Heterogeneous unmanned cluster control method and system combining dynamic topology and hierarchical game

Abstract

The application provides a heterogeneous unmanned cluster control method and system combining dynamic topology and hierarchical game, relates to the technical field of unmanned cluster control, and first acquires the node attribute category and node motion parameter of each unmanned node in the heterogeneous unmanned cluster, and constructs a topology connection relationship set; then constructs a hierarchical game control structure based on the topology connection relationship set, which contains an upper strategy game domain and a lower action execution domain; then distributes a strategy income function and a game interaction rule through the upper strategy game domain to generate a hierarchical game control instruction; finally, the lower action execution domain generates real-time motion control parameters according to the hierarchical game control instruction and sends them to the unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion. The application improves the adaptability and cooperative control ability of the cluster to complex environments and improves the task completion efficiency.

Description

Technical Field

[0001] This invention relates to the field of unmanned swarm control technology, and more specifically, to a heterogeneous unmanned swarm control method and system that combines dynamic topology and hierarchical game theory. Background Technology

[0002] In today's era of rapid technological advancement, heterogeneous unmanned swarms have demonstrated enormous application potential in numerous fields, such as military reconnaissance, environmental monitoring, and logistics delivery. Heterogeneous unmanned swarms typically consist of various unmanned nodes with different functional types, which work collaboratively within a detection area to accomplish complex tasks.

[0003] Currently, control methods for unmanned swarms have many limitations. On the one hand, traditional control methods often employ static topologies, which are ill-suited to dynamically changing environments and task requirements. In practical applications, the relative positions and communication status of unmanned nodes change continuously over time and as the task progresses. Static topologies cannot reflect these changes in a timely manner, leading to low swarm collaboration efficiency and even the inability to complete predetermined tasks. On the other hand, most existing control methods lack effective hierarchical game theory mechanisms. When dealing with global task decomposition and local motion control, it is difficult to achieve organic unity between the global and local aspects, and it fails to fully consider the interest dynamics and collaborative cooperation among unmanned nodes. This makes it difficult for the swarm to make optimal decisions and responses when facing complex tasks and unexpected situations. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory, the method comprising: Obtain the node attribute category and node motion parameters of each unmanned node in the heterogeneous unmanned cluster. The node attribute category is used to distinguish the functional type of the unmanned node. The node motion parameters include the node spatial position coordinates, the node velocity vector, and the node acceleration vector. A set of topological connections for a heterogeneous unmanned cluster is constructed based on the node attribute categories and the node motion parameters. The set of topological connections includes the existence identifier of communication links between each unmanned node and other unmanned nodes within the detection area, as well as the relative motion constraint boundary parameters. Based on the aforementioned set of topological connections, a hierarchical game control structure is constructed. The hierarchical game control structure includes an upper-level strategy game domain and a lower-level action execution domain. The upper-level strategy game domain is used for global task decomposition and collaborative path orchestration of heterogeneous unmanned clusters. The lower-level action execution domain is used for local motion control and obstacle avoidance response processing of each unmanned node. The strategy payoff function and game interaction rules are assigned to each unmanned node through the upper-level strategy game domain, and the strategy payoff function of each unmanned node is updated according to the changes in the set of topological connections, generating a hierarchical game control instruction containing the optimized strategy selection results of each unmanned node. The lower-level action execution domain receives the hierarchical game control command and generates real-time motion control parameters for each unmanned node based on the hierarchical game control command and the set of topological connections. The real-time motion control parameters are then sent to the corresponding unmanned nodes to drive the heterogeneous unmanned cluster to perform cooperative motion.

[0005] Furthermore, embodiments of the present invention also provide a heterogeneous unmanned swarm control system combining dynamic topology and hierarchical game theory, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a heterogeneous unmanned swarm control system combining dynamic topology and hierarchical game theory reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the heterogeneous unmanned swarm control system combining dynamic topology and hierarchical game theory to execute the above-described heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory.

[0007] Based on the above, by acquiring the node attribute categories and motion parameters of each unmanned node in a heterogeneous unmanned cluster, and constructing a dynamic set of topological connections accordingly, the communication links and relative motion constraints between unmanned nodes can be accurately reflected in real time. This allows the cluster topology to dynamically adjust with changes in environment and task, greatly improving the cluster's adaptability and flexibility to complex environments. A hierarchical game control structure is constructed based on this set of topological connections. Global task decomposition and collaborative path orchestration are placed in the upper-level strategy game domain, while local motion control and obstacle avoidance response processing are placed in the lower-level action execution domain, achieving organic coordination between the global and local aspects. The upper-level strategy game domain generates optimized strategy selection results by allocating strategy payoff functions and game interaction rules, and updating the strategy payoff functions according to topological changes. This fully considers the game-playing interests of each unmanned node, enabling the cluster to make optimal decisions at the global level. The lower-level action execution domain generates real-time motion control parameters based on the hierarchical game control instructions and the set of topological connections, driving unmanned nodes to perform collaborative movements. This ensures the accuracy and timeliness of local actions, thereby significantly improving the collaborative control capability, task completion efficiency, and ability to cope with complex situations in the heterogeneous unmanned cluster. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the heterogeneous unmanned swarm control method that combines dynamic topology and hierarchical game theory provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of a heterogeneous unmanned swarm control system that combines dynamic topology and hierarchical game theory, provided in an embodiment of the present invention. Detailed Implementation

[0010] Figure 1 This is a flowchart illustrating a heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory, provided in one embodiment of the present invention. A detailed description follows.

[0011] This embodiment provides a heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory. It is applied to a heterogeneous unmanned swarm system composed of various types of unmanned nodes, and is suitable for unmanned swarm systems in special industrial scenarios such as oil and gas field development, deep-sea operations, unmanned vehicles, unmanned vessels, and drones. For example, it can be deployed in a complex target area including mountains, forests, and building ruins to perform collaborative search and rescue missions. Specifically, the heterogeneous unmanned swarm includes aerial unmanned aerial vehicle nodes of a first attribute category and ground unmanned vehicle nodes of a second attribute category. The aerial unmanned aerial vehicle nodes possess high-altitude wide-area perspective, beyond-line-of-sight communication relay capabilities, and rapid cross-terrain maneuverability; the ground unmanned vehicle nodes possess fine terrain traversal capabilities, close-range vital sign detection capabilities, and rescue material transport capabilities.

[0012] Step S110: Obtain the node attribute category and node motion parameters of each unmanned node in the heterogeneous unmanned cluster. The node attribute category is used to distinguish the functional type of the unmanned node. The node motion parameters include the node spatial position coordinates, the node velocity vector, and the node acceleration vector.

[0013] During the initialization phase and at the start of each subsequent control cycle, each unmanned node in the heterogeneous unmanned swarm broadcasts its own status information via its onboard communication module. The broadcast status information includes the node's attribute category, encoded using predefined enumeration values: value A1 represents an aerial unmanned aerial vehicle (UAV) node, and value A2 represents a ground unmanned vehicle (UAV) node. The broadcast status information also includes node motion parameters. The node's spatial coordinates are calculated by fusing data from a Global Navigation Satellite System (GNSS) receiver and an inertial measurement unit (INS), forming a three-dimensional coordinate vector. For aerial UAV nodes, the three dimensions of this coordinate vector represent longitude, latitude, and altitude, respectively; for ground UAV nodes, the three dimensions represent the x-coordinate, y-coordinate, and ground elevation in a predefined Cartesian coordinate system, respectively. The node velocity vector is a three-dimensional vector, obtained by integrating data from the INS and performing Kalman filtering on Doppler shift measurements from the GNSS. The node acceleration vector is a three-dimensional vector, obtained by directly measuring data from a triaxial accelerometer and performing temperature compensation and zero-bias correction. All broadcast node motion parameters are accompanied by a timestamp.

[0014] Step S120: Construct a set of topological connections for a heterogeneous unmanned cluster based on the node attribute category and the node motion parameters. The set of topological connections includes the existence identifier of the communication link between each unmanned node and other unmanned nodes in the detection area, as well as the relative motion constraint boundary parameters.

[0015] Step S121: Based on the node attribute categories, perform attribute classification processing on all unmanned nodes in the heterogeneous unmanned cluster to obtain a first attribute category node affiliation set and a second attribute category node affiliation set. The unmanned nodes in the first attribute category node affiliation set have a first motion capability parameter set and a first perception capability parameter set. The unmanned nodes in the second attribute category node affiliation set have a second motion capability parameter set and a second perception capability parameter set. The first motion capability parameter set includes a first maximum velocity amplitude parameter and a first maximum acceleration amplitude parameter. The second motion capability parameter set includes a second maximum velocity amplitude parameter and a second maximum acceleration amplitude parameter. The first perception capability parameter set includes a first detection distance amplitude parameter and a first detection angle range parameter. The second perception capability parameter set includes a second detection distance amplitude parameter and a second detection angle range parameter.

[0016] First, based on the node attribute category of each unmanned node obtained in step S110, all unmanned nodes are classified into two different sets. All unmanned nodes whose node attribute category is "airborne unmanned aerial vehicle node" are assigned to the first attribute category set; all unmanned nodes whose node attribute category is "ground unmanned vehicle node" are assigned to the second attribute category set. For each airborne unmanned aerial vehicle node in the first attribute category set, a first set of motion capability parameters and a first set of perception capability parameters are read from its local storage. The first maximum velocity amplitude parameter defines the upper limit of the node's velocity vector magnitude; the first maximum acceleration amplitude parameter defines the upper limit of its acceleration vector magnitude. The first detection distance amplitude parameter defines the maximum effective detection radius of the detection equipment carried by the node; the first detection angle range parameter defines the scanning angle range of the detection equipment in the horizontal and vertical directions. For each ground unmanned vehicle node in the second attribute category set, a second set of motion capability parameters and a second set of perception capability parameters are read. The second maximum velocity amplitude parameter represents the vehicle's maximum speed; the second maximum acceleration amplitude parameter represents its maximum acceleration and deceleration capability. The second detection distance amplitude parameter defines the effective detection distance of its sensing sensor; the second detection angle range parameter defines the horizontal and vertical field of view of its sensing sensor.

[0017] Step S122: For each unmanned node in the first attribute category node belonging set, traverse other unmanned nodes within its detection area, and obtain the first spatial interval distance parameter and the first relative velocity difference parameter between each unmanned node and other unmanned nodes within its detection area. The first spatial interval distance parameter is used to characterize the straight-line distance between two unmanned nodes in the spatial position coordinates, and the first relative velocity difference parameter is used to characterize the magnitude and direction of the difference vector between the two unmanned nodes in the velocity vector.

[0018] For each UAV node in the set to which the first attribute category node belongs, a spherical detection area is delineated with its node spatial coordinates as the center and its first detection distance amplitude parameter as the radius. Other UAV nodes in the cluster are traversed to determine whether their node spatial coordinates fall within this detection area. For each other UAV node that falls within the detection area, the first spatial interval distance parameter is obtained by calculating the three-dimensional Euclidean distance between the node spatial coordinates of the two nodes. Simultaneously, a difference vector is obtained by vector subtraction of the node velocity vectors of the two nodes. The magnitude of this difference vector is calculated to obtain the magnitude part of the first relative velocity difference parameter. The direction part of the first relative velocity difference parameter is obtained by dividing each component of this difference vector by its magnitude.

[0019] Step S123: Based on the comparison between the first spatial interval distance parameter and the first detection distance amplitude parameter, determine the existence identifier of the first communication link between each unmanned node in the first attribute category node affiliation set and other unmanned nodes. The first communication link existence identifier includes a first link establishment identifier and a first link maintenance duration parameter. The first link maintenance duration parameter is used to characterize the cumulative length of time from the establishment of the communication link to the current moment.

[0020] The value of the first spatial interval distance parameter is compared with the value of the first detection range amplitude parameter. If the first spatial interval distance parameter is less than or equal to the first detection range amplitude parameter, the first link establishment flag is set to 1; otherwise, it is set to 0. When the first link establishment flag is 1, the locally stored link status history is queried. If the link did not exist in the previous time, the current time is recorded as the link establishment start time; if it already exists, the original start time is maintained. The first link maintenance duration parameter is calculated by subtracting the link establishment start time timestamp from the current time timestamp.

[0021] Step S124: Determine the first relative motion constraint boundary parameter between each unmanned node in the first attribute category node affiliation set and other unmanned nodes based on the direction of the difference vector in the first relative velocity difference parameter and the first attribute category node affiliation set. The first relative motion constraint boundary parameter includes a first relative motion direction angle range parameter and a first relative motion velocity ratio range parameter. The first relative motion direction angle range parameter is used to limit the acceptable range of the motion direction angle between two unmanned nodes, and the first relative motion velocity ratio range parameter is used to limit the acceptable range of the ratio of the velocity magnitudes between two unmanned nodes.

[0022] For each unmanned aerial vehicle (UAV) node in the set to which the first attribute category node belongs, the first relative motion constraint boundary parameter between it and other UAV nodes within the detection area is determined based on the direction of the difference vector in the first relative velocity difference parameter. The first relative motion direction angle range parameter defines an acceptable angle range, which is determined by the cosine of the angle between the velocity vectors of the two nodes being within a preset range. The first relative motion velocity ratio range parameter defines an acceptable velocity ratio range, which is limited by the numerical range of the ratio of the velocity magnitudes of the two nodes.

[0023] Step S125: For each unmanned node in the set to which the second attribute category node belongs, traverse the other unmanned nodes within its detection area, and obtain the second spatial interval distance parameter and the second relative velocity difference parameter between each unmanned node and the other unmanned nodes within its detection area. The second spatial interval distance parameter is used to characterize the linear distance between two unmanned nodes in the spatial position coordinates, and the second relative velocity difference parameter is used to characterize the magnitude and direction of the difference vector between the two unmanned nodes in the velocity vector.

[0024] For each ground-based unmanned vehicle node in the set to which the second attribute category nodes belong, a spherical detection area is delineated with its node spatial coordinates as the center point and its second detection distance amplitude parameter as the radius. Other unmanned nodes in the cluster are traversed to determine whether their node spatial coordinates fall within this detection area. For each other unmanned node that falls within the detection area, the second spatial interval distance parameter is obtained by calculating the three-dimensional Euclidean distance between the node spatial coordinates of the two nodes. Simultaneously, a difference vector is obtained by vector subtraction of the node velocity vectors of the two nodes. The magnitude of this difference vector is calculated to obtain the magnitude part of the second relative velocity difference parameter. The direction part of the second relative velocity difference parameter is obtained by dividing each component of this difference vector by its magnitude.

[0025] Step S126: Based on the comparison between the second spatial interval distance parameter and the second detection distance amplitude parameter, determine the existence identifier of the second communication link between each unmanned node in the second attribute category node affiliation set and other unmanned nodes. The existence identifier of the second communication link includes the second link establishment identifier and the second link maintenance duration parameter. The second link maintenance duration parameter is used to characterize the cumulative length of time from the establishment of the communication link to the current time.

[0026] The value of the second spatial interval distance parameter is compared with the value of the second detection range amplitude parameter. If the second spatial interval distance parameter is less than or equal to the second detection range amplitude parameter, the second link establishment flag is set to 1; otherwise, it is set to 0. When the second link establishment flag is 1, the locally stored link status history is queried. If the link did not exist in the previous time, the current time is recorded as the link establishment start time; if it already exists, the original start time is maintained. The second link maintenance duration parameter is calculated by subtracting the link establishment start time timestamp from the current time timestamp.

[0027] Step S127: Determine the second relative motion constraint boundary parameter between each unmanned node in the second attribute category node belonging set and other unmanned nodes based on the direction of the difference vector in the second relative velocity difference parameter and the second attribute category node belonging set. The second relative motion constraint boundary parameter includes a second relative motion direction angle range parameter and a second relative motion velocity ratio range parameter. The second relative motion direction angle range parameter is used to limit the acceptable range of the motion direction angle between two unmanned nodes, and the second relative motion velocity ratio range parameter is used to limit the acceptable range of the ratio of the velocity magnitudes between two unmanned nodes.

[0028] For each ground-based unmanned vehicle node in the set to which the second attribute category nodes belong, the second relative motion constraint boundary parameter between it and other unmanned nodes within the detection area is determined based on the direction of the difference vector in the second relative velocity difference parameter. The second relative motion direction angle range parameter defines an acceptable angle range, which is determined by the cosine of the angle between the velocity vectors of the two nodes being within a preset range. The second relative motion velocity ratio range parameter defines an acceptable velocity ratio range, which is limited by the numerical range of the ratio of the velocity magnitudes of the two nodes.

[0029] Step S128: Based on the existence identifier of the first communication link and the boundary parameter of the first relative motion constraint, perform a first association mapping process on each unmanned node in the first attribute category node belonging set and other unmanned nodes in its detection area to generate a first local topology connection relationship subset corresponding to the first attribute category node.

[0030] For each UAV node in the set to which the first attribute category node belongs, a first local topological connection subset is constructed based on the existence identifier of the first communication link and the first relative motion constraint boundary parameter. This first local topological connection subset, with the UAV node as the central node, stores information about all its neighboring nodes in the form of a data table. For each other UAV node whose first link establishment identifier is value 1, a new record is added to this data table, containing the unique identifier of that neighboring node and the edge attribute parameter between that node and that neighboring node. Specifically, the edge attribute parameter includes the existence identifier of the first communication link and the first relative motion constraint boundary parameter.

[0031] Step S129: Based on the second communication link existence identifier and the second relative motion constraint boundary parameter, perform a second association mapping process on each unmanned node in the second attribute category node belonging set and other unmanned nodes in its detection area to generate a second local topology connection relationship subset corresponding to the second attribute category node.

[0032] For each ground-based unmanned vehicle node in the set to which the second attribute category nodes belong, a second local topology connection subset is constructed based on the existence identifier of the second communication link and the boundary parameters of the second relative motion constraint. This second local topology connection subset, centered on the ground-based unmanned vehicle node, stores information about all its neighboring nodes in the form of a data table. For each other unmanned node whose second link establishment identifier is value 1, a new record is added to this data table, containing the unique identifier of that neighboring node and the edge attribute parameters between that node and its neighboring node. These edge attribute parameters specifically include the existence identifier of the second communication link and the boundary parameters of the second relative motion constraint.

[0033] Step S1210: Merge the first local topology connection subset and the second local topology connection subset to generate a topology connection set of the heterogeneous unmanned cluster. The topology connection set includes a list of neighbor node identifiers for each unmanned node and a set of edge attribute parameters between each unmanned node and its neighbor nodes. The set of edge attribute parameters includes the existence identifier of the communication link and the relative motion constraint boundary parameters.

[0034] The first local topological connection subset of all aerial unmanned aerial vehicle nodes and the second local topological connection subset of all ground unmanned vehicle nodes are merged to form a topological connection set covering the entire heterogeneous unmanned cluster. This topological connection set is organized in the form of an adjacency list, containing multiple entries equal to the total number of unmanned nodes in the cluster. Each entry corresponds to one unmanned node, and each entry stores a list of neighbor node identifiers for that unmanned node, as well as a set of edge attribute parameters corresponding to that list. Each element in the set of edge attribute parameters encapsulates the existence identifier of the communication link between the unmanned node and its corresponding neighbor node, and the relative motion constraint boundary parameters.

[0035] Step S130: Construct a hierarchical game control structure based on the set of topological connections. The hierarchical game control structure includes an upper-level strategy game domain and a lower-level action execution domain. The upper-level strategy game domain is used to perform global task decomposition and collaborative path orchestration on the heterogeneous unmanned cluster. The lower-level action execution domain is used to perform local motion control and obstacle avoidance response processing on each unmanned node.

[0036] Based on the set of topological connections constructed in step S120, a two-layer collaborative control structure, namely a hierarchical game control structure, is constructed. This structure decomposes the complex cluster collaborative control problem into two sub-problems with different time scales and information granularities, which are handled by the upper-layer strategy game domain and the lower-layer action execution domain, respectively.

[0037] Step S131: Extract the neighbor node identifier list of all unmanned nodes in the heterogeneous unmanned cluster according to the topological connection relationship set, and obtain the relative motion constraint boundary parameters in the edge attribute parameter set between each unmanned node and its neighbor nodes. The relative motion constraint boundary parameters include the relative motion direction angle range parameter and the relative motion speed ratio range parameter.

[0038] From the set of topological connections generated in step S1210, extract the list of neighbor node identifiers for each unmanned node. For each neighbor node in this list, further extract the relative motion constraint boundary parameters corresponding to that neighbor node from the set of edge attribute parameters. These relative motion constraint boundary parameters specifically include the range parameters of the relative motion direction angle and the range parameters of the relative motion velocity ratio.

[0039] Step S132: Based on the node attribute category and the relative motion constraint boundary parameter, perform game hierarchy division processing on the unmanned nodes in the heterogeneous unmanned cluster to obtain the first game node set corresponding to the upper-level strategy game domain and the second game node set corresponding to the lower-level action execution domain. The unmanned nodes in the first game node set have global information interaction capabilities, and the unmanned nodes in the second game node set have local perception and action execution capabilities.

[0040] Based on the node attribute category of each unmanned node and its relative motion constraint boundary parameters in the set of topological connections, its game level is determined. Nodes whose attribute category is aerial unmanned aerial vehicle (UAV) node and whose number of neighboring nodes exceeds a preset threshold are judged to have strong communication and coordination capabilities, suitable for undertaking global coordination tasks, and are assigned to the first game node set corresponding to the upper-level strategy game domain. The remaining nodes, especially ground unmanned vehicle nodes or aerial UAV nodes with fewer neighboring nodes, are judged to be more suitable for performing local autonomous tasks and are assigned to the second game node set corresponding to the lower-level action execution domain.

[0041] Step S133: Assign a strategy generator module in the upper-level strategy game domain to each unmanned node in the first set of game nodes. The strategy generator module is used to generate a candidate strategy set for each unmanned node based on the global topology structure information in the set of topology connections. The candidate strategy set includes multiple candidate path planning schemes and multiple candidate task allocation schemes.

[0042] For each unmanned node in the first game node set, a strategy generator module is instantiated on its onboard computing unit. This strategy generator module takes the global topology information from the topology connection set generated in step S1210 as input. This global topology information includes the node connections and edge attributes of the entire cluster. Internally, the strategy generator module runs a Monte Carlo tree search-based planning algorithm. It explores paths on the graph structure defined by the topology connection set, simulating possible paths to different neighboring nodes starting from the current node position. Each simulation extends by a certain number of steps, forming a candidate path planning scheme, represented by a sequence of spatial coordinates of the nodes along the path. Simultaneously, based on the unfinished parts of the current global task and the capabilities of each node, the algorithm generates multiple candidate task allocation schemes. Each candidate task allocation scheme is represented by a mapping table between unmanned node identifiers and subtask identifiers. After multiple simulations, the strategy generator module outputs a candidate strategy set, where each element contains a candidate path planning scheme and a candidate task allocation scheme.

[0043] Step S134: Assign a motion controller module in the lower-level action execution domain to each unmanned node in the second set of game nodes. The motion controller module is used to generate real-time motion control parameters for each unmanned node based on the strategy selection result output by the upper-level strategy game domain. The real-time motion control parameters include velocity vector adjustment parameters and acceleration vector adjustment parameters.

[0044] For each unmanned node in the second set of game nodes, a motion controller module is instantiated on its onboard computing unit. The core of this motion controller module is a model predictive controller (MMC), which receives the policy selection results from the upper-level policy game domain and transforms them into real-time motion control commands executable in the actual physical environment. The MMC maintains a predictive model of the unmanned node's kinematics, capable of predicting its trajectory over a future period based on its current state and control input. Using a sequence of path points in a selected path planning scheme as a reference trajectory, the controller solves an optimization problem in each control cycle. The goal of this optimization problem is to minimize the deviation between the predicted trajectory and the reference trajectory, while satisfying the node's own motion capability constraints and relative motion constraint boundary parameters. The solution to this optimization problem yields the real-time motion control parameters, which include velocity vector adjustment parameters and acceleration vector adjustment parameters.

[0045] Step S135: Establish an information transmission channel between the strategy generator module in the upper-level strategy game domain and the motion controller module in the lower-level action execution domain. The information transmission channel is used to transmit the strategy selection results generated in the upper-level strategy game domain to the lower-level action execution domain. The strategy selection results include the path planning scheme identifier and task allocation scheme identifier selected by each unmanned node.

[0046] Above the cluster's communication network layer, a logical information transmission channel is established, implemented using a topic-based publish-subscribe pattern. In the upper-layer strategy game domain, the strategy generator module, after optimizing the strategy, encapsulates the strategy selection result into a message of a specific format. This message contains the path planning scheme identifier and task allocation scheme identifier selected by each unmanned node, and publishes this message to a predefined topic. In the lower-layer action execution domain, the motion controller module subscribes to this topic. When a new message is published, the motion controller module receives the message and parses out its own path planning scheme identifier and task allocation scheme identifier to obtain the reference trajectory to follow.

[0047] Step S136: Generate an update frequency parameter for the topology connection set based on the change frequency of the neighbor node identifier list in the topology connection set. The update frequency parameter is used to indicate the degree of drastic change of the topology connection set in the time dimension.

[0048] Within a preset time window, the topology connection set generated in step S1210 is continuously monitored. At fixed sampling intervals, the current topology connection set is recorded and compared with the topology connection set at the previous sampling time. The number of neighbor node identifiers and edge attribute parameters that changed within the time window are counted. The total number of changes is divided by the length of the time window to obtain the average number of changes per unit time, which serves as the update frequency parameter for the topology connection set.

[0049] Step S137: Based on the transmission delay parameter of the information transmission channel and the update frequency parameter of the topology connection relationship set, set a collaborative time window parameter between the upper-level strategy game domain and the lower-level action execution domain. The collaborative time window parameter is used to synchronize the strategy update time of the upper-level strategy game domain and the control command issuance time of the lower-level action execution domain.

[0050] The transmission delay parameter of the information transmission channel is measured. This parameter is obtained by recording the timestamp difference between the message publication and reception times and averaging the values. The reciprocal of the update frequency parameter of the topology connection set generated in step S136 is taken to obtain the average time interval of topology changes. The value of the collaborative time window parameter is set as the weighted sum of this transmission delay parameter and the average time interval of topology changes. This collaborative time window parameter defines the length of time after the upper-level strategy game domain performs a strategy update, during which the strategy result is effectively applied in the lower-level action execution domain. The strategy update operation of the upper-level strategy game domain is triggered periodically according to the value of this collaborative time window parameter, and the control command issuance time of the lower-level action execution domain is also synchronized accordingly.

[0051] Step S138: Generate a hierarchical interaction density parameter for the hierarchical game control structure based on the number of unmanned nodes in the first game node set and the number of unmanned nodes in the second game node set. The hierarchical interaction density parameter is used to indicate the information interaction frequency between the upper-level strategy game domain and the lower-level action execution domain.

[0052] The total number of nodes in the cluster is obtained by summing the number of unattended nodes in the first set of game nodes and the number of unattended nodes in the second set of game nodes. The value of the hierarchical interaction density parameter is obtained by dividing the number of unattended nodes in the first set of game nodes by the total number of nodes in the cluster; this value is between 0 and 1.

[0053] Step S139: Based on the changing trend of the set of edge attribute parameters in the set of topological connections, dynamically adjust the numerical range of the collaborative time window parameter to generate a dynamically adjusted collaborative time window parameter. Based on the dynamically adjusted collaborative time window parameter and the hierarchical interaction density parameter, construct a time synchronization mechanism between the upper-level strategy game domain and the lower-level action execution domain. The time synchronization mechanism is used to ensure the alignment of strategy update operation and control instruction issuance operation in the time dimension.

[0054] The changing trend of the edge attribute parameter set in the topological connection set generated in step S1210 is continuously monitored. The number of edges that change in the edge attribute parameter set within each control cycle is recorded, and the moving average of this number is calculated. If the moving average shows an upward trend, the value of the collaborative time window parameter set in step S137 is multiplied by an adjustment factor less than 1 to obtain the dynamically adjusted collaborative time window parameter; if it shows a downward trend, it is multiplied by an adjustment factor greater than 1. Based on the dynamically adjusted collaborative time window parameter and the hierarchical interaction density parameter generated in step S138, a timing synchronization mechanism is constructed. It is driven by a global clock synchronization signal. The strategy update operation of the upper-level strategy game domain is triggered on each rising edge of this synchronization signal. The calculation period of the strategy update is the value of the dynamically adjusted collaborative time window parameter. The control command issuance operation of the lower-level action execution domain is triggered after a fixed delay after the strategy update is completed.

[0055] Step S140: Assign a strategy payoff function and game interaction rules to each unmanned node through the upper-level strategy game domain, and update the strategy payoff function of each unmanned node according to the changes in the set of topological connections, thereby generating a hierarchical game control instruction containing the optimized strategy selection results of each unmanned node.

[0056] In the upper-level strategy game domain, the driving force and behavioral criterion for each unmanned node participating in the game are defined, and these driving forces are adjusted in real time based on the dynamically changing topology, ultimately calculating an optimal decision scheme for each node.

[0057] Step S141: Assign an initial strategy benefit function to each unmanned node in the heterogeneous unmanned cluster. The initial strategy benefit function is constructed through the following steps: determine the task completion progress metric based on the node attribute category of the unmanned node and the current task allocation result, and convert the task completion progress metric into a dimensionless task completion benefit utility value according to a preset first normalization function; determine the energy consumption metric based on the velocity vector and acceleration vector in the node motion parameters of the unmanned node, and convert the energy consumption metric into a dimensionless energy consumption cost utility value according to a preset second normalization function; add the task completion benefit utility value and the energy consumption cost utility value according to a preset weight to obtain the initial strategy benefit function, which is a dimensionless comprehensive utility value.

[0058] An initial policy benefit function is constructed for each unmanned node in the heterogeneous unmanned cluster. The list of subtasks currently assigned to the node is obtained from the global task management module, and the proportion of completed subtasks to the total number of assigned subtasks is calculated; this proportion is the task completion progress metric. A preset first normalization function is used to convert this task completion progress metric into a dimensionless task completion benefit utility value in the range of 0 to 1. Velocity and acceleration vectors are extracted from the node motion parameters obtained in step S110, and the weighted sum of the squares of the magnitudes of the velocity and acceleration vectors is calculated to obtain the energy consumption metric. A preset second normalization function is used to map this energy consumption metric to the range of 0 to 1, obtaining a dimensionless energy consumption cost utility value. The task completion benefit utility value and the energy consumption cost utility value are added according to preset weights to obtain the initial policy benefit function.

[0059] Step S142: Based on the neighbor node identifier list of each unmanned node in the topological connection relationship set, obtain the relative motion constraint boundary parameters in the edge attribute parameter set between each unmanned node and its neighbor nodes. Perform a first correction process on the initial strategy benefit function of each unmanned node based on the relative motion constraint boundary parameters to obtain a first corrected strategy benefit function. The first correction process is performed through the following steps: Calculate the motion direction consistency metric and motion speed matching degree metric between the unmanned node and its neighbor nodes based on the relative motion constraint boundary parameters; convert the motion direction consistency metric into a dimensionless direction collaborative utility value according to a preset third normalization function, and convert the motion speed matching degree metric into a dimensionless speed collaborative utility value according to a preset fourth normalization function; combine the direction collaborative utility value and the speed collaborative utility value according to a preset weight to obtain a neighbor node collaborative benefit utility value; add the neighbor node collaborative benefit utility value to the initial strategy benefit function to obtain the first corrected strategy benefit function, which is a dimensionless comprehensive utility value.

[0060] Obtain the neighbor node identifier list from the topological connection set generated in step S1210, and extract the relative motion constraint boundary parameters from the edge attribute parameter set for each neighbor node in the list. Calculate the motion direction consistency metric between the node and all neighbor nodes. For each neighbor node, calculate the cosine of the angle between the node's velocity vector and the velocity vectors of its neighbors, compare this cosine value with a preset ideal value, and average the consistency index of all neighbor nodes. Calculate the motion speed matching degree metric between the node and all neighbor nodes. For each neighbor node, calculate the ratio of the node's velocity modulus to the velocity modulus of its neighbors, compare this ratio with a preset ideal ratio range, and average the matching degree index of all neighbor nodes. Use a preset third normalization function to convert the motion direction consistency metric into a direction cooperation utility value, and use a preset fourth normalization function to convert the motion speed matching degree metric into a speed cooperation utility value. Sum the direction cooperation utility value and the speed cooperation utility value according to preset weights to obtain the neighbor node cooperation benefit utility value. The collaborative benefit utility value of the neighbor node is added to the initial policy benefit function obtained in step S141 to obtain the first modified policy benefit function.

[0061] Step S143: Assign game interaction rules to each unmanned node in the heterogeneous unmanned cluster. The game interaction rules include strategy update trigger conditions and strategy selection interaction methods. The strategy update trigger conditions are determined according to the update frequency parameter of the topological connection relationship set. The strategy selection interaction methods include non-cooperative game interaction modes and cooperative game interaction modes.

[0062] Game interaction rules are assigned to each unmanned node. The strategy update trigger condition is determined based on the update frequency parameter of the topology connection set generated in step S136. A dynamic threshold is set, which is proportional to the topology update frequency parameter. When the topology update frequency parameter exceeds the threshold, the strategy update trigger condition is set to update the strategy every control cycle. When it is below the threshold, the trigger cycle is inversely proportional to the topology update frequency parameter. The strategy selection interaction mode is dynamically determined according to the game level to which the node belongs and the current task stage. If the node belongs to the first game node set and is currently in the global task coordination stage, a cooperative game interaction mode is adopted. In this mode, when making strategy selection, the node exchanges payoff evaluation information with other first game nodes through the information transmission channel to jointly seek the strategy combination that maximizes the total group payoff. If the node belongs to the second game node set or is currently in the local obstacle avoidance stage, a non-cooperative game interaction mode is adopted. In this mode, the node makes strategy selection independently only based on its own perceived local information and payoff function.

[0063] Step S144: Monitor the change events of the neighbor node identifier list of each unmanned node in the topological connection relationship set. When a change in the neighbor node identifier list is detected, re-acquire the relative motion constraint boundary parameters corresponding to the changed neighbor node based on the changed neighbor node identifier list. Perform a second correction process on the first correction strategy benefit function of the unmanned node corresponding to the changed neighbor node based on the re-acquired relative motion constraint boundary parameters to obtain the second correction strategy benefit function. The second correction process is performed through the following steps: determine the topological structure change impact metric based on the re-acquired relative motion constraint boundary parameters and the event type of neighbor node joining or leaving; convert the topological structure change impact metric into a dimensionless topological structure change compensation utility value according to the preset fifth normalization function; add the topological structure change compensation utility value to the first correction strategy benefit function to obtain the second correction strategy benefit function, which is a dimensionless comprehensive utility value.

[0064] The neighbor node identifier list of each unmanned node in the topology connection set generated in step S1210 is continuously monitored. When a change in the neighbor node identifier list of a node is detected, the relative motion constraint boundary parameters corresponding to the changed neighbor node are re-acquired based on the changed neighbor node identifier list, and the type of change event, i.e., joining event or leaving event, is recorded. The calculation method of the topology change impact metric is as follows: For joining events, the deviation of the relative motion direction angle and velocity ratio between the newly joined neighbor node and the node from their respective constraint boundary parameters is calculated. The smaller the deviation, the greater the impact metric. For leaving events, the deviation of the relative motion state of the leaving neighbor node at the last moment before leaving from the constraint boundary parameters is calculated. The greater the deviation, the greater the impact metric. The topology change impact metric is converted into a dimensionless topology change compensation utility value between -1 and 1 using a preset fifth normalization function. Positive impact corresponds to positive compensation value, and negative impact corresponds to negative compensation value. The topology change compensation utility value is added to the first correction strategy benefit function obtained in step S142 to obtain the second correction strategy benefit function.

[0065] Step S145: Input the second modified strategy payoff function into the strategy generator module of the upper-level strategy game domain, and use the strategy generator module to perform payoff evaluation processing on each candidate strategy in the candidate strategy set of each unmanned node to obtain the payoff evaluation value corresponding to each candidate strategy.

[0066] The second modified strategy benefit function obtained in step S144 is used as the evaluation basis and input into the strategy generator module allocated to the node in step S133. The strategy generator module traverses each candidate strategy in its internally stored candidate strategy set, simulates the execution of each candidate strategy, and calculates the expected benefit that can be obtained after executing the strategy according to the second modified strategy benefit function. This benefit evaluation value is the dimensionless comprehensive utility value calculated by substituting the various indicators of the strategy into the second modified strategy benefit function.

[0067] Step S146: According to the strategy selection interaction method in the game interaction rules, compare and select the payoff evaluation values ​​of all candidate strategies in the candidate strategy set of each unmanned node to obtain the optimized strategy selection result of each unmanned node. The optimized strategy selection result includes the selected path planning scheme and the selected task allocation scheme.

[0068] The strategy selection is performed according to the strategy selection interaction mode in the game interaction rules allocated in step S143. In a cooperative game interaction mode, each node sends the payoff evaluation value of its candidate strategies to other first-player nodes through the information transmission channel. All first-player nodes jointly execute a distributed consensus algorithm, iteratively exchanging their respective candidate strategy payoff information, and finally converging to a strategy combination that maximizes the total group payoff. The strategy corresponding to each node is its optimized strategy selection result. In a non-cooperative game interaction mode, each node independently sorts the payoff evaluation values ​​of all candidate strategies in its own candidate strategy set in descending order and selects the candidate strategy with the highest payoff evaluation value as its optimized strategy selection result. The optimized strategy selection result includes a selected path planning scheme and a selected task allocation scheme.

[0069] Step S147: Based on the selected path planning scheme and the selected task allocation scheme in the optimization strategy selection results, generate a strategy execution instruction for each unmanned node, package all the strategy execution instructions of the unmanned nodes, and generate a hierarchical game control instruction containing the optimization strategy selection results of each unmanned node. The strategy execution instruction includes a path point sequence and the coordinates of the task target point.

[0070] For each unmanned node, a corresponding strategy execution instruction is generated based on the optimization strategy selection result obtained in step S146. This instruction contains two fields: a path point sequence field storing the specific path point coordinate sequence corresponding to the selected path planning scheme in the optimization strategy selection result; and a task target point coordinate field storing the target location coordinates of the subtask pointed to by the selected task allocation scheme in the optimization strategy selection result. All unmanned node strategy execution instructions are indexed according to their node identifiers and packaged into a single data frame, which is the hierarchical game control instruction.

[0071] Step S150: Receive the hierarchical game control command through the lower-level action execution domain, generate real-time motion control parameters for each unmanned node based on the hierarchical game control command and the set of topological connections, and send the real-time motion control parameters to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion.

[0072] The lower-level action execution domain receives the hierarchical game control command generated in step S147, and uses the command and the latest set of topological connections to calculate accurate real-time motion control commands that satisfy physical constraints for each unmanned node.

[0073] Step S151: Receive the hierarchical game control instructions through the motion controller module of the lower-level action execution domain, and parse the strategy execution instructions of each unmanned node from the hierarchical game control instructions. The strategy execution instructions include the path point sequence and the coordinates of the task target point.

[0074] In the lower-level action execution domain, the motion controller module of each unmanned node listens to the information transmission channel. When the hierarchical game control command is issued, it receives the command and parses its own strategy execution command from the data packet of the command according to its own node identifier, and obtains the path point sequence and task target point coordinates contained in the command.

[0075] Step S152: Obtain the current node motion parameters of each unmanned node, wherein the current node motion parameters include the current node spatial position coordinates, the current node velocity vector, and the current node acceleration vector.

[0076] The motion controller module obtains the latest node motion parameters from the local state estimation unit, namely the node spatial position coordinates, node velocity vector, and node acceleration vector at the current moment.

[0077] Step S153: Based on the current node spatial coordinates of each unmanned node and the path point sequence in the strategy execution instruction, calculate the expected velocity vector and expected acceleration vector required for each unmanned node to move from the current node spatial coordinates to the next path point in the path point sequence.

[0078] The motion controller module parses the path point sequence in the strategy execution instructions and extracts the first path point in the sequence that has not yet been reached as the current target point. It calculates the position error vector pointing from the current node's spatial coordinates to the target point, and uses a proportional-derivative controller to calculate the desired velocity vector. The desired velocity vector equals the position error vector multiplied by the proportional coefficient, plus the difference between the current node's velocity vector and the target point's moving velocity multiplied by the derivative coefficient. The desired acceleration vector is obtained by performing time difference calculation on the desired velocity vector, i.e., subtracting the desired velocity vector of the previous cycle from the desired velocity vector of the current cycle and then dividing by the control cycle duration.

[0079] Step S154: Based on the neighbor node identifier list of each unmanned node in the topological connection relationship set, obtain the relative motion constraint boundary parameters in the edge attribute parameter set between each unmanned node and its neighbor nodes. Based on the relative motion constraint boundary parameters, perform a first constraint processing on the expected velocity vector and the expected acceleration vector to obtain the first-constrained velocity vector and the first-constrained acceleration vector. The first constraint processing is used to ensure that the relative motion direction angle between the unmanned node and its neighbor nodes is within the relative motion direction angle range parameter, and the relative motion velocity ratio is within the relative motion velocity ratio range parameter.

[0080] The neighbor node identifier list of the node is obtained from the topological connection relationship set generated in step S1210, and the relative motion constraint boundary parameters in the edge attribute parameter set corresponding to each neighbor node are extracted. The first constraint processing is implemented through an iterative projection algorithm. For each neighbor node, the ratio of the relative motion direction between the node and the neighbor node and the velocity is calculated when the node moves according to the expected velocity vector calculated in step S153. If the angle or ratio exceeds the range of the corresponding constraint boundary parameter, a correction vector is calculated to project the expected velocity vector into the feasible space that satisfies the constraint. The constraints of all neighbor nodes are projected one by one and iterated multiple times so that the expected velocity vector gradually converges to the feasible region that simultaneously satisfies the constraints of all neighbor nodes, thus obtaining the velocity vector after the first constraint. The acceleration vector after the first constraint is obtained by time difference of the velocity vector after the first constraint.

[0081] Step S155: Perform a second constraint process on the first constrained velocity vector and the first constrained acceleration vector according to the set of motion capability parameters corresponding to the node attribute category of each unmanned node, to obtain the second constrained velocity vector and the second constrained acceleration vector. The second constraint process is used to ensure that the magnitude of the velocity vector of the unmanned node does not exceed the maximum velocity magnitude parameter corresponding to the node attribute category, and the magnitude of the acceleration vector does not exceed the maximum acceleration magnitude parameter corresponding to the node attribute category.

[0082] Based on the node attribute category of the unmanned node, the corresponding set of motion capability parameters is read from step S121. The second constraint processing first applies amplitude constraints to the velocity vector after the first constraint, calculating its magnitude. If this magnitude is greater than the maximum velocity amplitude parameter corresponding to the node category, each component of the velocity vector is proportionally reduced until its magnitude equals the maximum velocity amplitude parameter, resulting in the second constrained velocity vector. Similarly, amplitude constraints are applied to the acceleration vector after the first constraint, and its magnitude is calculated. If the magnitude is greater than the corresponding maximum acceleration amplitude parameter, each component is proportionally reduced until its magnitude equals the maximum acceleration amplitude parameter, resulting in the second constrained acceleration vector.

[0083] Step S156: Input the second constrained velocity vector and the second constrained acceleration vector into the motion planning submodule of the motion controller module, and generate real-time motion control parameters for each unmanned node through the motion planning submodule. The real-time motion control parameters include velocity vector control quantity and acceleration vector control quantity.

[0084] The motion controller module contains a motion planning submodule, the core of which is a trajectory tracker based on model predictive control. This trajectory tracker uses the second constrained velocity vector and the second constrained acceleration vector obtained in step S155 as desired reference inputs. It uses the kinematic model of the unmanned node to predict the state trajectory in the future finite time domain and solves an optimization problem. The goal of this optimization problem is to minimize the deviation between the predicted state and the desired reference while considering the physical constraints of the actuator. The solution to the optimization problem is a control sequence, the first of which is output as the real-time motion control parameter at the current moment. This real-time motion control parameter specifically includes velocity vector control and acceleration vector control.

[0085] Step S157: Based on the changing trend of the edge attribute parameter set in the topological connection relationship set, dynamically adjust the output frequency of the real-time motion control parameters to generate a dynamically adjusted real-time motion control parameter output frequency. Based on the dynamically adjusted real-time motion control parameter output frequency, group the real-time motion control parameters according to the time sequence to obtain the grouped real-time motion control parameter sequence and send it to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion.

[0086] The changing trend of the set of edge attribute parameters in the topological connection set generated in step S1210 is monitored, and the moving average of the number of changes in edge attribute parameters within each control cycle is calculated. When this moving average increases, the command output frequency of the motion planning submodule is increased to obtain the dynamically adjusted real-time motion control parameter output frequency. According to this output frequency, the motion planning submodule groups the real-time motion control parameters generated in step S156 in a time series. Every output cycle, the control quantity of the current cycle is packaged into a data packet and sent to the underlying flight controller or motion controller of the corresponding unmanned node through the wireless communication module, driving the node to adjust its own motion state according to the command.

[0087] Step S210: Monitor the node motion parameter change events of each unmanned node in the heterogeneous unmanned cluster. When a node motion parameter change event is detected, obtain the changed node motion parameters. The changed node motion parameters include the changed node spatial position coordinates, the changed node velocity vector, and the changed node acceleration vector.

[0088] Throughout the operation of the heterogeneous unmanned cluster, the onboard computing unit of each unmanned node continuously monitors its own node motion parameters at high frequency. When the change in any component of the node motion parameters exceeds a preset threshold, it is considered that a node motion parameter change event has occurred. Once the event is detected, the latest node motion parameters at the current moment are immediately obtained as the changed node motion parameters.

[0089] Step S220: Based on the changed spatial location coordinates of the nodes and the neighbor node identifier list of each unmanned node in the topological connection relationship set, recalculate the spatial distance parameter and relative velocity difference parameter between each unmanned node and other unmanned nodes in its detection area. Based on the recalculated spatial distance parameter and relative velocity difference parameter, update the communication link existence identifier and relative motion constraint boundary parameter between each unmanned node in the topological connection relationship set and other unmanned nodes in its detection area, and obtain the updated topological connection relationship set.

[0090] When the motion parameters of an unmanned node change, steps S122 to S127 are re-executed based on the changed spatial coordinates of the node, updating the existence identifiers of all communication links and the relative motion constraint boundary parameters related to that node. For the node itself, the detection area is redefined with its changed position as the center. The spatial distance parameters and relative velocity difference parameters of other nodes are recalculated, and the existence identifiers of communication links and the relative motion constraint boundary parameters between the node and other nodes within the detection area are updated accordingly. Simultaneously, for each of the node's original neighboring nodes, the changed node is re-evaluated to determine if it is still within its detection area, and the corresponding link identifiers and constraint boundary parameters are updated. After all updates are completed, a completely new set of topological connections is obtained.

[0091] Step S230: Compare the updated set of topology connections with the set of topology connections before the update, identify the list of neighbor node identifiers that have changed and the set of edge attribute parameters that have changed, and generate a topology change event notification based on the identified list of neighbor node identifiers that have changed and the set of edge attribute parameters that have changed. The topology change event notification includes changed node identifier information and change type identifier information.

[0092] The updated topology connection set obtained in step S220 is compared item by item with the topology connection set before the last update. The comparison objects include the neighbor node identifier list of each node and the set of edge attribute parameters for each edge. The comparison identifies which nodes' neighbor node identifier lists have changed and which edges' edge attribute parameters have changed. For each change, a topology change event notification is generated, which includes the identifier information of the changed node and the change type identifier information.

[0093] Step S240: Send the topology change event notification to the upper-level strategy game domain, triggering the upper-level strategy game domain to recalculate the strategy payoff function of the unmanned node according to the updated topology connection relationship set, and receive the updated hierarchical game control instruction generated by the upper-level strategy game domain after recalculation according to the updated topology connection relationship set.

[0094] The topology change event notification generated in step S230 is sent to the upper-level strategy game domain via the information transmission channel established in step S135. Upon receiving the notification, the upper-level strategy game domain triggers a recalculation of the strategy payoff function for the relevant unattended nodes. For the changed nodes mentioned in the notification, the second correction process in step S144 is re-executed based on the updated topology connection set to update their second corrected strategy payoff function. Subsequently, steps S145 and S146 are re-executed based on the updated payoff function to generate new optimized strategy selection results for the relevant nodes. Finally, step S147 is re-executed based on the new optimized strategy selection results to generate updated hierarchical game control instructions and send them back to the lower-level action execution domain.

[0095] Step S250: Receive the updated hierarchical game control command through the lower-level action execution domain, and generate updated real-time motion control parameters for each unmanned node based on the updated hierarchical game control command and the updated topology connection relationship set, and send them to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion after dynamic topology changes.

[0096] After receiving the updated hierarchical game control command sent in step S240, the lower-level action execution domain, based on the command and the updated topological connection set obtained in step S220, re-executes steps S151 to S157 to calculate the updated real-time motion control parameters for each relevant unmanned node and sends them to the corresponding unmanned node.

[0097] Step S260: Repeatedly execute the operations of monitoring node motion parameter change events, updating the topology connection relationship set, generating topology change event notifications, receiving updated hierarchical game control instructions, generating updated real-time motion control parameters, and sending updated real-time motion control parameters until the heterogeneous unmanned cluster completes all collaborative tasks.

[0098] Steps S210 to S250 form a closed-loop feedback loop that is continuously executed during cluster operation. Every significant change in node motion parameters triggers topology updates and policy re-optimization until the pre-defined global collaborative task is marked as completed.

[0099] Step S310: Based on the node attribute category and the node motion parameters, identify the key unmanned nodes in the heterogeneous unmanned cluster. The key unmanned nodes have information relay and forwarding capabilities and global task coordination capabilities.

[0100] Traverse all unmanned nodes in the heterogeneous unmanned cluster. Nodes whose attribute category is aerial unmanned aerial vehicle node and whose altitude component of their spatial location coordinates exceeds a preset altitude threshold, and whose number of nodes in their neighbor node identifier list exceeds a preset number threshold, are identified as having information relay forwarding capabilities. Nodes whose attribute category is aerial unmanned aerial vehicle node and whose remaining energy parameter exceeds a preset energy threshold are identified as having the potential to undertake additional computing and coordination tasks. Nodes that meet the above capability conditions are marked as key unmanned nodes.

[0101] Step S320: Assign the global strategy aggregator module in the upper-level strategy game domain to the key unmanned node. The global strategy aggregator module is used to collect the candidate strategy set and payoff evaluation value of all unmanned nodes in the first game node set.

[0102] For each unmanned aerial vehicle node marked as a key unmanned node, a global policy aggregator module is instantiated on its onboard computing unit. It is a lightweight distributed consensus service component that periodically sends requests to all other unmanned nodes in the first game node set through the information transmission channel established in step S135 to collect the candidate policy sets and corresponding profit evaluation values ​​generated by them in the current period.

[0103] Step S330: The global strategy aggregator module aggregates the collected candidate strategy set and the return evaluation values ​​to generate a global strategy evaluation result, which includes the global optimal strategy combination identifier and the global second-best strategy combination identifier.

[0104] After collecting candidate strategies and payoffs for all first-game nodes, the global strategy aggregator module executes an aggregation processing algorithm to combine all collected candidate strategies into all possible global strategy combinations. For each global strategy combination, it calculates its total global payoff, which is the weighted sum of the payoff evaluation values ​​of all nodes in the combination. Then, it sorts the total global payoffs of all global strategy combinations in descending order. The global strategy combination ranked first is marked as the global optimal strategy combination, and the global strategy combination ranked second is marked as the global suboptimal strategy combination. The global strategy evaluation result is the identifier corresponding to these two combinations.

[0105] Step S340: Generate a global strategy coordination instruction based on the global optimal strategy combination identifier in the global strategy evaluation result. The global strategy coordination instruction is used to instruct the unmanned nodes in the first game node set to perform cooperative operations according to the global optimal strategy combination.

[0106] Based on the global optimal strategy combination identifier obtained in step S330, the specific strategy combination content corresponding to the identifier is retrieved from the local cache, and a global strategy coordination instruction is generated. The broadcast message content instructs all first game nodes to switch to the global optimal strategy combination and is broadcast to all first game nodes through the information transmission channel.

[0107] Step S350: The global strategy coordination instruction is sent to each unmanned node in the first game node set, replacing the original optimization strategy selection result of each unmanned node in the first game node set. The hierarchical game control instruction is regenerated based on the replaced optimization strategy selection result to obtain the hierarchical game control instruction after global coordination.

[0108] After receiving the global strategy coordination instruction issued in step S340, each unmanned node in the first game node set parses out the global optimal strategy combination identifier and extracts the specific strategy assigned to itself from the local cache based on the identifier. It then replaces the optimized strategy selection result obtained autonomously in step S146 with this strategy. After all the first game nodes have completed the replacement, they re-execute step S147 to generate the global coordination hierarchical game control instruction based on these new strategy selection results.

[0109] Step S360: Receive the globally coordinated hierarchical game control command through the lower-level action execution domain, and generate the globally coordinated real-time motion control parameters for each unmanned node based on the globally coordinated hierarchical game control command and the set of topological connections, and send them to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform globally coordinated cooperative motion.

[0110] After receiving the global coordination hierarchical game control command generated in step S350, the lower-level action execution domain re-executes steps S151 to S157 based on the command and the latest set of topological connections to calculate the global coordination real-time motion control parameters for each relevant unmanned node and send them to the corresponding unmanned node.

[0111] Step S370: Monitor the motion state of the heterogeneous unmanned cluster driven by the real-time motion control parameters after global coordination, obtain motion state feedback parameters, adjust the aggregation weight parameters of the global policy aggregator module according to the motion state feedback parameters, and generate the adjusted aggregation weight parameters for the next global policy evaluation process. The motion state feedback parameters include the deviation parameters between the actual motion trajectory and the planned motion trajectory of each unmanned node.

[0112] During the collaborative motion driven by step S360, the motion state of the cluster is continuously monitored. For each unmanned node, the actual motion trajectory generated after the execution of global coordination and real-time motion control parameters is compared with the planned motion trajectory in the globally optimal strategy combination assigned to it in step S350. The Euclidean distance between the actual position and the planned position at each sampling point is calculated, and the root mean square value of the distance sequence is used as the deviation parameter of the node. The deviation parameters of all first game nodes are weighted and averaged to obtain the overall motion state feedback parameter. The motion state feedback parameter is used as a feedback signal to adjust the weight used in the aggregation process in step S330. If the deviation parameter of a node is large, its weight in the calculation of the global total gain is reduced in the next global strategy evaluation.

[0113] Step S410: Based on the neighbor node identifier list of each unmanned node in the topological connection relationship set, identify the topological connection-dense region and the topological connection-sparse region in the heterogeneous unmanned cluster. The number of neighbor nodes of the unmanned node in the topological connection-dense region is not less than a first threshold parameter, and the number of neighbor nodes of the unmanned node in the topological connection-sparse region is less than a second threshold parameter.

[0114] Iterate through the list of neighbor node identifiers for all unmanned nodes in the topology connection set generated in step S1210, and count the number of neighbor nodes for each node. Delineate a spatial region with a preset radius centered on the node's spatial location coordinates, and average the number of neighbor nodes for all nodes within this region to obtain the connection density of the region. Regions with a connection density value not lower than a first threshold parameter are marked as densely connected regions, and regions with a connection density value lower than a second threshold parameter are marked as sparsely connected regions.

[0115] Step S420: Assign a dense region collaborative controller module in the upper-level strategy game domain to the unmanned nodes in the densely connected topology region. The dense region collaborative controller module is used to perform group strategy coordination processing on the unmanned nodes in the densely connected topology region.

[0116] For all unmanned nodes marked as densely connected regions, a dense region cooperative controller module is instantiated on a key unmanned node selected as the cluster head within the region. This module runs a consensus-based distributed optimization algorithm whose objective function is the weighted sum of the policy reward functions of all nodes within the region. The constraint condition is that the relative motion constraint boundary parameters between nodes within the region must be satisfied.

[0117] Step S430: Obtain the candidate strategy set and benefit evaluation value of all unmanned nodes in the densely connected topology region through the densely connected topology collaborative controller module, and perform joint optimization processing on the candidate strategy set of all unmanned nodes in the densely connected topology region to generate a joint optimization strategy combination for the dense region.

[0118] The dense region collaborative controller module collects the candidate policy set and corresponding benefit evaluation values ​​of all unmanned nodes in the region through the communication network within the region. The above data is input into the distributed optimizer running inside it. The optimizer iteratively solves the problem. In each iteration, each node updates its own policy based on the policies of its neighboring nodes and the Lagrange multiplier until the policies of all nodes converge to a stable state that satisfies the constraints. The policy set of all nodes corresponding to this stable state is the joint optimization policy combination for the dense region.

[0119] Step S440: The combined optimization strategy for the dense region is distributed to each unmanned node in the densely connected topology region, replacing the original optimization strategy selection result of each unmanned node in the densely connected topology region.

[0120] The dense area collaborative controller module distributes the corresponding part of the dense area joint optimization strategy combination generated in step S430 to each unmanned node in the area through intra-area communication. After receiving the new strategy, each node replaces its original optimization strategy selection result with the new strategy.

[0121] Step S450: Assign an independent decision-making module for the sparse region in the topological connection sparse region to the unmanned node in the topological connection sparse region. The independent decision-making module for the sparse region is used to perform independent policy decision-making processing on the unmanned node in the topological connection sparse region.

[0122] For each unmanned node marked as a sparse region of topological connection, a sparse region independent decision-making module is instantiated on its own onboard computing unit. This is a simplified decision-maker that does not run complex distributed coordination algorithms, but makes decisions based solely on its own perceived local information and preset heuristic rules.

[0123] Step S460: Obtain the candidate strategy set and benefit evaluation value of each unmanned node in the topology-connected sparse region through the sparse region independent decision module, perform independent optimization processing on the candidate strategy set of each unmanned node in the topology-connected sparse region, and generate the sparse region independent optimization strategy selection result.

[0124] The sparse region independent decision module obtains the candidate strategy set and benefit evaluation value of the node itself, executes a greedy search algorithm to traverse all candidate strategies in the candidate strategy set, and directly selects the strategy with the highest benefit evaluation value as the sparse region independent optimization strategy selection result of the node.

[0125] Step S470: The sparse region independent optimization strategy selection result is sent to each unmanned node in the topologically connected sparse region, replacing the original optimization strategy selection result of each unmanned node in the topologically connected sparse region. Based on the replaced dense region joint optimization strategy combination and sparse region independent optimization strategy selection result, a new hierarchical game control instruction is generated to obtain a regionally differentiated hierarchical game control instruction. The regionally differentiated hierarchical game control instruction is received through the lower-level action execution domain, and regionally differentiated real-time motion control parameters for each unmanned node are generated based on the regionally differentiated hierarchical game control instruction and the topological connection relationship set. The regionally differentiated real-time motion control parameters are sent to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform regionally differentiated cooperative motion.

[0126] The sparse region independent decision-making module sends the generated sparse region independent optimization strategy selection result to the node itself to replace the original optimization strategy selection result. It merges the strategy after replacement in the dense region in step S440 with the strategy after replacement in the sparse region in step S460 to form a regionally differentiated strategy set covering the entire cluster. Based on this strategy set, it re-executes step S147 to generate regionally differentiated hierarchical game control instructions. The lower-level action execution domain receives the instructions and executes steps S151 to S157 to generate the corresponding regionally differentiated real-time motion control parameters and sends them to each unmanned node.

[0127] Step S510: Based on the node attribute category and the node motion parameters, obtain the remaining energy parameter and task completion progress parameter of each unmanned node in the heterogeneous unmanned cluster. The remaining energy parameter is used to characterize the energy reserve value currently available to the unmanned node, and the task completion progress parameter is used to characterize the proportion of the task amount currently completed by the unmanned node to the total number of allocated tasks.

[0128] The remaining energy parameters of each unmanned node are read in real time through the battery management system interface, which reflect the current available energy reserves in milliampere-hours or joules. At the same time, the list of subtasks assigned to the node and the status of the completed subtasks are obtained from the global task management module, and the task completion progress parameter is obtained by calculating the ratio of the number of completed subtasks to the total number of assigned subtasks.

[0129] Step S520: Based on the remaining energy parameter and the task completion progress parameter, dynamically adjust the energy consumption cost utility value in the initial strategy benefit function of each unmanned node to obtain the dynamically adjusted strategy benefit function. In the dynamically adjusted strategy benefit function, unmanned nodes with lower remaining energy parameters have higher weights in the energy consumption cost utility value.

[0130] For each unmanned node, its remaining energy parameter is compared with the preset energy warning threshold. If the remaining energy parameter is lower than the threshold, the weight of the energy consumption cost utility value in the initial strategy benefit function in step S141 is adjusted. The weight of the energy consumption cost utility value is increased and the weight of the task completion benefit utility value is reduced accordingly. The adjusted weight is substituted into the initial strategy benefit function formula in step S141 to generate the dynamically adjusted strategy benefit function.

[0131] Step S530: Input the dynamically adjusted strategy payoff function into the strategy generator module of the upper-level strategy game domain. The strategy generator module performs energy-aware payoff evaluation on each candidate strategy in the candidate strategy set of each unmanned node to obtain an energy-aware payoff evaluation value. Based on the energy-aware payoff evaluation value, the candidate strategies in the candidate strategy set of each unmanned node are reordered to obtain an energy-aware optimized strategy selection result. Based on the energy-aware optimized strategy selection result, an energy-aware hierarchical game control instruction is generated.

[0132] The dynamically adjusted strategy payoff function obtained in step S520 is input into the strategy generator module. The strategy generator module uses this payoff function to replace the original payoff function and re-executes the payoff evaluation process in step S145 to calculate the energy-aware payoff evaluation value for each candidate strategy. Then, the strategy selection process in step S146 is executed, but the new payoff evaluation values ​​are used for sorting and selection to obtain the energy-aware optimized strategy selection result. Finally, step S147 is executed to generate the energy-aware hierarchical game control instruction based on the result.

[0133] Step S540: Receive the energy-sensing hierarchical game control command through the lower-level action execution domain, generate energy-sensing real-time motion control parameters for each unmanned node based on the energy-sensing hierarchical game control command and the set of topological connections, and send the energy-sensing real-time motion control parameters to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform energy-equalizing cooperative motion.

[0134] The lower-level action execution domain receives the energy-sensing hierarchical game control command and executes steps S151 to S157 to generate the corresponding energy-sensing real-time motion control parameters and sends them to each unmanned node.

[0135] Step S550: Monitor the changes in the remaining energy parameters of each unmanned node during the execution of the energy equalization cooperative motion, obtain the sequence of remaining energy parameter changes, identify energy-limited unmanned nodes whose remaining energy parameters are lower than the third threshold parameter based on the sequence of remaining energy parameter changes, and generate a list of energy-limited node identifiers.

[0136] During the coordinated motion execution driven by step S540, the remaining energy parameter of each unmanned node is continuously monitored at high frequency and its time-varying sequence is recorded. When the remaining energy parameter value of a node drops below the preset third threshold parameter, the unique identifier of that node is added to the list of energy-limited nodes.

[0137] Step S560: Based on the list of energy-constrained node identifiers, adjust the task allocation scheme of the energy-constrained unmanned nodes, redistribute some tasks of the energy-constrained unmanned nodes to other unmanned nodes whose remaining energy parameters are higher than the fourth threshold parameter, generate task redistribution results, and regenerate energy-sensing hierarchical game control instructions based on the task redistribution results to drive the heterogeneous unmanned cluster to perform cooperative movement after task redistribution.

[0138] Based on the list of energy-constrained node identifiers generated in step S550, the global task management module is triggered to perform task reallocation. For each energy-constrained node in the list, its currently unfinished tasks are evaluated and tasks with high energy consumption are identified. From the nodes with remaining energy parameters higher than the fourth threshold parameter, the node closest to the task location and with matching capabilities is selected, and the task is transferred to that node. The task reallocation result is generated. Based on this mapping table, the task allocation scheme of the relevant nodes is updated, and steps S530 and S540 are re-executed to generate new energy-aware hierarchical game control instructions.

[0139] For example, the method may further include: step S610: obtaining global task description information of the heterogeneous unmanned cluster to be executed, wherein the global task description information includes task type identifiers of multiple subtasks, task spatial location coordinates and task execution timing constraint parameters.

[0140] Before the collaborative search and rescue mission begins, the central mission scheduling unit broadcasts global mission description information to the heterogeneous unmanned swarm. This global mission description information is a structured data set containing detailed definitions of multiple subtasks. Each subtask is described by a mission type identifier, mission spatial coordinates, and mission execution timing constraints. The mission type identifier is an enumerated value, the mission spatial coordinates are three-dimensional coordinate vectors, and the mission execution timing constraints include the earliest start timestamp and the latest completion timestamp.

[0141] Step S620: Based on the node attribute category and the node motion parameters, perform initial task allocation processing on multiple subtasks in the global task description information to obtain an initial task allocation result. The initial task allocation result includes a list of subtask identifiers assigned to each unmanned node.

[0142] The central task scheduling unit executes a greedy allocation algorithm based on distance and capability according to the node attribute category and current node motion parameters of each unmanned node. It traverses all subtasks in the global task description information. For each subtask, it selects a set of nodes with the capability to execute the task based on its task type identifier. In the selected set of nodes, it calculates the Euclidean distance from the current node spatial coordinates to the task spatial coordinates of each node. It selects the node with the closest distance as the execution node of the subtask. After all subtasks are allocated, the initial task allocation result is generated.

[0143] Step S630: Based on the neighbor node identifier list of each unmanned node in the topology connection relationship set, identify unmanned node pairs with direct communication links, generate a direct communication node pair list, and perform task allocation negotiation processing for each unmanned node pair in the direct communication node pair list. The task allocation negotiation processing includes exchanging their respective sub-task identifier lists and remaining capability parameters, and adjusting the sub-task identifier lists according to the exchange results to generate negotiated task allocation results.

[0144] Based on the topology connection set generated in step S1210, all links are established with edges marked with a value of 1. Each edge connects two unmanned nodes to form a direct communication node pair, generating a list of direct communication node pairs. For each unmanned node pair in this list, task allocation negotiation is triggered. The two nodes exchange their current subtask identifier lists and remaining capacity parameters via the direct communication link. After exchanging information, each node independently checks if there are any subtasks in the other node's subtasks that are more efficient than its own. If so, it sends a task transfer request to the other node. Upon receiving the request, the other node evaluates its current load and task importance. If it agrees to the transfer, it updates its own subtask identifier list and sends a confirmation message to the requester. Upon receiving the confirmation, the requester also updates its own subtask identifier list. After all direct communication node pairs complete one round of negotiation, a negotiated task allocation result is generated.

[0145] Step S640: For unmanned nodes that do not have direct communication links, obtain the subtask identifier list and remaining capability parameters of other unmanned nodes through the multi-hop communication paths in the topology connection relationship set, perform multi-hop task allocation negotiation processing, and generate the task allocation result after multi-hop negotiation.

[0146] For unmanned node pairs without direct communication links, negotiation is conducted via multi-hop communication paths. Each node calculates the shortest path to all other nodes using a distributed routing protocol based on the topology connection set generated in step S1210. It then sends a request message to the target node via this shortest path to obtain its subtask identifier list and remaining capacity parameters. After collecting information from all nodes, each node independently executes a market-based negotiation algorithm, treating each subtask as a commodity. Each node bids based on its remaining capacity parameters and the execution cost of the subtask. All nodes broadcast their bids via multi-hop paths. Ultimately, each subtask is assigned to the node with the highest bid. All nodes update their subtask identifier lists based on this global negotiation result, generating the multi-hop negotiated task allocation result.

[0147] Step S650: Merge the negotiated task allocation result with the multi-hop negotiated task allocation result to obtain the global negotiated task allocation result. Update the task allocation scheme of each unmanned node in the upper-level strategy game domain according to the global negotiated task allocation result to generate the updated task allocation scheme.

[0148] The negotiated task allocation result generated in step S630 is merged with the multi-hop negotiated task allocation result generated in step S640. If the same subtask is assigned to different nodes in the two results, the multi-hop negotiated task allocation result shall prevail. The merged result is the global negotiated task allocation result. This result is input into the upper-level strategy game domain to replace the original task allocation scheme in the optimization strategy selection result of each unmanned node in step S146 to generate an updated task allocation scheme.

[0149] Step S660: Regenerate the hierarchical game control instructions according to the updated task allocation scheme to obtain the hierarchical game control instructions after task negotiation. Receive the hierarchical game control instructions after task negotiation through the lower-level action execution domain, and generate real-time motion control parameters for each unmanned node after task negotiation based on the hierarchical game control instructions after task negotiation and the set of topological connections.

[0150] Based on the updated task allocation scheme generated in step S650, the original path planning scheme is retained, and step S147 is re-executed to generate the hierarchical game control instruction after task negotiation. The lower-level action execution domain receives the instruction and executes steps S151 to S157 to generate the corresponding real-time motion control parameters after task negotiation.

[0151] Step S670: Send the real-time motion control parameters after task negotiation to the corresponding unmanned nodes, drive the heterogeneous unmanned cluster to execute the coordinated motion after task negotiation, monitor the task completion progress parameters of each unmanned node during the execution of the coordinated motion after task negotiation, obtain the task completion progress parameter sequence, identify the lagging unmanned nodes whose task completion progress parameters are lower than the fifth threshold parameter according to the task completion progress parameter sequence, generate a list of lagging unmanned node identifiers, and allocate additional task execution auxiliary resources to the lagging unmanned nodes in the list of lagging unmanned node identifiers.

[0152] The real-time motion control parameters generated in step S660 after task negotiation are sent to each unmanned node to drive the cluster to execute the coordinated motion after task negotiation. During the motion execution, the task completion progress parameter of each unmanned node is continuously monitored and its sequence of changes over time is recorded. When the growth rate of the task completion progress parameter of a node is lower than a preset rate threshold or its absolute value is lower than a preset fifth threshold parameter within a certain period of time, the node is identified as a lagging unmanned node and its unique identifier is added to the lagging unmanned node identification list. For the lagging unmanned nodes in the list, additional task execution auxiliary resources are allocated to them through the central task scheduling unit.

[0153] For example, the method may further include: step S710: collecting a snapshot of the real-time node motion parameter sequence and the real-time topological connection relationship set of each unmanned node in the heterogeneous unmanned cluster during the execution of cooperative motion; inputting the real-time node motion parameter sequence and the real-time topological connection relationship set snapshot into a pre-constructed spatiotemporal graph neural network model; the spatiotemporal graph neural network model includes a graph convolutional encoder module and a temporal convolutional encoder module; the graph convolutional encoder module is used to perform spatial feature extraction processing on the spatial topological structure between each unmanned node and its neighboring nodes in the real-time topological connection relationship set snapshot to generate a spatial topological embedding vector corresponding to each unmanned node; the temporal convolutional encoder module is used to perform temporal dimension feature extraction processing on the real-time node motion parameter sequence to generate a temporal motion embedding vector corresponding to each unmanned node.

[0154] During the cooperative motion of a heterogeneous unmanned swarm, real-time node motion parameter sequences and real-time topological connection set snapshots of each unmanned node are continuously collected. The collected data is input into a pre-built and trained spatiotemporal graph neural network model. This model includes a graph convolutional encoder module and a temporal convolutional encoder module. The graph convolutional encoder module receives the real-time topological connection set snapshot as input, represented as an adjacency matrix. The module contains multiple graph convolutional layers. Each layer updates the node feature representation by aggregating the feature information of each node and its neighboring nodes. After multiple layers are stacked, the output is a spatial topological embedding vector corresponding to each unmanned node. This vector encodes the node's local structural information in the topological graph and the feature aggregation results of its neighboring nodes. The temporal convolutional encoder module receives real-time node motion parameter sequences as input, which consist of continuous multi-frame motion parameter data within a time window. The temporal convolutional encoder module contains multiple dilated convolutional layers. Each dilated convolutional layer extracts temporal dependency features at different time scales through convolutional kernels with different dilation rates. After multiple layers are stacked, the temporal motion embedding vector corresponding to each unmanned node is output. This temporal motion embedding vector encodes the change pattern of the node's motion state in the time dimension.

[0155] Step S720: Input the spatial topology embedding vector and the temporal motion embedding vector into the feature fusion module of the spatiotemporal graph neural network model. The feature fusion module performs weighted fusion processing on the spatial topology embedding vector and the temporal motion embedding vector through a cross-modal attention mechanism to generate a spatiotemporal fusion representation vector corresponding to each unmanned node. The cross-modal attention mechanism includes the interactive calculation process of the spatial feature query matrix and the temporal feature key value matrix.

[0156] The feature fusion module of the spatiotemporal graph neural network model receives the spatial topology embedding vector and temporal motion embedding vector generated in step S710 as input. Internally, the feature fusion module implements a cross-modal attention mechanism. First, it constructs a spatial feature query matrix and a temporal feature key-value matrix. The spatial feature query matrix is ​​obtained by linearly transforming the spatial topology embedding vector, and the temporal feature key-value matrix is ​​obtained by linearly transforming the temporal motion embedding vector to obtain the key matrix and value matrix. Then, it calculates the attention weight matrix between the spatial feature query matrix and the temporal feature key matrix, where each element represents the correlation strength between spatial and temporal features. Finally, it multiplies the attention weight matrix with the temporal feature value matrix to obtain the weighted temporal features. The weighted temporal features are then concatenated with the original spatial topology embedding vector along the feature dimension to generate the spatiotemporal fusion representation vector corresponding to each unmanned node.

[0157] Step S730: Input the spatiotemporal fusion representation vector corresponding to each unmanned node into the game strategy prediction head module of the spatiotemporal graph neural network model. The game strategy prediction head module includes a multilayer perceptron structure. The multilayer perceptron structure is used to map the spatiotemporal fusion representation vector to the strategy preference distribution vector of each unmanned node. The strategy preference distribution vector includes preference probability values ​​corresponding to multiple strategy selection dimensions.

[0158] The game strategy prediction head module of the spatiotemporal graph neural network model receives the spatiotemporal fusion representation vector generated in step S720 as input. The game strategy prediction head module consists of a multilayer perceptron structure, which includes an input layer, multiple hidden layers, and an output layer. The input layer receives the values ​​of each dimension of the spatiotemporal fusion representation vector. After nonlinear transformation and feature extraction by multiple hidden layers, the output layer maps the network output to a policy preference distribution vector through a Softmax activation function. The dimension of this policy preference distribution vector is the same as the number of policies in the candidate policy set, and the value in each dimension represents the probability of the node selecting the corresponding candidate policy.

[0159] Step S740: Perform policy filtering on the candidate policy set of each unmanned node according to the policy preference distribution vector to obtain the policy filtering result. The policy filtering result includes the priority sequence of candidate policies for each unmanned node after sorting according to the preference probability value.

[0160] For each unmanned node, the preference probability values ​​in the policy preference distribution vector generated in step S730 are sorted from high to low. Based on the sorting results, the candidate policies in the candidate policy set are reordered to generate a candidate policy priority sequence for each unmanned node as the policy selection result.

[0161] Step S750: Input the strategy selection results into the upper-level strategy game domain, replace the original candidate strategy set to participate in the strategy payoff function evaluation and game interaction rule execution process, and generate hierarchical game control instructions based on the spatiotemporal graph neural network model.

[0162] The strategy selection results generated in step S740 are input into the upper-level strategy game domain, replacing the original candidate strategy set. The upper-level strategy game domain uses the updated candidate strategy priority sequence to execute the strategy payoff function evaluation and game interaction rule execution process from steps S145 to S147, generating hierarchical game control instructions based on the spatiotemporal graph neural network model.

[0163] Step S760: Receive the hierarchical game control command enhanced by the spatiotemporal graph neural network model through the lower-level action execution domain, and generate enhanced real-time motion control parameters for each unmanned node according to the hierarchical game control command enhanced by the spatiotemporal graph neural network model and the set of topological connections.

[0164] After receiving the hierarchical game control command based on the spatiotemporal graph neural network model generated in step S750, the lower-level action execution domain executes steps S151 to S157 based on the command and the latest set of topological connections to generate enhanced real-time motion control parameters for each unmanned node.

[0165] Step S770: The enhanced real-time motion control parameters are sent to the corresponding unmanned nodes to drive the heterogeneous unmanned cluster to perform cooperative motion. At the same time, a snapshot of the new real-time node motion parameter sequence and the new real-time topology connection relationship set generated during the execution is collected as incremental training sample data for the spatiotemporal graph neural network model. The spatiotemporal graph neural network model is subjected to online incremental update processing based on the incremental training sample data. The online incremental update processing includes calculating the deviation loss value between the predicted policy preference distribution vector corresponding to the incremental training sample data and the actual execution policy selection result, and updating the network weight parameters of the spatiotemporal graph neural network model by backpropagation based on the deviation loss value. The updated spatiotemporal graph neural network model is deployed to the key unmanned nodes in the heterogeneous unmanned cluster for policy selection processing in the next round of cooperative motion.

[0166] The enhanced real-time motion control parameters generated in step S760 are sent to the corresponding unmanned nodes to drive the heterogeneous unmanned swarm to perform cooperative motion. During execution, snapshots of new real-time node motion parameter sequences and new real-time topology connection sets are simultaneously collected, and these data are used as incremental training samples. For each incremental training sample, the spatiotemporal graph neural network model calculates the predicted policy preference distribution vector, and simultaneously obtains the actual policy selection result of the node from the actual execution record as the label, calculating the cross-entropy loss value between the predicted preference distribution and the actual selection result as the bias loss value. Based on this bias loss value, the gradient of each layer parameter of the spatiotemporal graph neural network model is calculated using the backpropagation algorithm, and the network weight parameters are updated using the stochastic gradient descent optimizer. The updated spatiotemporal graph neural network model is deployed to the key unmanned nodes in the heterogeneous unmanned swarm for policy selection processing in the next round of cooperative motion.

[0167] For example, the method may further include: step S810: obtaining the historical motion trajectory database of each unmanned node in the heterogeneous unmanned cluster, wherein the historical motion trajectory database contains a set of trajectory point sequences and corresponding topological connection relationship sets recorded by each unmanned node in the past multiple task cycles, and historical snapshots.

[0168] Historical data is extracted from the central storage unit of the cluster or the local storage of each node to construct a historical motion trajectory database. This database contains a set of trajectory point sequences recorded by each unmanned node over multiple past task cycles. Each trajectory point sequence consists of the spatial location coordinates of the node under consecutive timestamps, and also includes a set of historical snapshots of topological connections aligned with the time of each trajectory point sequence.

[0169] Step S820: Construct an adversarial game prediction network structure, which includes a trajectory generator network module, a topology discriminator network module, and a cooperative evaluator network module. The trajectory generator network module is used to generate a sequence of predicted trajectory points based on the input initial node motion parameters and the initial set of topological connections. The topology discriminator network module is used to distinguish between the set of topological connections corresponding to the generated sequence of predicted trajectory points and the historical snapshot of the actual set of topological connections. The cooperative evaluator network module is used to evaluate the cooperative motion efficiency index corresponding to the generated sequence of predicted trajectory points.

[0170] An adversarial game prediction network structure is constructed, consisting of three sub-network modules. The trajectory generator network module employs an encoder-decoder architecture. The encoder uses a long short-term memory (LSTM) network to encode the initial sequence of node motion parameters and the initial set of topological connections. The decoder uses another LSM network to progressively generate the predicted trajectory point sequence. The topology discriminator network module uses a graph convolutional network structure. Its input is either the set of topological connections corresponding to the predicted trajectory point sequence generated by the trajectory generator or a snapshot of the actual historical topological connection set. The output is a discrimination result indicating whether the input is real data or generated data. The cooperative evaluator network module uses a multilayer perceptron structure. Its input is the predicted trajectory point sequence generated by the trajectory generator, and its output is a cooperative motion efficiency score.

[0171] Step S830: Using the historical snapshots of the trajectory point sequence set and the corresponding topological connection relationship set in the historical motion trajectory database as training data, adversarial training is performed on the adversarial game prediction network structure. The adversarial training process includes a first adversarial training stage of training the topology discriminator network module with fixed trajectory generator network module parameters and a second adversarial training stage of training the trajectory generator network module with fixed topology discriminator network module parameters. The first adversarial training stage is used to maximize the discrimination accuracy of the topology discriminator network module in distinguishing between generated data and real data. The second adversarial training stage is used to minimize the discrimination probability that the data generated by the trajectory generator network module is identified as generated data by the topology discriminator network module.

[0172] Using historical snapshots of trajectory point sequences and corresponding topological connectivity sets from a historical motion trajectory database as training data, adversarial training is performed on the adversarial game prediction network structure. In the first adversarial training phase, the parameters of the trajectory generator network module are fixed, and historical real data and predicted data generated by the trajectory generator network module are input into the topology discriminator network module. The discriminant loss is calculated, and the parameters of the topology discriminator network module are updated. The goal is to maximize the accuracy of the topology discriminator network module in distinguishing between real and generated data. In the second adversarial training phase, the parameters of the topology discriminator network module are fixed, and only the parameters of the trajectory generator network module are updated. The goal is to maximize the probability that the data generated by the trajectory generator network module is misclassified as real data by the topology discriminator network module. The first and second adversarial training phases are alternately executed until the network converges.

[0173] Step S840: After completing the adversarial training process, obtain the current node motion parameters and current topology connection set of each unmanned node in the current heterogeneous unmanned cluster. Input the current node motion parameters and current topology connection set into the trained trajectory generator network module to generate multiple candidate predicted trajectory sequences for each unmanned node. Input the multiple candidate predicted trajectory sequences into the trained cooperative evaluator network module. The cooperative evaluator network module performs cooperative motion efficiency index evaluation processing on each candidate predicted trajectory sequence to generate a cooperative efficiency score value corresponding to each candidate predicted trajectory sequence.

[0174] After adversarial training, the set of current node motion parameters and current topological connections for each unmanned node at the current moment is obtained and input into the trained trajectory generator network module. The trajectory generator network module, through its internal encoder-decoder structure and random noise vectors, generates multiple different candidate predicted trajectory sequences. Each generated candidate predicted trajectory sequence is input into the trained cooperative evaluator network module. The cooperative evaluator network module, through its multilayer perceptron structure, evaluates the cooperative motion efficiency index of each trajectory sequence and outputs a scalar value as the cooperative efficiency score for that trajectory.

[0175] Step S850: Sort and filter the multiple candidate predicted trajectory sequences according to the collaborative efficiency score value, select the candidate predicted trajectory sequence with the highest collaborative efficiency score value as the optimal predicted trajectory sequence, and input the optimal predicted trajectory sequence into the upper-level strategy game domain as the predicted benefit utility value in the strategy benefit function of each unmanned node to participate in the strategy benefit function evaluation and game interaction rule execution process. The predicted benefit utility value is used to measure the expected collaborative benefit utility value that the unmanned node can obtain when it executes according to the optimal predicted trajectory sequence.

[0176] The collaborative efficiency scores corresponding to each candidate predicted trajectory sequence generated in step S840 are sorted from high to low, and the candidate predicted trajectory sequence with the highest collaborative efficiency score is selected as the optimal predicted trajectory sequence. This optimal predicted trajectory sequence is input into the upper-level strategy game domain. When constructing the initial strategy payoff function in step S141, a predicted payoff utility term is added. This predicted payoff utility value is calculated by substituting the optimal predicted trajectory sequence into a preset payoff evaluation function and is used to measure the expected collaborative payoff that a node can obtain when executing according to the predicted trajectory.

[0177] Step S860: Generate a prediction-enhanced hierarchical game control instruction based on the strategy payoff function containing the predicted payoff utility value, receive the prediction-enhanced hierarchical game control instruction through the lower-level action execution domain, generate prediction-enhanced real-time motion control parameters for each unmanned node based on the prediction-enhanced hierarchical game control instruction and the set of topological connections, send the prediction-enhanced real-time motion control parameters to the corresponding unmanned node, and drive the heterogeneous unmanned cluster to perform prediction-optimized cooperative motion.

[0178] The upper-level strategy game domain uses a strategy payoff function that includes predicted payoff utility values ​​to re-execute steps S145 to S147 to generate prediction-enhanced hierarchical game control instructions. The lower-level action execution domain receives these instructions and executes steps S151 to S157 to generate prediction-enhanced real-time motion control parameters for each unmanned node, which are then sent to the corresponding unmanned nodes to drive the cluster to execute prediction-optimized cooperative motion.

[0179] Based on the same inventive concept, please refer to Figure 2 This diagram illustrates a schematic block diagram of a heterogeneous unmanned swarm control system combining dynamic topology and hierarchical game theory, as provided in an embodiment of this application. It includes a central processing unit (CPU), a system memory comprising random access memory (RAM) and read-only memory (ROM), and a system bus connecting the system memory and the CPU. The heterogeneous unmanned swarm control system also includes a basic input / output system to facilitate information transmission between various devices within the computer, and a large-capacity storage device for storing the operating system, applications, and other program modules.

[0180] A basic input / output system includes a display for showing information and input devices such as a mouse and keyboard for user input. Both the display and the input devices are connected to the central processing unit via an input / output controller connected to the system bus. The basic input / output system may also include an input / output controller for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller also provides output to a display screen, printer, or other types of output devices.

[0181] Mass storage devices are connected to the central processing unit via a mass storage controller connected to the system bus. The mass storage devices and their associated computer-readable media provide non-volatile storage for heterogeneous unmanned swarm control combining dynamic topology and hierarchical game theory. Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. According to various embodiments of this application, heterogeneous unmanned swarm control combining dynamic topology and hierarchical game theory can also be connected to a remote computer on a network, such as the Internet. That is, heterogeneous unmanned swarm control combining dynamic topology and hierarchical game theory can be connected to a network via a network interface unit connected to the system bus, or a network interface unit can be used to connect to other types of networks or remote computer systems.

[0182] In addition, in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0183] The above are merely exemplary embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory, characterized in that, The method includes: Obtain the node attribute category and node motion parameters of each unmanned node in the heterogeneous unmanned cluster. The node attribute category is used to distinguish the functional type of the unmanned node. The node motion parameters include the node spatial position coordinates, the node velocity vector, and the node acceleration vector. A set of topological connections for a heterogeneous unmanned cluster is constructed based on the node attribute categories and the node motion parameters. The set of topological connections includes the existence identifier of communication links between each unmanned node and other unmanned nodes within the detection area, as well as the relative motion constraint boundary parameters. Based on the aforementioned set of topological connections, a hierarchical game control structure is constructed. The hierarchical game control structure includes an upper-level strategy game domain and a lower-level action execution domain. The upper-level strategy game domain is used for global task decomposition and collaborative path orchestration of heterogeneous unmanned clusters. The lower-level action execution domain is used for local motion control and obstacle avoidance response processing of each unmanned node. The strategy payoff function and game interaction rules are assigned to each unmanned node through the upper-level strategy game domain, and the strategy payoff function of each unmanned node is updated according to the changes in the set of topological connections, generating a hierarchical game control instruction containing the optimized strategy selection results of each unmanned node. The lower-level action execution domain receives the hierarchical game control command, generates real-time motion control parameters for each unmanned node based on the hierarchical game control command and the set of topological connections, and sends the real-time motion control parameters to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion.

2. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The step of constructing a set of topological connections for a heterogeneous unmanned cluster based on the node attribute categories and the node motion parameters includes: Based on the node attribute categories, all unmanned nodes in the heterogeneous unmanned cluster are classified according to their attributes, resulting in a first attribute category node affiliation set and a second attribute category node affiliation set. The unmanned nodes in the first attribute category node affiliation set have a first set of motion capability parameters and a first set of perception capability parameters. The unmanned nodes in the second attribute category node affiliation set have a second set of motion capability parameters and a second set of perception capability parameters. The first set of motion capability parameters includes a first maximum velocity amplitude parameter and a first maximum acceleration amplitude parameter. The second set of motion capability parameters includes a second maximum velocity amplitude parameter and a second maximum acceleration amplitude parameter. The first set of perception capability parameters includes a first detection distance amplitude parameter and a first detection angle range parameter. The second set of perception capability parameters includes a second detection distance amplitude parameter and a second detection angle range parameter. For each unmanned node in the first attribute category node belonging set, traverse other unmanned nodes within its detection area, and obtain the first spatial interval distance parameter and the first relative velocity difference parameter between each unmanned node and other unmanned nodes within its detection area. The first spatial interval distance parameter is used to characterize the straight-line distance between two unmanned nodes in the spatial position coordinates, and the first relative velocity difference parameter is used to characterize the magnitude and direction of the difference vector between the two unmanned nodes in the velocity vector. Based on the comparison between the first spatial interval distance parameter and the first detection distance amplitude parameter, the existence identifier of the first communication link between each unmanned node in the first attribute category node affiliation set and other unmanned nodes is determined. The existence identifier of the first communication link includes the first link establishment identifier and the first link maintenance duration parameter. The first link maintenance duration parameter is used to characterize the cumulative length of time from the establishment of the communication link to the current moment. Based on the direction of the difference vector in the first relative velocity difference parameter and the first attribute category node affiliation set, the first relative motion constraint boundary parameter between each unmanned node in the first attribute category node affiliation set and other unmanned nodes is determined. The first relative motion constraint boundary parameter includes a first relative motion direction angle range parameter and a first relative motion velocity ratio range parameter. The first relative motion direction angle range parameter is used to limit the acceptable range of the motion direction angle between two unmanned nodes, and the first relative motion velocity ratio range parameter is used to limit the acceptable range of the ratio of the velocity magnitudes between two unmanned nodes. For each unmanned node in the set to which the second attribute category node belongs, traverse the other unmanned nodes within its detection area, and obtain the second spatial interval distance parameter and the second relative velocity difference parameter between each unmanned node and the other unmanned nodes within its detection area. The second spatial interval distance parameter is used to characterize the straight-line distance between two unmanned nodes in the spatial position coordinates, and the second relative velocity difference parameter is used to characterize the magnitude and direction of the difference vector between the two unmanned nodes in the velocity vector. Based on the comparison between the second spatial interval distance parameter and the second detection distance amplitude parameter, the existence identifier of the second communication link between each unmanned node in the second attribute category node affiliation set and other unmanned nodes is determined. The existence identifier of the second communication link includes the second link establishment identifier and the second link maintenance duration parameter. The second link maintenance duration parameter is used to characterize the cumulative length of time from the establishment of the communication link to the current moment. Based on the direction of the difference vector in the second relative velocity difference parameter and the second attribute category node affiliation set, the second relative motion constraint boundary parameter between each unmanned node in the second attribute category node affiliation set and other unmanned nodes is determined. The second relative motion constraint boundary parameter includes a second relative motion direction angle range parameter and a second relative motion velocity ratio range parameter. The second relative motion direction angle range parameter is used to limit the acceptable range of the motion direction angle between two unmanned nodes, and the second relative motion velocity ratio range parameter is used to limit the acceptable range of the ratio of the velocity magnitudes between two unmanned nodes. Based on the existence identifier of the first communication link and the boundary parameter of the first relative motion constraint, a first association mapping process is performed on each unmanned node in the first attribute category node belonging set and other unmanned nodes in its detection area to generate a first local topological connection relationship subset corresponding to the first attribute category node. Based on the existence identifier of the second communication link and the boundary parameter of the second relative motion constraint, a second association mapping process is performed on each unmanned node in the set to which the second attribute category node belongs and other unmanned nodes in its detection area to generate a second local topology connection relationship subset corresponding to the second attribute category node. The first local topology connection subset and the second local topology connection subset are merged to generate a topology connection set of heterogeneous unmanned cluster. The topology connection set includes a list of neighbor node identifiers for each unmanned node and a set of edge attribute parameters between each unmanned node and its neighbor nodes. The set of edge attribute parameters includes the existence identifier of the communication link and the relative motion constraint boundary parameters.

3. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The construction of a hierarchical game control structure based on the set of topological connections includes: Based on the set of topological connections, extract the list of neighbor node identifiers of all unmanned nodes in the heterogeneous unmanned cluster, and obtain the relative motion constraint boundary parameters in the set of edge attribute parameters between each unmanned node and its neighbor nodes. The relative motion constraint boundary parameters include the range parameters of the relative motion direction angle and the range parameters of the ratio of the relative motion speed. Based on the node attribute categories and the relative motion constraint boundary parameters, the unmanned nodes in the heterogeneous unmanned cluster are divided into game hierarchy processes to obtain the first game node set corresponding to the upper-level strategy game domain and the second game node set corresponding to the lower-level action execution domain. The unmanned nodes in the first game node set have global information interaction capabilities, and the unmanned nodes in the second game node set have local perception and action execution capabilities. Each unmanned node in the first set of game nodes is assigned a strategy generator module in the upper-level strategy game domain. The strategy generator module is used to generate a candidate strategy set for each unmanned node based on the global topology information in the set of topology connections. The candidate strategy set includes multiple candidate path planning schemes and multiple candidate task allocation schemes. Each unmanned node in the second set of game nodes is assigned a motion controller module in the lower action execution domain. The motion controller module is used to generate real-time motion control parameters for each unmanned node based on the strategy selection result output by the upper strategy game domain. The real-time motion control parameters include velocity vector adjustment parameters and acceleration vector adjustment parameters. An information transmission channel is established between the strategy generator module in the upper-level strategy game domain and the motion controller module in the lower-level action execution domain. The information transmission channel is used to transmit the strategy selection results generated in the upper-level strategy game domain to the lower-level action execution domain. The strategy selection results include the path planning scheme identifier and task allocation scheme identifier selected by each unmanned node. Based on the frequency of change of the neighbor node identifier list in the topological connection set, an update frequency parameter of the topological connection set is generated, which is used to indicate the degree of drastic change of the topological connection set in the time dimension. Based on the transmission delay parameter of the information transmission channel and the update frequency parameter of the topology connection set, a collaborative time window parameter is set between the upper-level strategy game domain and the lower-level action execution domain. The collaborative time window parameter is used to synchronize the strategy update time of the upper-level strategy game domain and the control command issuance time of the lower-level action execution domain. The hierarchical interaction density parameter of the hierarchical game control structure is generated based on the number of unmanned nodes in the first game node set and the number of unmanned nodes in the second game node set. The hierarchical interaction density parameter is used to indicate the information interaction frequency between the upper-level strategy game domain and the lower-level action execution domain. Based on the changing trend of the set of edge attribute parameters in the set of topological connections, the numerical range of the collaborative time window parameter is dynamically adjusted to generate a dynamically adjusted collaborative time window parameter. Based on the dynamically adjusted collaborative time window parameter and the hierarchical interaction density parameter, a time synchronization mechanism between the upper-level strategy game domain and the lower-level action execution domain is constructed. The time synchronization mechanism is used to ensure the alignment of strategy update operations and control command issuance operations in the time dimension.

4. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The process of assigning a strategy payoff function and game interaction rules to each unmanned node through the upper-level strategy game domain, updating the strategy payoff function of each unmanned node according to changes in the set of topological connections, and generating a hierarchical game control instruction containing the optimized strategy selection results of each unmanned node includes: An initial strategy benefit function is assigned to each unmanned node in a heterogeneous unmanned cluster. The initial strategy benefit function is constructed through the following steps: A task completion progress metric is determined based on the node attribute category and the current task allocation result of the unmanned node; the task completion progress metric is then converted into a dimensionless task completion benefit utility value using a preset first normalization function; an energy consumption metric is determined based on the velocity and acceleration vectors in the node motion parameters of the unmanned node; the energy consumption metric is then converted into a dimensionless energy consumption cost utility value using a preset second normalization function; the task completion benefit utility value and the energy consumption cost utility value are added together with preset weights to obtain the initial strategy benefit function, which is a dimensionless comprehensive utility value. Based on the neighbor node identifier list of each unmanned node in the topological connection relationship set, the relative motion constraint boundary parameters in the edge attribute parameter set between each unmanned node and its neighbor nodes are obtained. The initial strategy benefit function of each unmanned node is then modified based on these relative motion constraint boundary parameters to obtain a first modified strategy benefit function. This first modification involves the following steps: calculating the motion direction consistency metric and motion speed matching metric between the unmanned node and its neighbor nodes based on the relative motion constraint boundary parameters; converting the motion direction consistency metric into a dimensionless direction cooperation utility value using a preset third normalization function, and converting the motion speed matching metric into a dimensionless speed cooperation utility value using a preset fourth normalization function; combining the direction cooperation utility value and the speed cooperation utility value according to preset weights to obtain a neighbor node cooperation benefit utility value; and adding the neighbor node cooperation benefit utility value to the initial strategy benefit function to obtain the first modified strategy benefit function, which is a dimensionless comprehensive utility value. Each unmanned node in the heterogeneous unmanned cluster is assigned a game interaction rule, which includes a strategy update trigger condition and a strategy selection interaction mode. The strategy update trigger condition is determined according to the update frequency parameter of the topological connection relationship set, and the strategy selection interaction mode includes a non-cooperative game interaction mode and a cooperative game interaction mode. The system monitors changes in the neighbor node identifier list of each unmanned node in the topological connection set. When a change in the neighbor node identifier list is detected, the relative motion constraint boundary parameters corresponding to the changed neighbor nodes are re-acquired based on the changed neighbor node identifier list. The first correction strategy benefit function of the unmanned node corresponding to the changed neighbor nodes is then subjected to a second correction process based on the re-acquired relative motion constraint boundary parameters to obtain a second correction strategy benefit function. The second correction process is performed through the following steps: determining the topological change impact metric based on the re-acquired relative motion constraint boundary parameters and the event type of neighbor node joining or leaving; converting the topological change impact metric into a dimensionless topological change compensation utility value based on a preset fifth normalization function; and adding the topological change compensation utility value to the first correction strategy benefit function to obtain the second correction strategy benefit function, which is a dimensionless comprehensive utility value. The second modified strategy payoff function is input to the strategy generator module of the upper-level strategy game domain. The strategy generator module performs payoff evaluation on each candidate strategy in the candidate strategy set of each unmanned node to obtain the payoff evaluation value corresponding to each candidate strategy. According to the strategy selection interaction method in the game interaction rules, the payoff evaluation values ​​of all candidate strategies in the candidate strategy set of each unmanned node are compared and selected to obtain the optimized strategy selection result of each unmanned node. The optimized strategy selection result includes the selected path planning scheme and the selected task allocation scheme. Based on the selected path planning scheme and the selected task allocation scheme in the optimization strategy selection results, a strategy execution instruction for each unmanned node is generated. All strategy execution instructions of unmanned nodes are packaged and processed to generate a hierarchical game control instruction containing the optimization strategy selection results of each unmanned node. The strategy execution instruction includes a path point sequence and the coordinates of the task target point.

5. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The process of receiving the hierarchical game control command through the lower-level action execution domain and generating real-time motion control parameters for each unmanned node based on the hierarchical game control command and the set of topological connections includes: The motion controller module of the lower-level action execution domain receives the hierarchical game control instructions and parses the strategy execution instructions for each unmanned node from the hierarchical game control instructions. The strategy execution instructions include a path point sequence and the coordinates of the task target point. Obtain the current node motion parameters for each unmanned node, which include the current node's spatial position coordinates, current node velocity vector, and current node acceleration vector; Based on the current spatial coordinates of each unmanned node and the path point sequence in the strategy execution instruction, calculate the expected velocity vector and expected acceleration vector required for each unmanned node to move from its current spatial coordinates to the next path point in the path point sequence. Based on the neighbor node identifier list of each unmanned node in the set of topological connections, the relative motion constraint boundary parameters in the set of edge attribute parameters between each unmanned node and its neighbor nodes are obtained. Based on the relative motion constraint boundary parameters, the desired velocity vector and the desired acceleration vector are subjected to a first constraint process to obtain the first-constrained velocity vector and the first-constrained acceleration vector. The first constraint process is used to ensure that the relative motion direction angle between the unmanned node and its neighbor nodes is within the range parameter of the relative motion direction angle, and that the relative motion velocity ratio is within the range parameter of the relative motion velocity ratio. Based on the set of motion capability parameters corresponding to the node attribute category of each unmanned node, the velocity vector and acceleration vector after the first constraint are subjected to a second constraint process to obtain the velocity vector and acceleration vector after the second constraint. The second constraint process is used to ensure that the magnitude of the velocity vector of the unmanned node does not exceed the maximum velocity magnitude parameter corresponding to the node attribute category, and the magnitude of the acceleration vector does not exceed the maximum acceleration magnitude parameter corresponding to the node attribute category. The second constrained velocity vector and the second constrained acceleration vector are input to the motion planning submodule of the motion controller module. The motion planning submodule generates real-time motion control parameters for each unmanned node. The real-time motion control parameters include velocity vector control and acceleration vector control. Based on the changing trend of the set of edge attribute parameters in the set of topological connections, the output frequency of the real-time motion control parameters is dynamically adjusted to generate a dynamically adjusted output frequency of the real-time motion control parameters. Based on the dynamically adjusted output frequency of the real-time motion control parameters, the real-time motion control parameters are grouped according to the time sequence to obtain the grouped real-time motion control parameter sequence, which is then sent to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform cooperative motion.

6. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The method further includes: Monitor the node motion parameter change events of each unmanned node in the heterogeneous unmanned cluster. When a node motion parameter change event is detected, obtain the changed node motion parameters. The changed node motion parameters include the changed node spatial position coordinates, the changed node velocity vector, and the changed node acceleration vector. Based on the changed spatial coordinates of the nodes and the neighbor node identifier list of each unmanned node in the topological connection set, the spatial distance parameter and relative velocity difference parameter between each unmanned node and other unmanned nodes in its detection area are recalculated. Based on the recalculated spatial distance parameter and relative velocity difference parameter, the communication link existence identifier and relative motion constraint boundary parameter between each unmanned node in the topological connection set and other unmanned nodes in its detection area are updated to obtain the updated topological connection set. The updated set of topological connections is compared with the set of topological connections before the update. The list of neighboring node identifiers that have changed and the set of edge attribute parameters that have changed are identified. Based on the identified list of neighboring node identifiers that have changed and the set of edge attribute parameters that have changed, a topological change event notification is generated. The topological change event notification includes the identifier information of the changed node and the identifier information of the change type. The topology change event notification is sent to the upper-level strategy game domain, triggering the upper-level strategy game domain to recalculate the strategy payoff function of the unmanned node based on the updated set of topology connections. The updated hierarchical game control instruction generated by the upper-level strategy game domain after recalculation based on the updated set of topology connections is received. The updated hierarchical game control command is received by the lower-level action execution domain, and the updated real-time motion control parameters of each unmanned node are generated according to the updated hierarchical game control command and the updated topology connection relationship set and sent to the corresponding unmanned node, thereby driving the heterogeneous unmanned cluster to perform cooperative motion after dynamic topology changes. Repeatedly execute the following operations: monitor node motion parameter change events, update the topology connection relationship set, generate topology change event notifications, receive updated hierarchical game control instructions, generate updated real-time motion control parameters, and send updated real-time motion control parameters, until the heterogeneous unmanned cluster completes all collaborative tasks.

7. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The method further includes: Based on the node attribute category and the node motion parameters, key unmanned nodes in a heterogeneous unmanned cluster are identified. The key unmanned nodes have information relay and forwarding capabilities and global task coordination capabilities. Assign a global strategy aggregator module in the upper-level strategy game domain to the key unmanned nodes. The global strategy aggregator module is used to collect the candidate strategy set and payoff evaluation value of all unmanned nodes in the first game node set. The global strategy aggregator module aggregates the collected candidate strategy set and the return evaluation values ​​to generate a global strategy evaluation result, which includes a global optimal strategy combination identifier and a global second-best strategy combination identifier. Based on the global optimal strategy combination identifier in the global strategy evaluation result, a global strategy coordination instruction is generated. The global strategy coordination instruction is used to instruct the unmanned nodes in the first game node set to perform cooperative operations according to the global optimal strategy combination. The global strategy coordination instruction is sent to each unmanned node in the first game node set, replacing the original optimization strategy selection result of each unmanned node in the first game node set. Based on the replaced optimization strategy selection result, the hierarchical game control instruction is regenerated to obtain the hierarchical game control instruction after global coordination. The lower-level action execution domain receives the globally coordinated hierarchical game control command, and generates the globally coordinated real-time motion control parameters for each unmanned node based on the globally coordinated hierarchical game control command and the set of topological connections, and sends them to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform globally coordinated cooperative motion. The motion state of the heterogeneous unmanned cluster is monitored under the real-time motion control parameters after global coordination. Motion state feedback parameters are obtained, and the aggregation weight parameters of the global policy aggregator module are adjusted according to the motion state feedback parameters to generate the adjusted aggregation weight parameters for the next global policy evaluation. The motion state feedback parameters include the deviation parameters between the actual motion trajectory and the planned motion trajectory of each unmanned node.

8. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The method further includes: Based on the neighbor node identifier list of each unmanned node in the set of topological connections, the topological connection-dense region and topological connection-sparse region in the heterogeneous unmanned cluster are identified. The number of neighbor nodes of the unmanned node in the topological connection-dense region is not less than a first threshold parameter, and the number of neighbor nodes of the unmanned node in the topological connection-sparse region is less than a second threshold parameter. Assign a dense region collaborative controller module in the upper-level strategy game domain to the unmanned nodes in the densely connected topology region. The dense region collaborative controller module is used to perform group strategy coordination processing on the unmanned nodes in the densely connected topology region. The dense region collaborative controller module obtains the candidate strategy set and benefit evaluation value of all unmanned nodes in the densely connected topology region, and performs joint optimization processing on the candidate strategy set of all unmanned nodes in the densely connected topology region to generate a joint optimization strategy combination for the dense region. The combined optimization strategy for the dense region is distributed to each unmanned node in the densely connected topology region, replacing the original optimization strategy selection result of each unmanned node in the densely connected topology region. Assign an independent decision-making module for the sparse region in the topological connection sparse region to the unmanned node in the topological connection sparse region. The independent decision-making module for the sparse region is used to perform independent policy decision-making processing on the unmanned node in the topological connection sparse region. The sparse region independent decision-making module obtains the candidate strategy set and benefit evaluation value of each unmanned node in the topologically connected sparse region, performs independent optimization processing on the candidate strategy set of each unmanned node in the topologically connected sparse region, and generates the sparse region independent optimization strategy selection result. The sparse region independent optimization strategy selection result is sent to each unmanned node in the topologically connected sparse region, replacing the original optimization strategy selection result of each unmanned node in the topologically connected sparse region. Based on the replaced dense region joint optimization strategy combination and sparse region independent optimization strategy selection result, a new hierarchical game control instruction is generated to obtain a regionally differentiated hierarchical game control instruction. The regionally differentiated hierarchical game control instruction is received through the lower-level action execution domain, and regionally differentiated real-time motion control parameters for each unmanned node are generated based on the regionally differentiated hierarchical game control instruction and the topological connection relationship set. The regionally differentiated real-time motion control parameters are sent to the corresponding unmanned node to drive the heterogeneous unmanned cluster to perform regionally differentiated cooperative motion.

9. The heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory according to claim 1, characterized in that, The method further includes: Based on the node attribute category and the node motion parameters, the remaining energy parameter and task completion progress parameter of each unmanned node in the heterogeneous unmanned cluster are obtained. The remaining energy parameter is used to characterize the energy reserve value currently available to the unmanned node, and the task completion progress parameter is used to characterize the proportion of the task amount currently completed by the unmanned node to the total number of allocated tasks. Based on the remaining energy parameter and the task completion progress parameter, the energy consumption cost-utility value in the initial strategy benefit function of each unmanned node is dynamically adjusted to obtain the dynamically adjusted strategy benefit function. The dynamically adjusted strategy payoff function is input to the strategy generator module in the upper-level strategy game domain. The strategy generator module performs energy-aware payoff evaluation on each candidate strategy in the candidate strategy set of each unmanned node to obtain an energy-aware payoff evaluation value. Based on the energy-aware payoff evaluation value, the candidate strategies in the candidate strategy set of each unmanned node are reordered to obtain the energy-aware optimized strategy selection result. Based on the energy-aware optimized strategy selection result, an energy-aware hierarchical game control instruction is generated. The lower-level action execution domain receives the energy-sensing hierarchical game control command, and generates energy-sensing real-time motion control parameters for each unmanned node based on the energy-sensing hierarchical game control command and the set of topological connections. The energy-sensing real-time motion control parameters are then sent to the corresponding unmanned nodes to drive the heterogeneous unmanned cluster to perform energy-equalizing cooperative motion. Monitor the changes in the remaining energy parameters of each unmanned node during the execution of the energy equalization cooperative motion, obtain the sequence of remaining energy parameter changes, identify energy-limited unmanned nodes whose remaining energy parameters are lower than the third threshold parameter based on the sequence of remaining energy parameter changes, and generate a list of energy-limited node identifiers. Based on the energy-constrained node identifier list, the task allocation scheme of the energy-constrained unmanned nodes is adjusted, and some tasks of the energy-constrained unmanned nodes are reassigned to other unmanned nodes whose remaining energy parameters are higher than the fourth threshold parameter. The task reassignment result is generated, and the energy-aware hierarchical game control command is regenerated based on the task reassignment result to drive the heterogeneous unmanned cluster to perform the cooperative movement after the task reassignment.

10. A heterogeneous unmanned swarm control system combining dynamic topology and hierarchical game theory, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the heterogeneous unmanned swarm control method combining dynamic topology and hierarchical game theory as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images