinating unit. To ensure the link guarantee capability of candidate collaborative units, The switching costs brought about by the restructuring , , , , These are the corresponding weighting coefficients.

[0029] As can be seen from the above technical solution, compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1. This method, through collaborative requirement identification and collaborative unit construction, can automatically determine whether multi-agent collaboration is needed based on the requirement vector of lunar operation tasks and the capability vector of heterogeneous agents, and optimize the construction of the minimum sufficient collaborative unit. Thus, under the condition of limited computing power on the edge, it avoids unnecessary collaborative overhead and resource waste, and improves the matching efficiency of task execution and resource utilization.

[0030] 2. By leveraging precise collaborative control in the generation of self-organizing collaborative relationships and the maintenance of consistent states, it is possible to autonomously generate collaborative topology, role division, and global waypoints in a lunar environment with intermittent communication links and time-varying topology. Based on local reference states, it dynamically adjusts the consistency convergence error of each agent, achieving stable convergence of multiple agents in terms of task objectives, behavioral timing, and spatial coordination, thereby improving the quality and reliability of collaborative execution.

[0031] 3. Through anomaly detection and adaptive correction and reorganization mechanism, it is possible to distinguish between locally correctable anomalies and overall reorganization anomalies without relying on continuous centralized control, prioritize local correction, and trigger collaborative reorganization when necessary. This enhances the robustness of the lunar heterogeneous multi-agent system under complex disturbances and local faults, and ensures the continuous execution capability of operational tasks. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0033] Figure 1 This is a flowchart of a heterogeneous multi-agent self-organizing and coordinating method for lunar surface operations, provided by an embodiment of the present invention. Figure 2 The lunar surface heterogeneous multi-agent collaborative operation system scenario and composition architecture diagram provided in the embodiments of the present invention; Figure 3 A schematic diagram of collaborative requirement identification and collaborative unit construction provided in this embodiment of the invention; Figure 4 A schematic diagram of heterogeneous multi-agent self-organizing cooperative relationship generation provided in this embodiment of the invention; Figure 5 A schematic diagram of consistency state maintenance and local reference state generation provided in this embodiment of the invention; Figure 6 The flowchart of local correction and reorganization recovery triggered by collaborative anomaly provided in this embodiment of the invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] like Figure 1 As shown, this embodiment provides a heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations, including the following steps: S1. Based on the current lunar surface operation task and the capability information of each heterogeneous intelligent agent, identify the collaborative requirements and construct collaborative units for specific tasks and their corresponding member attribute groups. S2. Based on the cooperative unit and its corresponding member attribute group, a cooperative graph structure is constructed and a joint cooperative matrix is ​​generated to generate self-organizing cooperative relationships, thereby obtaining the cooperative topology, sub-task allocation results and global waypoint sequence of each node. S3. During the collaborative execution process, based on the collaborative topology, subtask allocation results and global waypoint sequence of each node, combined with the real-time status of each node, a local reference state is dynamically generated, and collaborative control is performed by evaluating and adjusting the consistency convergence error to obtain the updated node status and collaborative deviation value. S4. Based on the updated node status and coordination deviation value, perform anomaly detection. When it is determined that there is a risk of coordination mismatch, trigger and execute local correction or coordination reorganization to achieve continuous execution of the coordination task.

[0036] By identifying collaborative needs and constructing collaborative units, it automatically determines the necessity of collaboration and optimizes member combinations under the condition of limited computing power on the edge, avoiding resource waste. Through precise collaborative control in the generation of self-organized collaborative relationships and maintenance of consistent state, it autonomously generates collaborative topology and guidance routes in the lunar environment with intermittent communication links and time-varying topology, and dynamically adjusts the consistency convergence error based on the local reference state, improving the collaborative quality and stability of multi-agents in terms of task objectives, behavioral timing and spatial coordination. Through anomaly detection and adaptive correction and reorganization mechanisms, it distinguishes between local correction and global reorganization, enhancing the system's robustness to complex disturbances and local failures without relying on continuous centralized control, and can effectively ensure the continuous execution of operational tasks.

[0037] The following provides further explanation of each step in the above method and its related technical features; In this embodiment S1, based on the current lunar surface operation task and the capability information of each heterogeneous intelligent agent, the collaborative requirements are identified, and collaborative units for specific tasks and their corresponding member attribute groups are constructed. In this embodiment, a system is constructed to address typical operational scenarios such as lunar surface exploration, sample collection, equipment deployment, material transportation, and emergency response. Figure 2 The diagram illustrates the scenario and architecture of a heterogeneous multi-agent collaborative operation system on the lunar surface. This system is divided into a task object layer, a support facility layer, an agent layer, and a scenario layer. The task object layer includes sample collection points, equipment deployment points, material transportation routes, and emergency response points, clearly defining the specific objectives and spatial locations required for collaborative operations. The support facility layer comprises communication relay facilities, energy supply facilities, and fixed work stations, providing necessary link guarantees, energy replenishment, and task scheduling support for the collaborative execution of agents. The agent layer consists of heterogeneous agents such as rovers, lunar transport vehicles, quadrupedal robotic dogs, and work units with robotic arms, each undertaking complementary functions such as rapid reconnaissance, heavy-load transportation, complex terrain exploration, and precise operation. The scenario layer is divided into multiple operational areas. Agents within each area achieve task collaboration through self-organizing relationships. Due to the characteristics of the lunar environment, such as large terrain undulations, unstable communication links, difficult energy replenishment, frequent dynamic changes in tasks, and significant differences in individual capabilities, a single agent often struggles to independently complete complex tasks. Therefore, autonomous organization and dynamic collaboration among multiple agents are necessary to achieve efficient execution, stable cooperation, and anomaly recovery during operations.

[0038] To uniformly describe the task object, the executing entity, and the environmental support conditions, we first define the heterogeneous multi-agent set as:

[0039] The task set is as follows:

[0040] in, This represents the overall set of entities executing on the lunar surface. This represents the overall set of tasks to be completed. For any intelligent agent Define its time at time The state vector is:

[0041] Among them, location and speed Describe the agent's maneuverability; remaining energy Reflecting its ability to continuously perform tasks; communication quality Characterizes the effectiveness of node access to the collaborative network; task execution status. This reflects the current stage of the task undertaken by the intelligent agent.

[0042] like Figure 3As shown, the implementation process of S1 is as follows: First, for the current lunar operation task, a task requirement vector is constructed, and the capability vectors and real-time status information of each candidate agent are obtained. The adaptation calculation module calculates the individual agent adaptation degree based on the task requirement vector and the agent capability vector, and determines whether the current task requires multi-agent collaborative execution based on the collaborative requirement determination function. After determining that collaboration is required, the collaborative requirement determination module selects agents that meet the adaptation degree threshold, remaining energy, communication quality and spatial reachability constraints to form a candidate collaborative member set. After determining that collaboration is required, the collaborative unit optimization module optimizes and selects members from the candidate collaborative member set by minimizing the comprehensive cost function, constructs a collaborative unit, and outputs the member attribute group corresponding to the collaborative unit.

[0043] S1.1 Task requirements and agent capability representation; For the task First, construct its task requirement vector:

[0044] This vector is used to derive from capacity requirements. Load requirements Time constraints Spatial range , communication guarantee and risk level The task is described in a unified manner from six aspects; Corresponding to the task requirement vector, for each agent Construct its capability vector:

[0045] These vectors represent the agent's maneuverability. Operational skills Carrying capacity Perception ability , communication ability and energy support capabilities In this way, heterogeneous entities with different structures and functions, such as lunar transport rovers, robot dogs, rovers, and robotic arms, can be mapped to the same comparable dimensional space for measurement. In obtaining the task requirement vector and capability vector Next, we introduce single-unit adaptability:

[0046] Used for quantifying intelligent agents For the task The degree of adaptation, in the formula, Represents intelligent agents In the Capability values ​​in each dimension Indicates task The required value in the corresponding dimension, Indicates the first The importance weight of each capability dimension To avoid a denominator of zero, a small positive quantity is introduced, satisfying... ,use The form; Individual compatibility On the one hand, it is used to determine whether a single agent can complete the task independently; on the other hand, it is used to construct a collaborative candidate set, determine core execution members, and screen alternative nodes when abnormal personnel are added. If the individual node's adaptability is higher than the preset threshold for independent execution, the task can be determined to have the potential to be completed independently. When the individual node adaptability of all candidate nodes is lower than the threshold, the collaborative requirement determination process is initiated. The threshold for independent execution should be determined based on the task level, task risk, and fault tolerance requirements.

[0047] S1.2 Collaborative needs assessment and candidate member selection; To further determine the task Does it need to initiate a collaboration mechanism? Build a collaboration requirement determination function:

[0048] In the formula, to This represents the weighting coefficient for different decision factors. to This indicates the trigger threshold for the corresponding factor. This represents the total threshold for collaborative demand, when At that time, determine the task Collaborative execution is required; after determining that collaboration is needed, a set of candidate collaborative members is constructed:

[0049] in, This represents the minimum fitness threshold for entering the candidate set. Indicates the lower limit of safe energy. Indicates the minimum communication quality requirements. This represents the maximum allowable approach radius. By using the above constraints, nodes with high execution risks can be eliminated before constructing collaborative units, thereby improving the feasibility of subsequent collaborative structures.

[0050] S1.3 Optimized construction of collaborative units and output of member attribute groups; From the candidate set Further selection of members to form collaborative units And establish the following optimization objectives:

[0051] In the formula, to These are the corresponding weight coefficients; To make the optimization objective computable, the maneuver cost, energy cost, communication cost, and overall capability gain are defined as follows:

[0052] in, Represents a node Maximum permissible speed of movement; , and Representing nodes respectively To complete the task The estimated energy consumption for mobility, operations, and communication support; Represents a node With nodes Between moments Normalized link quality; Indicates the first Weights for each capability dimension; Indicates task In the Requirement values ​​in each capability dimension; Accordingly, the capability coverage constraint is explicitly written as:

[0053] definition ,in , Let be the state vector of the aforementioned agent. For the aforementioned capability vector, For intelligent agents For the task The individual unit adaptability is determined, and this step ultimately outputs the cooperative unit. and the member attribute group corresponding to this collaborative unit. The output results serve as input for subsequent steps such as collaborative topology generation, role organization, task allocation, and phase guidance generation.

[0054] In this embodiment S2, based on the cooperative unit and its corresponding member attribute group, a cooperative graph structure is constructed and a joint cooperative matrix is ​​generated to generate self-organizing cooperative relationships, thereby obtaining the cooperative topology, sub-task allocation results and the global waypoint sequence of each node; like Figure 4As shown, the S2 implementation process is as follows: The collaborative unit is abstracted as a graph structure, where the agent is the graph node; the collaborative edge weight is calculated based on the complementarity of capabilities between nodes, communication link quality, spatial distance, collaboration credibility, and task relevance, and an adjacency matrix is ​​constructed to generate a collaborative topology; then, the role layer is entered, the main coordinating node is determined based on the comprehensive role score of the nodes, and the roles of the remaining nodes are organized; on this basis, the current task is decomposed into a set of subtasks, a subtask dependency matrix and an information interaction matrix are constructed, a subtask allocation matrix is ​​generated based on the matching degree between nodes and subtasks, and a joint collaboration matrix is ​​constructed to uniformly represent the coupling relationship between the task chain, information chain, and topology chain; finally, the task sequence of each node is generated based on the subtask allocation matrix and the subtask dependency matrix, and a global waypoint sequence of each node is generated by combining the role correction term and the collaboration correction term, providing global constraints for subsequent precise collaborative control.

[0055] S2.1 Collaborative Graph Structure Modeling and Topology Construction; First, the collaborative unit Abstracted into a graph structure:

[0056] Among them, the node set edge set Representing collaborative connection relationships, the set of edge weights This indicates the connection strength; For any two intelligent agents and Define its task-oriented The collaborative edge weights are:

[0057] In the formula, to These are the corresponding weight coefficients; Indicates the time of the two nodes. Spatial distance; Used to measure the degree to which the combined capabilities of two nodes meet the task requirements. Used to characterize the real-time link quality between two nodes. Used to characterize the credibility of historical collaborations Used to characterize the degree of potential collaborative coupling between two nodes under the current task decomposition; The capability complementarity term in the above edge weights can be further expressed as:

[0058] In the formula, Indicates the first Complementary evaluation weights for each capability dimension, and Representing nodes respectively With nodes In the Ability values ​​in each ability dimension Indicates task The required value in the corresponding dimension; After obtaining the edge weights, an adjacency matrix can be constructed based on the threshold:

[0059] when Greater than or equal to the threshold If a valid collaborative connection exists between nodes, it is assumed that a direct connection is established; otherwise, no direct connection is established. If necessary, the collaborative topology can be constrained through sparse optimization to suppress redundant connections while maintaining high-value connections, thereby obtaining a collaborative topology that has both key connection relationships and low communication complexity.

[0060] S2.2 Node Role Organization and Determination of the Main Coordinating Node: After topology generation, it is also necessary to organize roles within the collaborative units and define the overall role rating of nodes:

[0061] And order:

[0062] As the primary coordinating node to These are the corresponding energy weighting coefficients; To make the evaluation metrics in the node comprehensive role rating calculable, the following definitions are added:

[0063] in, Indicates the overall capability level of the node. This represents the average connectivity strength of nodes in the current collaborative topology. Indicates energy margin, Indicates the degree of operational stability. This represents the fluctuation amplitude of key node state variables within a set time window. It represents the average matching degree of a node to the current set of subtasks, and the basis for generating the main coordinating node has been transformed from abstract evaluation into a directly calculable scoring system.

[0064] S2.3 Subtask Decomposition and Collaborative Relationship Mapping: To ensure that task relationships have clearly defined modeling objects, the tasks are further... Decomposed into a set of subtasks:

[0065] in, Indicates task The number of subtasks obtained from the decomposition is used to further define the task dependency matrix of the subtask layer:

[0066] When subtask Must precede subtasks During execution, ;otherwise, The task dependency matrix is ​​used to describe the sequence of operations between subtasks. It applies to subtasks belonging to the same task and is not directly used to represent the connection relationship between nodes. In the subtask set Given the given information, the matching degree between nodes and subtasks is further calculated. For each node... sub-tasks Its matching degree is defined as:

[0067] In the formula, Subtasks In the Matching weights across each capability dimension Subtasks The required value in the corresponding dimension; To clarify the relationship between nodes and subtasks, define subtask assignment variables:

[0068] This results in a node-subtask allocation matrix:

[0069] Based on the aforementioned subtask matching degree By combining node role categories, ability constraints, and task load constraints, we can obtain the node... The set of subtasks:

[0070] The mapping between nodes and tasks is represented by the allocation matrix. Decisions, rather than those based on the task dependency matrix. Launch directly; To uniformly represent the coupling relationship between the task chain, information chain, and topology chain, an information interaction matrix is ​​further defined. And based on the allocation matrix Projecting the dependencies of the subtask layer onto the node layer yields the node task coupling matrix:

[0071] in, The elements are used to characterize the strength of cooperative coupling formed between nodes due to the relationship between preceding and subsequent subtasks, and a joint cooperative matrix is ​​further constructed:

[0072] In the formula, , , These represent the fusion weights of task-dependent projection, information interaction, and topological connectivity, respectively. Based on the above definitions, the joint coordination matrix... It can uniformly describe the overall collaborative coupling strength between nodes.

[0073] S2.4 Node Task Sequence and Global Boot Information Generation; Considering that a single node may undertake multiple subtasks, and that these subtasks typically have a sequential execution relationship, further analysis is conducted based on the task dependency matrix. and allocation matrix The task execution order for generating nodes is set as follows: Represents a node The task sequence, then The set of subtasks undertaken by a node Under the premise of satisfying task dependency constraints, the subtasks are sorted according to priority and matching degree to obtain:

[0074] in, Represents a node The number of task phases undertaken; For any stage of the node task sequence Let its basic target operation location be... The character modification item is The co-correction term is The guiding target point corresponding to the node in this stage is defined as:

[0075] in, Indicates the location of the basic target corresponding to the subtask in the current stage; This represents the positional adjustment amount resulting from the node's role and responsibilities; This represents the cooperative positional correction amount resulting from cooperative topology and neighborhood relationships; Character modification items can be defined according to the node character type:

[0076] The cooperative correction term can be defined based on the node adjacency relationship and the ideal relative cooperative relationship:

[0077] in, Represents a node The set of neighboring nodes in a collaborative topology Representing neighboring nodes For nodes The influence weight of the guiding point in the current stage Represents a node With nodes The ideal relative positional relationship that should be satisfied in collaborative execution; Furthermore, joint synergy matrices can be utilized. The importance of each guiding point in each stage is weighted, and nodes are set. In the stage The guiding point weight is Then the node The global waypoint sequence is represented as:

[0078] The global waypoint sequence reflects the key stages that a node should sequentially traverse in the current task scenario. This step ultimately outputs the cooperative topology. Adjacency matrix Main Coordinating Node Subtask allocation matrix Joint Synergy Matrix and the global waypoint sequence of each node. The aforementioned topological and matrix results are used to describe the collaborative organizational structure and node task sequences. with waypoint sequence Used to describe the stage advancement path of nodes at the global task level, together they form the input basis for dynamic consistency convergence and local reference state generation in step S3.

[0079] In this embodiment S3, during the collaborative execution process, based on the collaborative topology, subtask allocation results, and global waypoint sequence of each node, combined with the real-time status of each node, a local reference state is dynamically generated, and collaborative control is performed by evaluating and adjusting the consistency convergence error to obtain the updated node state and collaborative deviation value. like Figure 5As shown, the S3 implementation process is as follows: Using the sequence of each agent's role and the global waypoint sequence as input, in the stage selection layer, the current activity guidance point is selected from the waypoint sequence based on the current task execution state of each agent. In the execution scenario layer and the local reference state layer, obstacles such as rocks, craters, and danger zones in the lunar environment, as well as the adjacency and information interaction relationships between the transport vehicle, rover, and robotic arm, are considered. A local reference state containing reference position, reference task progress, reference energy state, and reference link state is dynamically generated for each agent. Subsequently, in the deviation layer, the tracking deviation of each agent to the local reference state and the relative cooperative deviation with neighboring agents are calculated to form a consistency convergence error. Based on this error, a precise cooperative control input is generated to drive node state updates, obtaining the updated basic state vector and cooperative execution state. The overall consistency cost function and cooperative deviation function are calculated, and the cooperative deviation value is output, thereby achieving consistent convergence and precise maintenance of the cooperative state of multiple agents in the dynamic lunar environment.

[0080] S3.1 Cooperative execution state mapping and local reference state generation; For nodes The aforementioned basic state vector The system already includes information such as location status, remaining energy, communication quality, and task execution status. The cooperative execution state of a node is constructed by selecting and mapping the state components directly related to cooperative execution from the basic state vector.

[0081] in, This indicates the current spatial position or pose state of the node. Indicates the current remaining energy state of the node; Task execution state from the basic state vector The mapping is used to characterize the current task progress of the node; Communication quality from the basic state vector The mapped value is used to characterize the current communication link state of the node. node The global waypoint sequence is as follows:

[0082] And the node task sequence is:

[0083] Set nodes The total number of phase tasks undertaken within the current collaboration cycle is And its discrete task execution state This indicates that the node is currently at the th position. Each task phase, for any point... Define the node at time The task progress status is as follows:

[0084]

[0085] The task progress state of a node is then defined as:

[0086] Based on the above definition, the execution state of discrete tasks can be... The unified mapping is to the advancement amount on the interval [0,1], so that the advancement degree of different nodes under different stages of tasks is comparable; If communication quality Since it is already a normalized communication quality index, we can directly set:

[0087] If communication quality The overall communication quality is formed by combining multiple link metrics. First, a comprehensive communication quality assessment can be constructed based on received signal strength, communication delay, and packet loss rate. Then, this assessment can be normalized and used as the link state. It provides a unified representation of the current communication validity of nodes; To clarify the stage objectives that a node should track at the current moment, define the node at time [time]. The current activity navigation point is:

[0088] in, This indicates the phased global target position that the node should prioritize approaching at the current moment. The current activity guidance point is directly determined by the node's current discrete task execution state. In the global waypoint sequence The selected state is the task execution status when a node completes its current stage task and meets the stage switching conditions. Update to the next stage, so that the current activity guide point will be synchronously switched to the guide point corresponding to the next stage; Get the current activity guide point Then, by combining the node's current location, neighborhood collaboration relationships, and local environmental constraints, a local reference location for the node is generated. :

[0089] in, This represents the advancement coefficient of a node toward the current active guide point. Represents the neighborhood cooperative correction coefficient. This represents the local obstacle avoidance correction coefficient. Represents a node With nodes The ideal relative positional relationship that should be satisfied in collaborative execution This represents the obstacle avoidance correction amount caused by a node being affected by local obstacles, dangerous areas, or impassable areas; Among them, the propulsion coefficient :

[0090] In the formula, Represents a node Maximum permissible propulsion speed Indicates the discrete update step size; In terms of task progress, the local reference progress status is determined by the current activity guide point. The corresponding stage position is determined, due to the current activity guidance point. With global waypoint sequence If a node corresponds to a certain stage guide point, then the node's reference progress status is:

[0091] Reference progress status is used to characterize the target stage progress level that a node should achieve at the current moment, thus aligning it with the actual task progress status. These constitute directly comparable error quantities; In terms of resources, the reference energy state of the generating node is as follows:

[0092] in, Indicates the node's role The required lower limit of safe energy. This represents the estimated remaining energy required for a node to complete all subsequent stages from the current stage. It can be obtained by summing the energy consumption of path maneuvering and operation from the current stage to the final stage; In terms of communication, the reference link state of the generating node:

[0093] in, This indicates the minimum level of link assurance required by the node under its current role. Represents a node In the current stage, to maintain contact with neighboring nodes The link requirements that need to be met for the collaborative relationship, and the reference link status is used to reflect the minimum communication guarantee requirements that nodes should meet when maintaining neighborhood collaboration and phase advancement; The local reference state of the constructible node at the current moment is:

[0094] In this way, a clear data transfer relationship is established between S2 and S3: S2 provides phase guidance information and collaborative organization information, while S3 provides the reference state that can be executed at the current moment.

[0095] S3.2 Consistency Deviation Assessment and Precise Collaborative Control; To measure the deviation of a node's current state from the target reference state and its neighborhood cooperative relationships, the node's consistency convergence error is defined as:

[0096] The first term represents the deviation of the node's current state from its local reference state, and the second term represents the deviation of the relative cooperative relationship between the node and its neighboring nodes. Each node is required to satisfy the preset neighborhood cooperative relationship under the constraints of its own reference state and jointly enter the target cooperative stable domain. To measure the overall coordination level of the collaborative units, a general consistency cost function is further constructed:

[0097] like If the value continues to decrease and remains within a small range, it indicates that the coordinating unit has gradually entered a stable coordinating state; if the value continues to increase, it indicates that the coordinating process is showing a trend of losing synchronization, dispersion, or instability. Based on adjacency matrix Define the consistency control terms for the nodes as follows:

[0098] Based on the consistency control items, precise collaborative control inputs are formed:

[0099] Each node must coordinate with its collaborating neighbors and continuously converge toward the local reference state corresponding to the current activity phase. Since the role requirements have already been reflected in the role organization of S2, the generation of phase guiding points, and the reference energy state and reference link state of S3, there is no need to introduce an independent role reference state parallel to the local reference state. S3.3 State Update and Coordination Deviation Output; After receiving control input, the node state is updated in a discrete manner:

[0100] in, This represents the node's own dynamic evolution function. This represents the external environmental disturbance term; To further evaluate the current collaborative execution quality from multiple dimensions, a collaborative deviation function is defined:

[0101] in, , , and These respectively represent the degree of deviation of members in terms of spatial location, mission progress, energy distribution, and link status; The output of this step includes the updated control inputs for each node. Basic state Collaborative execution status Local reference state Consistency Cost and the co-operational deviation value It is used to drive the continued execution of the next time step and serves as input for S4 anomaly detection and reorganization recovery.

[0102] In this embodiment S4, anomaly detection is performed based on the updated node status and coordination deviation value. When it is determined that there is a risk of coordination mismatch, a local correction or coordination reorganization is triggered and executed to achieve continuous execution of the coordination task. like Figure 6 As shown, the implementation process of S4 is as follows: In the execution scenario layer, abnormal signals such as link deterioration, insufficient energy, node failure, increased coordination deviation, and role failure are continuously monitored; the detection layer calculates the coordination anomaly detection index based on the control input, basic state, coordination execution state, and coordination deviation value output by S3; the first judgment layer determines whether an anomaly is triggered based on whether the index exceeds a preset threshold. If no anomaly is triggered, normal coordination execution is returned; if an anomaly is triggered, the structure discrimination layer is entered, and the coordination structure integrity criterion is used to distinguish between locally correctable anomalies and overall reorganization anomalies; for locally correctable anomalies, the problem of minimizing the local correction cost function is solved, and correction operations including control parameter adjustment, adjacency relationship fine-tuning, local subtask reallocation, or local update of waypoints are performed; for overall reorganization anomalies, based on the current set of available agents and the set of remaining tasks, a new coordination unit is re-determined by maximizing the reorganization objective function, and steps S1 to S3 are re-executed to achieve continuous execution of the coordination task.

[0103] S4.1 Collaborative anomaly detection and anomaly type identification; To uniformly characterize the degree of abnormal risk in the current collaborative unit, a collaborative anomaly detection index is defined:

[0104] in, to These represent the corresponding weight coefficients; Used to characterize the overall degree of deviation on key dimensions; Used to characterize the overall coordination level of the coordinating units at the current moment; Used to characterize the degree of abrupt change in node control input at consecutive time points; and These are used to characterize the degree of inadequacy of the node's current energy state and link state relative to the local reference state, respectively; when When this occurs, the current collaborative unit is considered to have entered an abnormal triggering state; when This indicates that the collaborative execution is still within an acceptable range, and the system continues to maintain the consistency and precise collaborative process of S3. Furthermore, to differentiate the scope of the impact of anomalies, a collaborative structure integrity criterion is defined: if the current adjacency matrix... The corresponding key connectivity relationships remain valid, and the joint coordination matrix remains valid. The primary coupling relationship represented by the node was not fundamentally disrupted, and the node task sequence... With global waypoint sequence If execution can still be maintained through minor adjustments, it is determined to be a locally correctable anomaly; otherwise, it is determined to be an anomaly requiring overall reorganization.

[0105] S4.2 Implementation of local correction strategy; For anomalies that affect a local area but have not yet disrupted the overall collaborative structure, a local correction approach is prioritized, and a local correction cost function is constructed as follows:

[0106] in, This indicates the remaining cooperative deviation of the system after the partial correction is implemented. This indicates the cost of partial adjustment. Indicates the time required for recovery. Indicates the affected nodes and task scope. to These are the corresponding weighting coefficients; Local corrections are preferred based on the current subtask allocation matrix. Node task sequence Global waypoint sequence Adjacency matrix and joint synergy matrix This process is carried out to ensure that each node still has clear and feasible phase objectives in subsequent execution.

[0107] S4.3 Evaluation of the effects of synergistic recombination and recovery; After local correction, the local reference state of the affected nodes is regenerated. And recalculate the consistency cost. and co-operational deviation value If the revised Drop to threshold Below, and and If the deviation and consistency indicators continue to decline within the preset time window, the local correction is considered effective, and the consistency maintenance and precise collaborative execution will continue according to S3. If the deviation and consistency indicators still cannot be restored to the threshold range within a reasonable time after the local correction, the collaborative reorganization will be triggered. When anomalies have caused critical node failures, critical link interruptions, task dependency disruptions, or joint coordination matrices... When the main coupling relationship is unstable, or the overall coordination capability drops below a preset threshold, the system triggers a coordinated reorganization. This reorganization process is essentially a dynamic re-execution of S1 and S2, but its input conditions have changed: at this point, some nodes may have failed, some nodes may have decreased remaining energy, the task progress may have progressed to the middle, the current topology may be partially broken, and the original task allocation matrix may be affected. Node task sequence and global waypoint sequence It may no longer be applicable. Therefore, reorganization is not simply returning to the initial state, but rather reconstructing new collaborative units and collaborative relationships in combination with the current context. Set time The set of available intelligent agents is The remaining task set is The new collaborative units are redefined based on the following objectives:

[0108] in, This indicates the coverage capability of the candidate collaborative unit for the remaining task set. This indicates the connectivity of the candidate cooperating unit. Indicates the energy margin of candidate cooperating units. This indicates the link guarantee capability of the candidate coordinating unit. This indicates the switching costs resulting from the restructuring. to As corresponding to the weighting coefficients, this objective is used to reselect the set of nodes most suitable for undertaking subsequent collaborative tasks under the current remaining resources and tasks. In redefining new collaborative units Next, the corresponding member attribute groups are constructed:

[0109] Subsequently, S2 and S3 are executed again to form a complete closed loop. The system can quickly detect, correct hierarchically, and adaptively reorganize cooperative mismatches without relying on continuous centralized control. Through the closed-loop connection with S1, S2, and S3, it can continuously suppress the risks of local anomalies, structural mismatches, and task interruptions in the cooperative execution process of heterogeneous multi-agent systems on the lunar surface. This ensures that the system still has the ability to continuously cooperate and maintain task continuity under complex environments, dynamic disturbances, and local faults.

[0110] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0111] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations, characterized in that, Includes the following steps: S1. Based on the current lunar surface operation task and the capability information of each heterogeneous intelligent agent, identify the collaborative requirements and construct collaborative units for specific tasks and their corresponding member attribute groups. S2. Based on the cooperative unit and its corresponding member attribute group, a cooperative graph structure is constructed and a joint cooperative matrix is ​​generated to generate self-organizing cooperative relationships, thereby obtaining the cooperative topology, sub-task allocation results and global waypoint sequence of each node. S3. During the collaborative execution process, based on the collaborative topology, subtask allocation results and global waypoint sequence of each node, combined with the real-time status of each node, a local reference state is dynamically generated, and collaborative control is performed by evaluating and adjusting the consistency convergence error to obtain the updated node status and collaborative deviation value. S4. Based on the updated node status and coordination deviation value, perform anomaly detection. When it is determined that there is a risk of coordination mismatch, trigger and execute local correction or coordination reorganization to achieve continuous execution of the coordination task.

2. The heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 1, characterized in that, S1 includes: The individual agent fit is calculated based on the task requirement vector and the agent capability vector, and the collaborative requirement judgment function is used to determine whether the current task requires multi-agent collaborative execution. After determining that collaboration is needed, agents that meet the fitness threshold, remaining energy, communication quality and spatial reachability constraints are selected to form a candidate collaboration member set. By minimizing the comprehensive cost function, members are optimally selected from the set of candidate collaborative members, collaborative units are constructed, and the member attribute groups corresponding to the collaborative units are output.

3. The heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 2, characterized in that, The collaborative requirement determination function Represented as: in, , , , , These are weighting coefficients. , , , The trigger threshold for the corresponding factor. For indicator functions, For all candidate agents on the task Maximum single-unit adaptability, , , , These are respectively: load requirements, space range, risk level, and communication assurance. The maximum carrying capacity of all intelligent agents; when When it is determined that collaborative execution is required, This represents the total threshold for collaborative needs.

4. The heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 2, characterized in that, The comprehensive cost function Represented as: in, For the purpose of the task Constructed collaborative units For the cost of maneuver, For the cost of energy, For the cost of communication, For the size of the members of the collaborative unit, For the benefit of overall capabilities, , , , , These are the corresponding weighting coefficients; The collaborative unit Capability coverage constraints must be met: in, Let u be the capability value of the agent in the u-th dimension. For the task The demand value in the u-th dimension.

5. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 1, characterized in that, S2 includes: The collaborative unit is abstracted as a graph structure, where the agent is the graph node; the collaborative edge weight is calculated based on the complementarity of capabilities between nodes, communication link quality, spatial distance, collaboration credibility and task relevance, and an adjacency matrix is ​​constructed to generate the collaborative topology; The main coordinating node is determined based on the comprehensive role score of the nodes, and the roles of the remaining nodes are organized accordingly. The current task is decomposed into a set of subtasks, a subtask dependency matrix and an information interaction matrix are constructed, a subtask allocation matrix is ​​generated based on the matching degree between nodes and subtasks, and a joint coordination matrix is ​​constructed. Based on the subtask allocation matrix and subtask dependency matrix, a task sequence for each node is generated, and a global waypoint sequence for each node is generated by combining role correction terms and cooperation correction terms.

6. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 5, characterized in that, The matching degree between the node and the subtask Represented as: in, For subtasks The matching weight on the u-th capability dimension, For nodes The ability value in the u-th ability dimension. For subtasks The required value in the corresponding dimension, It is a small positive quantity; The global waypoint sequence includes the guidance points. Represented as: in, The basic target location for the subtask. This refers to the positional correction amount caused by the node role. This represents the positional correction caused by the cooperative topology.

7. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 1, characterized in that, S3 includes: Based on the node's current task execution status, the current active guidance point is selected from the global guidance waypoint sequence. Combined with the node's current position, neighborhood cooperation relationship, and local environmental constraints, a local reference position, reference progress status, reference energy status, and reference link status are constructed to form a local reference status. Define the consistency convergence error of the node, and generate precise cooperative control input based on the consistency convergence error to drive the node state update, thereby obtaining the updated basic state vector and cooperative execution state. Calculate the overall consistency cost function and the collaborative deviation function, and output the collaborative deviation value.

8. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations according to claim 7, characterized in that, The consistency convergence error Represented as: in, and This is the weight matrix. For nodes The collaborative execution status, For nodes Local reference state, For the set of neighboring nodes, For adjacency matrix elements, and They are nodes and neighboring nodes Current location For nodes with neighboring nodes The ideal relative positional relationship; The precise collaborative control input Represented as: in, For nodes Consistency control items To ensure consistent control gain, The gain is tracked for the reference state.

9. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations as described in claim 1, characterized in that, S4 includes: Based on the updated node status and coordination deviation value, a coordination anomaly detection index is calculated. When the index exceeds the anomaly detection threshold, it is determined that the current coordination unit has a risk of coordination mismatch. Based on the integrity criterion of the collaborative structure, distinguish between locally correctable anomalies and overall anomalies that require reorganization; For locally correctable anomalies, correction operations, including control parameter adjustment, adjacency relationship fine-tuning, local subtask reallocation, or local update of waypoints, are performed by solving the problem of minimizing the local correction cost function. For cases where overall reorganization is required, based on the current set of available agents and the set of remaining tasks, new collaborative units are determined by maximizing the reorganization objective function, and steps S1 to S3 are re-executed.

10. A heterogeneous multi-agent self-organizing and cooperative method for lunar surface operations according to claim 9, characterized in that, The local correction cost function Represented as: in, To correct the remaining coordination bias of the system after local correction, At the cost of partial adjustments, The time required for recovery For the affected nodes and task scope, , , , These are the corresponding weighting coefficients; The recombination objective function Represented as: in, For the newly defined collaborative unit. Let be the set of available agents at time t. For the remaining task set, Candidate collaborative unit The ability to cover the remaining task set. The connectivity of candidate coordinating units. The energy margin of the candidate coordinating unit. To ensure the link guarantee capability of candidate collaborative units, The switching costs brought about by the restructuring , , , , These are the corresponding weighting coefficients.

Images