Distributed multi-agent three-dimensional motion control model construction method

The distributed three-dimensional motion control model addresses formation ambiguity and communication-restricted scenarios by using a directed persistent network topology with minimal edges and projection-based control, ensuring stable and robust formation maintenance in multi-agent systems.

CN119065386APending Publication Date: 2024-12-03HUAZHONG UNIV OF SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411135751.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing three-dimensional multi-agent formation control methods face challenges in maintaining formation under communication-restricted conditions, particularly in scenarios where agents cannot maintain consistent reference directions, leading to formation ambiguity and local convergence issues.

Method used

A distributed multi-agent three-dimensional motion control model is constructed using a three-dimensional directed persistent network communication topology with minimal edges, incorporating node types like first, second, and third leaders, and followers, with motion control models based on relative position information and projection constraints, enabling formation maintenance and adaptation to limited communication scenarios.

Benefits of technology

The method ensures stable formation under communication limitations by using relative position information and local network reconstruction, achieving global convergence and robustness against agent failures or interference, suitable for drone and unmanned submarine clusters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065386A_ABST
    Figure CN119065386A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of multi-agent control, and particularly relates to a distributed multi-agent three-dimensional motion control model construction method, which is used for realizing a target formation, and comprises the following steps: constructing a three-dimensional directed persistent network communication topological graph with least edges according to a three-dimensional target formation of a to-be-controlled multi-agent system, wherein the node types comprise a first leader, a second leader, a third leader and a follower; according to the node type of each agent, correspondingly constructing a three-dimensional motion control model of which the input is the relative position information of all subsequent neighbor agents of the agent relative to the agent, and the output is the motion speed of the agent; wherein the three-dimensional motion control model of each follower is a three-dimensional formation control model based on projection. According to the method, the problem of three-dimensional distributed formation control under the condition that communication is limited can be solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of multi-agent control, and more specifically, relates to a distributed multi-agent three-dimensional motion control model construction method. Background Art

[0002] Multi-agent formation control refers to the process of manipulating the behavior of multiple agents to form a specific formation in space and maintain a stable formation. By selecting a suitable formation, the detection capability and collaborative combat capability of the multi-agent system can be effectively improved, thereby achieving an effect that is better than the simple accumulation of a single agent. Multi-agent formation control technology has a wide range of applications in robot control, drone formation, unmanned vehicle formation, autonomous driving and other fields. Compared with centralized formations, distributed formations have higher flexibility, intelligence and efficiency to cope with external changes and interference, and have wider application potential.

[0003] For the two-dimensional formation control method, the invention patent CN112558613B discloses a formation control method based on a complex Laplace matrix. The method first establishes a network communication topology map, then constructs a corresponding complex Laplace matrix based on the map and the target formation, and uses the complex Laplace matrix to design distributed control signals. However, like most displacement-based formation control methods, this method has an implicit constraint that all agents have the same reference direction (such as the compass direction), which is often not guaranteed in high-speed dynamic change scenarios and large-scale formations.

[0004] Reference [1] (Yu D, Chen CL P. Automatic leader–follower persistent formation generation with minimum agent-movement in various switching topologies [J]. IEEE transactions on cybernetics, 2018, 50 (4): 1569-1581) proposed an algorithm that uses triangulation combined with rigid graph and persistent graph theory to solve the initial formation generation and formation transformation problems of regular triangle formations. At the same time, a distance-based controller is used to control each agent. This method is suitable for situations where communication constraints are limited. However, this method does not mention how to control the formation in three dimensions.

[0005] Regarding the three-dimensional formation control method, the invention patent CN107632614B discloses a multi-UAV formation self-organizing cooperative control method based on rigid graph theory. The method first gives a specific step of constructing a multi-UAV target rigid network topology structure based on a rigid graph, and then gives an obstacle avoidance method when the UAV formation encounters obstacles during flight. However, this method cannot guarantee global convergence (there are some initial positions that cannot form a target formation) and does not involve a formation repair algorithm for multi-agent failures.

[0006] Compared with two-dimensional formation control, three-dimensional formation control needs to further consider the control in the height direction. When only local information is used for formation control, the design of the distributed controller is more complicated. In addition, most of the current three-dimensional formation control methods rely on the global position information (GPS positioning) of each agent or require each agent to maintain the same reference direction. They cannot complete the formation task under certain communication-restricted conditions, such as strong magnetic field interference or underwater clustering scenarios. In this case, each agent can only perceive the distance information of a limited number of local neighbors, so it is impossible to ensure that the reference direction of each agent is consistent. Although the formation control method based on rigid graph is applicable to the above scenarios, it theoretically only has local convergence and is sensitive to the initial position of the agent. There is a formation ambiguity problem in the formation and switching process of the formation at certain initial positions, which makes it impossible to guarantee the formation of the target formation. Summary of the invention

[0007] In view of the defects of the prior art and the need for improvement, the present invention provides a distributed multi-agent three-dimensional motion control model construction method, which aims to solve the problem of three-dimensional distributed formation control under communication restricted conditions.

[0008] To achieve the above object, according to one aspect of the present invention, a distributed multi-agent three-dimensional motion control model construction method is provided, which is used to realize the target formation, comprising:

[0009] According to the three-dimensional target formation of the multi-agent system to be controlled, a three-dimensional directed persistent network communication topology graph with the least edges is constructed, wherein the node types include the first leader, the second leader, the third leader and the follower;

[0010] According to the node type of each agent, a three-dimensional motion control model is constructed, in which the input of the agent is the relative position information of all the successor neighbor agents of the agent relative to the agent, and the output is the motion speed of the agent; wherein the three-dimensional motion control model of each follower is a three-dimensional formation control model based on projection, and the construction method is:

[0011] Based on the relative position vectors of the three successor neighbor intelligent agent nodes relative to the follower, two mutually orthogonal unit direction vectors located in the plane formed by the three successor neighbor intelligent agent nodes and a unit direction vector perpendicular to the plane are determined; one of the two mutually orthogonal unit direction vectors is parallel to the line connecting any two successor neighbors; the projections of the relative position vectors of any intelligent agent node among the three successor neighbor intelligent agent nodes on the three mutually orthogonal unit vectors tend to correspond to expected projection values ​​as three projection constraints, and the three-dimensional motion control model of the follower is obtained through weighted summation; the successor neighbor intelligent agent node is the end node of the directed edge corresponding to the follower, and the successor neighbor intelligent agent node and the expected projection value are both determined according to the three-dimensional directed persistent network communication topology graph with the least edges.

[0012] Furthermore, the three-dimensional directed persistent network communication topology graph is constructed as follows:

[0013] Based on the basic communication topology conditions, the target formation is tetrahedron-divided to obtain an undirected tetrahedron-divided graph, and the intelligent node as the first leader, the intelligent node as the second leader, the intelligent node as the third leader and the intelligent node as the follower are determined from the tetrahedron-divided graph. According to the predecessor-successor neighbor relationship, some node connections in the tetrahedron-divided graph are directed and redundant edges are deleted to obtain a three-dimensional directed persistent network communication topology with the least edges; wherein the basic communication topology conditions are: the undirected graph corresponding to the topology has rigidity, and the directed graph corresponding to the topology has persistence.

[0014] Furthermore, the three-dimensional directed persistent network communication topology graph is constructed in the following manner:

[0015] The target formation is partitioned by a tetrahedron partitioning algorithm to obtain an undirected tetrahedron partitioning graph G = (V, E), and a node set P and a directed edge set Q are established, which are initialized to an empty set; the node set V = {v1, v2, v3, …, v n} represents the set of agents in the multi-agent system to be controlled, and the edge set E = {e1, e2, e3, …, e m} represents all communication links between agents, and the directed edge e=(v i ,v j ) represents the agent node v i With agent node v j One-way communication and v i Get v j Relative position information, v i Called v j The previous neighbor agent node, v j Called vi The successor neighbor agent node of

[0016] According to the preset intelligent agent node v1 as the first leader, the tetrahedron structure including v1 is selected from the tetrahedron partition graph G = (V, E) as the initial tetrahedron, and any two intelligent agent nodes v2 and v3 other than the intelligent agent node v1 are selected from the initial tetrahedron as the second leader and the third leader respectively, and the intelligent agent node v4 other than v1, v2, and v3 in the initial tetrahedron is used as a follower; the intelligent agent nodes v1, v2, v3, and v4 corresponding to the four vertices of the initial tetrahedron are used as initial nodes, added to the node set P, and a directed edge e is constructed. 21 =(v2,v1),e 31 =(v3,v1),e 32 =(v3,v2),e 41 =(v4,v1),e 42 =(v4,v2) and e 43 =(v4,v3), added to the directed edge set Q;

[0017] Search for an agent node v from the remaining node set V\P that has an edge with any three nodes in the current node set P i , connect the three edges with v i Direct the predecessor neighbor agent node, add the directed edge set Q, and convert v i Add the current node set P; repeat this step until the remaining node set V\P is empty, delete the non-directed connections, and obtain the network communication topology diagram G'=(P,Q).

[0018] Furthermore, the three-dimensional motion control model of the second leader is a distance-based gradient control model, and the distance constraint it adopts is that the second leader is only subject to the distance constraint of the first leader; the three-dimensional motion control model of the third leader is a distance-based gradient control model, and the distance constraint it adopts is that the third leader is subject to the dual distance constraints of the first leader and the second leader.

[0019] Furthermore, the three-dimensional motion control model of the second leader is specifically as follows:

[0020]

[0021] In the formula, represents the movement speed of the second leader, x1 and x2 are the current position vectors of the first leader and the second leader respectively, x1 * ,x2 *are the target position vectors of the first leader and the second leader in the target formation, respectively; d represents the expected distance between the first leader and the second leader; the symbol ‖x‖ represents the length of the vector x; and c is a parameter used to control the formation speed of the target formation;

[0022] The three-dimensional motion control model of the third leader is specifically:

[0023]

[0024] In the formula, represents the movement speed of the third leader, x3 is the current position vector of the third leader, x3 * is the target position vector of the third leader in the target formation, d 13 ,d 23 They represent the expected distances between the third leader and the first leader and the second leader in the target formation respectively.

[0025] Furthermore, the three-dimensional motion control model of the follower is specifically:

[0026]

[0027] In the formula, Indicates follower v i The speed of movement; x j ,x k and x m For followers v i The three successor neighbor agent nodes v j 、v k and v m The current position vector of the agent; ε is a sufficiently small positive constant; the symbol × represents the vector product multiplication in three-dimensional space; h1, h2 and h3 are all intermediate variables, and the intermediate variable h1 represents the agent node v j 、v k The unit relative direction vector between them, h2 represents the relative direction vector with respect to the agent node v j 、v k and v m The three form a unit direction vector that is orthogonal to the plane. h3 represents a unit vector that is orthogonal to h1 and h2. ij =x i -x j , indicating v i and v j The relative position of the superscript symbol T indicates transposition, x i For followers v i The current position vector, x j Indicates v j The current position vector of x i *,x j * ,x k * ,x m * v i 、v j 、v k and v m The target position vector in the target formation; s1, s2, s3 are the projection parameters in the target formation, determined according to the target formation; c is the parameter used to control the formation speed of the target formation.

[0028] The present invention also provides a multi-agent distributed three-dimensional formation control method, which uses the three-dimensional motion control model of each agent constructed by the multi-agent three-dimensional motion control model construction method as described above to adjust the movement speed of the corresponding agent, and ultimately makes the formation formed by all agents in space consistent with the target formation formation.

[0029] Furthermore, when an agent failure occurs or communication between agents is disturbed, the method further comprises:

[0030] The faulty agent and the two agents whose communication edges are disturbed are regarded as faulty agents. In the order of the first leader, the second leader, the third leader, and the follower, the nodes corresponding to the faulty agents are removed from the three-dimensional directed persistent network communication topology with the least edges through the local network reconstruction algorithm, and the directed edges are re-added to ensure that the node removal does not weaken the stability and anti-interference of the formation, so as to achieve the adjustment of the three-dimensional directed persistent network communication topology.

[0031] According to the adjusted three-dimensional directed persistent network communication topology graph, the projection parameters or distance parameters in the three-dimensional motion control model of each intelligent agent are adjusted.

[0032] Furthermore, if the faulty intelligent agent is the first leader, the adjustment method of the three-dimensional directed persistent network communication topology graph is as follows: all directed edges connected to the intelligent agent node are deleted from the three-dimensional directed persistent network communication topology graph, the current second leader node is changed to a new first leader node, the third leader node is changed to a new second leader node, the follower node in the initial tetrahedron is changed to a new third leader node, and all follower predecessor neighbor intelligent agent nodes of the intelligent agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node;

[0033] If the faulty agent is the second leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node remains unchanged, the current third leader node is changed to the new second leader node, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node;

[0034] If the faulty agent is the third leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node and the second leader node remain unchanged, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node;

[0035] If the faulty agent is a follower, the network communication topology is adjusted as follows: all predecessor neighbor agent nodes of the agent node in the original network communication topology constitute a set V1, and the successor neighbor agent nodes constitute a set V2. Each node of V1 is traversed in turn, and a non-neighbor agent node of the node in the original network communication topology is searched from V2. A directed edge starting from the V1 node is established, and all directed edges connected to the agent node are deleted.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for constructing a multi-agent three-dimensional motion control model as described above or the method for controlling a multi-agent distributed three-dimensional formation as described above is implemented.

[0037] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0038] (1) The present invention proposes a distributed multi-agent three-dimensional motion control model construction method. According to the three-dimensional target formation of the multi-agent system to be controlled, a three-dimensional directed persistent network communication topology with the least edges is constructed, which is used to distribute the characterization of the successor neighbor relationship of each agent, wherein the node types include the first leader, the second leader, the third leader and the follower; according to the node type of each agent, a three-dimensional motion control model of the agent is constructed accordingly, wherein the input is the relative position information of all the successor neighbor agents of the agent relative to the agent, and the output is the motion speed of the agent. In particular, the motion control model of the follower proposed in the present invention is a projection-based three-dimensional formation control model. This control model only requires the relative displacement information of a small number of neighbor agents to complete the formation task and avoids the formation ambiguity problem. In addition, this method does not require that the reference direction of each agent be consistent during measurement, and can adapt to conditions with limited communication. Therefore, it can be applied to the formation maintenance and change of drone clusters or unmanned submarine clusters underwater, in the air or under communication-limited conditions.

[0039] (2) In order to ensure the robustness of the formation, the present invention also designs a local network reconstruction algorithm for the case of intelligent agent failure or interference between intelligent agents, so that the faulty intelligent agent or interference between intelligent agents will not affect the formation of normal intelligent agents, thereby improving the anti-interference ability of the method of the present invention. In addition, compared with re-constructing a three-dimensional directed persistent network communication topology graph for the remaining normal agents, the local network reconstruction algorithm proposed in the present invention achieves the same effect with the least edge addition or deletion operations, and therefore has a higher repair efficiency for the formation destruction caused by intelligent agent failure or interference between intelligent agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of a distributed multi-agent three-dimensional motion control model construction method provided by an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of constructing a three-dimensional directed persistent network communication topology graph with minimal edges provided by an embodiment of the present invention, wherein (a) represents an initial tetrahedron decomposition graph; (b) represents a directed network communication topology graph;

[0042] Figure 3 A schematic diagram of a projection-based motion control model provided by an embodiment of the present invention;

[0043] Figure 4 A network communication topology diagram corresponding to a target formation of a ten-agent system provided by an embodiment of the present invention;

[0044] Figure 5The initial formation diagram based on distance control provided by an embodiment of the present invention, wherein (a) is a diagram of the approximate trajectory of the intelligent body movement, and the numbered ones are the starting points; (b) is a diagram of the distance error change;

[0045] Figure 6 The initial formation diagram based on projection control provided by the embodiment of the present invention, wherein (a) is a diagram showing the approximate trajectory of the intelligent body movement; (b) is a diagram showing the distance error variation;

[0046] Figure 7 A schematic diagram of adjusting a three-dimensional directed persistent network communication topology provided by an embodiment of the present invention; wherein (a) is an original three-dimensional directed persistent network communication topology, and the numbers in the figure represent the numbers of the intelligent agent nodes; (b) is a new three-dimensional directed persistent network communication topology after local reconstruction when the intelligent agent node 3 fails;

[0047] Figure 8 Figure 1 is a diagram of the repair of a sudden fault formation of an intelligent agent provided by an embodiment of the present invention; wherein (a) is a diagram of the motion trajectory of the intelligent agent, in which the formation moves upward to the right as a whole, and a sudden fault occurs in the intelligent agent node F during the process and the formation begins to fall, triggering a network reconstruction algorithm to ensure the stability of the remaining intelligent agent formation; (b) is a diagram of the change in distance error, in which the solid line represents the change in distance error between non-faulty intelligent agent nodes, and the dotted line represents the distance error between the faulty intelligent agent node F and a normal intelligent agent node;

[0048] Fig. 9 A flow chart of multi-agent distributed three-dimensional formation control under communication-constrained conditions provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] Embodiment 1

[0051] A distributed multi-agent three-dimensional motion control model construction method, such as Figure 1 As shown, it is used to achieve the target formation, including:

[0052] According to the three-dimensional target formation of the multi-agent system to be controlled, a three-dimensional directed persistent network communication topology graph with the least edges is constructed to represent the successor neighbor relationship of each agent in a distributed manner, wherein the node types include the first leader, the second leader, the third leader and the follower;

[0053] According to the node type of each agent, a three-dimensional motion control model is constructed, in which the input of the agent is the relative position information of all the successor neighbor agents of the agent relative to the agent, and the output is the motion speed of the agent; wherein the three-dimensional motion control model of each follower is a three-dimensional formation control model based on projection, and the construction method is:

[0054] Based on the relative position vectors of the three successor neighbor intelligent agent nodes relative to the follower, two mutually orthogonal unit direction vectors located in the plane formed by the three successor neighbor intelligent agent nodes and a unit direction vector perpendicular to the plane are determined; one of the two mutually orthogonal unit direction vectors is parallel to the line connecting any two successor neighbors; the projections of the relative position vectors of any intelligent agent node among the three successor neighbor intelligent agent nodes on the three mutually orthogonal unit vectors tend to correspond to expected projection values ​​as three projection constraints, and the three-dimensional motion control model of the follower is obtained through weighted summation; the successor neighbor intelligent agent node is the end node of the directed edge corresponding to the follower, and the successor neighbor intelligent agent node and the expected projection value are both determined according to the three-dimensional directed persistent network communication topology graph with the least edges.

[0055] The motion control model of the first leader is determined by the motion trajectory of the target formation; the control models of other agents use it as a reference to form the target formation.

[0056] This embodiment designs a distributed control model for controlling the motion of each intelligent agent node according to the category of the intelligent agent node.

[0057] As a preferred implementation, the above three-dimensional directed persistent network communication topology graph is constructed as follows:

[0058] Based on the basic communication topology conditions, the target formation is tetrahedron-divided to obtain an undirected tetrahedron-divided graph, and the intelligent node as the first leader, the intelligent node as the second leader, the intelligent node as the third leader and the intelligent node as the follower are determined from the tetrahedron-divided graph. According to the predecessor-successor neighbor relationship, some node connections in the tetrahedron-divided graph are directed and redundant edges are deleted to obtain a three-dimensional directed persistent network communication topology with the least edges; wherein the basic communication topology conditions are: the undirected graph corresponding to the topology has rigidity, and the directed graph corresponding to the topology has persistence.

[0059] In order to cope with the situation of limited communication, this method is based on the basic communication topology conditions that the distance formation control method should meet, that is, the undirected graph satisfies rigidity (the graph will not be deformed while ensuring that the length of each edge remains unchanged), and the directed graph satisfies persistence (the persistence graph in the three-dimensional case satisfies the requirement that the number of outgoing edges of all nodes with out-degree greater than 3 is reduced to 3, and the corresponding undirected graph is still a rigid graph). The communication network topology diagram inside the multi-agent is constructed. At the same time, in order to save communication resources as much as possible, this method adopts a directed network with unidirectional communication and minimizes the number of edges in the constructed network. The goal of this method is to construct a directed persistent network communication topology diagram with the least edges according to the target formation.

[0060] As a preferred implementation, the above three-dimensional directed persistent network communication topology diagram is constructed in the following manner:

[0061] The target formation is partitioned by a tetrahedron partitioning algorithm to obtain an undirected tetrahedron partitioning graph G = (V, E), and a node set P and a directed edge set Q are established, which are initialized to an empty set; the node set V = {v1, v2, v3, …, v n} represents the set of agents in the multi-agent system to be controlled, and the edge set E = {e1, e2, e3, …, e m} represents all communication links between agents, and the directed edge e=(v i ,v j ) represents the agent node v i With agent node v j One-way communication and v i Get v j Relative position information, i.e. v i Can directly obtain v j Relative position information, and v j Cannot directly obtain v i Relative position information, v i Called v j The predecessor neighbor agent node, v j Called v i The successor neighbor agent node of

[0062] According to the preset intelligent agent node v1 as the first leader, the tetrahedron structure including v1 is selected from the tetrahedron partition graph G = (V, E) as the initial tetrahedron, and any two intelligent agent nodes v2 and v3 other than the intelligent agent node v1 are selected from the initial tetrahedron as the second leader and the third leader respectively, and the intelligent agent node v4 other than v1, v2, and v3 in the initial tetrahedron is used as a follower; the intelligent agent nodes v1, v2, v3, and v4 corresponding to the four vertices of the initial tetrahedron are used as initial nodes, added to the node set P, and a directed edge e is constructed. 21 =(v2,v1),e31 =(v3,v1),e 32 =(v3,v2),e 41 =(v4,v1),e 42 =(v4,v2) and e 43 =(v4,v3), added to the directed edge set Q;

[0063] Search for an agent node v from the remaining node set V\P that has an edge with any three nodes in the current node set P i , connect the three edges with v i Direct the predecessor neighbor agent node, add the directed edge set Q, and convert v i Add the current node set P; repeat this step until the remaining node set V\P is empty, delete the non-directed connections, and obtain the network communication topology diagram G'=(P,Q).

[0064] Figure 2 An example of the process of constructing a network communication topology diagram is given, where (a) is the initial tetrahedron decomposition diagram obtained after the target formation is tetrahedron-divided. Then, according to the algorithm flow of constructing the communication network in the present invention, tetrahedron 1234 is selected as the initial tetrahedron, and the final directed communication network is gradually expanded, as shown in FIG. Figure 2 As shown in (b) in the figure, the node number represents the node marking order in Algorithm 1, where nodes 1, 2, and 3 are the first, second, and third leaders respectively.

[0065] This method proposes a method to construct a three-dimensional minimum directed persistent network using tetrahedron partitioning. First, the target formation is processed using the tetrahedron partitioning algorithm to obtain the initial tetrahedron partitioning graph, which is an undirected graph. Then, according to the algorithm rules, it is directed and redundant edges are deleted to obtain the minimum directed persistent graph as the communication network for the formation control of the multi-agent system.

[0066] After the network communication topology is constructed, a three-dimensional minimum directed persistent network is obtained, and the first leader, second leader, third leader and followers are determined. Figure 2 As shown in (b), agent node 1 is the first leader with an out-degree of 0 and no successor node; agent node 2 is the second leader with an out-degree of 1 and the successor node is the first leader; agent node 3 is the third leader with an out-degree of 2 and the successor nodes are the first leader and the second leader; the remaining agent nodes are followers with an out-degree of 3 and three different successor nodes. The three-dimensional minimum directed persistent network obtained by the above method has only one leader of each type and the remaining nodes are followers.

[0067] According to the directed network communication topology constructed by the aforementioned method, all the successor neighbor agents of each agent are designated as the control input of the distributed controller of the agent, and different motion controller forms are adopted according to the corresponding node type of the agent.

[0068] The motion control model of each agent adjusts the movement speed (size and direction) of the agent through the position information of its neighbors (all subsequent neighbor agents that communicate with it), and finally makes the formation shape of all agents in space consistent with the target formation. The first leader (whose controller is the controller that determines how the formation moves) is used to determine the formation position, and other agents need to use it as a reference to form a specific formation.

[0069] The second leader is only constrained by the distance from the first leader ((||x1-x2|| 2 =d 2 )), directly using the existing distance-based gradient controller can ensure global convergence (no other ambiguous equilibrium points). The third leader is subject to the distance constraints from the first leader and the second leader, and can adopt a controller form similar to the distance-based controller in a two-dimensional plane. As a preferred implementation, the three-dimensional motion control model of the second leader is a distance-based gradient control model, and the distance constraint it adopts is that the second leader is only subject to the distance constraint of the first leader; the three-dimensional motion control model of the third leader is a distance-based gradient control model, and the distance constraint it adopts is that the third leader is subject to the dual distance constraints of the first leader and the second leader.

[0070] For all followers, it can be preferred that in order to ensure the global convergence of the formation at any initial position, this embodiment proposes a three-dimensional formation control model based on projection. Taking follower i as an example, the specific definition of the projection-based formation model is explained. First, the following intermediate variables are defined:

[0071]

[0072] The symbol × in the above formula represents the vector product multiplication in three-dimensional space. The specific operation rules are as follows:

[0073]

[0074] According to the intermediate variables defined above, the projection-based three-dimensional distributed formation control model can be simplified as follows:

[0075]

[0076] where x ij =x i -x j is the relative position of agent nodes i and j, symbol xT Represents the transpose of vector x. i * ,x j * ,x k * ,x m * is the position vector of i, j, k, m in the target formation, s1, s2, s3 are the projection parameters in the target formation. Since the target formation is known, the parameters can be uniquely determined by the calculation formula. It is worth noting that in the projection-based three-dimensional distributed formation control model, the position vector of each intelligent node in the target formation appears as a difference, and can be represented by the relative position of the successor neighbor relative to i, so no global position information is required. In addition, the above formula is also rotationally invariant, so there is no need for the intelligent node to identify the direction.

[0077] As a preferred implementation, the three-dimensional motion control model of the second leader is specifically as follows:

[0078]

[0079] In the formula, represents the movement speed of the second leader, x1 and x2 are the current position vectors of the first leader and the second leader respectively, x1 * ,x2 * are the target position vectors of the first leader and the second leader in the target formation, respectively; d represents the expected distance between the first leader and the second leader; the symbol ‖x‖ represents the length of the vector x; and c is a parameter used to control the formation speed of the target formation;

[0080] The three-dimensional motion control model of the third leader is specifically:

[0081]

[0082] In the formula, represents the movement speed of the third leader, x3 is the current position vector of the third leader, x3 * is the target position vector of the third leader in the target formation, d 13 ,d 23 They represent the expected distances between the third leader and the first leader and the second leader in the target formation respectively.

[0083] As a preferred implementation, the three-dimensional motion control model of the follower is specifically as follows:

[0084]

[0085] In the formula, Indicates follower v i The speed of movement; x j ,x k and x m For followers v i The three successor neighbor agent nodes v j 、v k and v m The current position vector of the agent; ε is a sufficiently small positive constant; the symbol × represents the vector product multiplication in three-dimensional space; h1, h2 and h3 are all intermediate variables, and the intermediate variable h1 represents the agent node v j、 v k The unit relative direction vector between them, h2 represents the relative direction vector with respect to the agent node v j 、v k and v m The three form a unit direction vector that is orthogonal to the plane. h3 represents a unit vector that is orthogonal to h1 and h2. ij =x i -x j , indicating v i and v j The relative position of the superscript symbol T indicates transposition, x i For followers v i The current position vector, x j Indicates v j The current position vector of x i * ,x j * ,x k * ,x m * v i 、v j 、v k and v m The target position vector in the target formation; s1, s2, s3 are the projection parameters in the target formation, determined according to the target formation; c is the parameter used to control the formation speed of the target formation.

[0086] The distance parameters and projection parameters required in the above control models can be obtained by pre-calculating the position vectors of each intelligent agent in the target formation in the reference coordinate system.

[0087] It is worth noting that in the follower's three-dimensional motion control model, the position vectors of each agent in the target formation appear as differences, and can be represented by the relative position of the successor neighbor relative to i, so no global position information is required. In addition, the above formula is rotationally invariant, so there is no need for the agent to identify the direction.

[0088] The geometric interpretation of the projection-based control model is given below, Figure 3 As shown, there is a target formation composed of five intelligent agent nodes i, j, k, m, and n. In order to simplify the analysis, we only focus on the control of intelligent agent node i. If a traditional distance-based controller is used, there are two possible equilibrium positions i and i′ in the plane. Specifically, when the initial position of intelligent agent node i is in the upper and lower half planes respectively, it will eventually move to two different equilibrium positions. When the number of intelligent agent nodes is 4, although it will not affect the formation shape, the addition of intelligent agent node n will obviously lead to the generation of two completely different formations. The situation is even worse when the number of intelligent agent nodes continues to increase. The present invention proposes to solve the problem of non-unique equilibrium point of the above-mentioned intelligent agent i in three-dimensional space by using a projection-based control model. Specifically, by constructing a direction vector h2 that is perpendicular to the plane where the three successor neighbor nodes j, k, and m are located, and through x ij The positive and negative distinction on the h2 projection is based on the various equilibrium points in the distance control method. Therefore, the relative position relationship between h2 and the neighboring nodes can be used to construct a three-dimensional orthogonal vector group h1, h2, h3, and the x ij The distance between agents is indirectly controlled by projecting values ​​on a three-dimensional orthogonal vector group, and the target formation is finally formed. Even in some communication-restricted situations, each agent node only needs to obtain a small amount of relative position information of the subsequent agent nodes to achieve a stable formation. Compared with the distance-based formation control, this method is insensitive to the initial position of the agent node, and compared with the consistency-based formation control, this method is insensitive to the reference direction of each agent when measuring.

[0089] In general, the projection-based control model makes the formation converge globally: where i′ is the mirror equilibrium point of the desired position i. In the distance-based control method, both i′ and i satisfy the distance constraints and therefore cannot be distinguished; however, the projection-based method can distinguish between i′ and i by the positivity of the projections of the orthogonal vector h2 that is perpendicular to the plane formed by the subsequent neighbor nodes j, k, and m.

[0090] In addition, from the transmission relationship and rigidity conditions of the generated communication topology graph, it can be seen that the second leader determines its relative position in the formation through the first leader, the third leader determines its relative position in the formation through the first and second leaders, and the remaining followers will directly or indirectly establish positional connections with the leaders through the transmission relationship of the directed graph. Therefore, independent control models can be set for each intelligent agent node. They control their own speed by obtaining the relative position information of the successor neighbor intelligent agent node relative to themselves, thereby adjusting the relative position relationship between themselves and the successor neighbor intelligent agent node to ensure that the final formation of the entire multi-agent system is the target formation. Specifically, the first leader control model is preset according to the task requirements, the second leader and the third leader control models adjust the relative position relationship between themselves and the successor neighbor intelligent agent node through the distance constraint between themselves and the successor neighbor intelligent agent node, and the follower control model adjusts the relative position relationship between itself and the successor neighbor intelligent agent node through the three projection constraints between itself and the successor neighbor intelligent agent node.

[0091] Next, a MATLAB simulation experiment is conducted to verify the actual effect of the three-dimensional multi-agent formation control method proposed in the present invention. Taking a ten-agent system as an example, assuming that the target formation is as follows: Figure 4 As shown, the dotted line in the figure is a network communication topology diagram constructed by the method of embodiment 1 (the direction arrow is omitted).

[0092] After that, the initial position of the random agent is given, and the formation effect is finally formed by comparing the traditional distance-based control method with the control method of the present invention. Figure 5 Shown are the motion trajectories of each agent under the distance control method (the dotted line is the final connection topology of the agent), and the distance error change diagram (distance error: the difference between the current distance between any two agents in the multi-agent system and the expected distance in the target formation). From the distance error change diagram, it can be seen that only some distance errors eventually tend to 0. After reaching the steady state, there are multiple pairs of agents whose distance errors are not 0, so they do not meet the formation requirements.

[0093] Figure 6 The motion trajectory and distance error change diagram of each agent under the control method of the present invention are shown. From the motion trajectory diagram, it can be seen that the final formation shape is consistent with the target formation. Further, from the distance error change diagram, it can be seen that the distance errors between agents eventually tend to 0, which meets the requirements of the target formation. Therefore, the control method of the present invention can overcome the problem that the traditional distance-based control method cannot form a target formation when the initial position of the agent is random.

[0094] Embodiment 2

[0095] A multi-agent distributed three-dimensional formation control method adopts the three-dimensional motion control model of each agent constructed by the multi-agent three-dimensional motion control model construction method described in the first embodiment above, adjusts the movement speed of the corresponding agent, and finally makes the formation formed by all agents in space consistent with the target formation formation, that is, it can be used to complete control tasks such as target formation generation, maintenance and switching.

[0096] As a preferred implementation, when an intelligent agent failure occurs or communication between intelligent agents is disturbed, the method further includes:

[0097] The faulty agent and the two agents whose communication edges are disturbed are regarded as faulty agents. In the order of the first leader, the second leader, the third leader, and the follower, the nodes corresponding to the faulty agents are removed from the three-dimensional directed persistent network communication topology with the least edges through the local network reconstruction algorithm, and the directed edges are re-added to ensure that the node removal does not weaken the stability and anti-interference of the formation, so as to achieve the adjustment of the three-dimensional directed persistent network communication topology.

[0098] According to the adjusted three-dimensional directed persistent network communication topology graph, the projection parameters or distance parameters in the three-dimensional motion control model of each intelligent agent are adjusted.

[0099] This method is only triggered when an intelligent agent fails seriously or the communication edge between intelligent agents is strongly disturbed. The original network communication topology is quickly adjusted through the local network reconstruction algorithm to remove the nodes corresponding to the faulty intelligent agent and ensure that the node removal will not weaken the stability and anti-interference of the formation.

[0100] As a preferred implementation, if the faulty agent is the first leader, the adjustment method of the three-dimensional directed persistent network communication topology graph is as follows: all directed edges connected to the agent node are deleted from the three-dimensional directed persistent network communication topology graph, the current second leader node is changed to a new first leader node, the third leader node is changed to a new second leader node, the follower node in the initial tetrahedron is changed to a new third leader node, and all follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node (the newly added edge must not exist in the original communication network topology);

[0101] If the faulty agent is the second leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node remains unchanged, the current third leader node is changed to the new second leader node, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node;

[0102] If the faulty agent is the third leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node and the second leader node remain unchanged, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node;

[0103] If the faulty agent is a follower, the network communication topology is adjusted as follows: all predecessor neighbor agent nodes of the agent node in the original network communication topology constitute a set V1, and the successor neighbor agent nodes constitute a set V2. Each node of V1 is traversed in turn, and a non-neighbor agent node of the node in the original network communication topology is searched from V2. A directed edge starting from the V1 node is established, and all directed edges connected to the agent node are deleted.

[0104] It should be noted that when only the communication fails, the agents at both ends of the communication will join the formation as followers after adjusting the network topology using the above method.

[0105] The solution of embodiment 1 ensures the anti-interference ability of the formation under the normal operation of the intelligent body. Next, in order to cope with the persistent collapse and stability destruction of the formation network caused by the failure of key intelligent bodies or the failure of certain communication edges, this embodiment proposes a network reconstruction algorithm, which adds a small number of communication edges in the network according to the type of fault, so that the communication topology diagram after reconstruction is still persistent, thereby ensuring the stability of the formation.

[0106] Figure 7An example of the reconstruction of the three-dimensional communication network topology when the third leader intelligent agent node 3 fails is shown. Since the failed node 3 is the third leader, corresponding to step 4, firstly, node 3 and all the directed edges connected to it are deleted, and the follower 4 in the initial tetrahedron is changed to the new third leader. Then, the follower predecessor neighbor nodes 9 and 5 of the failed intelligent agent node 3 are found, and a non-repeating directed edge pointing to the first leader 1 is added to obtain a stable three-dimensional communication network topology diagram without the failed intelligent agent after reconstruction.

[0107] Figure 8 The effectiveness of Algorithm 3 is verified through simulation experiments. Figure 8 As shown in (a) in the figure, the intelligent agent system starts to move to the upper right after forming the initial formation (the motion trajectory is represented by a solid line). However, an intelligent agent node F suddenly fails and starts to fall. At this time, Algorithm 3 is triggered to reconstruct the communication network (the topology before and after reconstruction are represented by dotted lines). After that, the remaining intelligent agents can maintain the original formation and continue to move to the upper right. Figure 8 (b) in the figure further gives the distance error diagram between some intelligent agents during the movement. It can be seen that a fault occurs at 20 seconds. After the fault occurs, the faulty intelligent agent cannot maintain the expected distance with the normal intelligent agent, and the distance error becomes larger and larger. The remaining normal intelligent agents repair the stability of the network due to network reconstruction, thus preventing the formation from collapsing.

[0108] In general, the overall process of the multi-agent distributed three-dimensional formation control method proposed by the present invention under communication-constrained conditions is as follows: Fig. 9 As shown in the figure, Algorithm 1 determines the overall communication network topology of the system according to the target formation and the initial position state of the agent, specifies the neighbor nodes of each agent, and provides conditions for the subsequent distributed controller deployment. Based on the communication network graph constructed by Algorithm 1, Algorithm 2 deploys the formation control model in a distributed manner on each agent. This control model only requires a small amount of neighbor relative position information, has no unified requirements for the reference direction of each agent, and can ensure a certain convergence to the target formation position. Finally, based on the previous two parts, Algorithm 3 provides a solution to the multi-agent formation control problem with agent failures and communication edge failures. It locally modifies the original communication network to maintain the persistence of the directed network and ensure the overall stability of the multi-agent formation.

[0109] The present invention proposes a method for improving some potential stability problems of a formation control method based on a combination of "rigid network topology + distance controller" in a three-dimensional scene. The formation control method based on a combination of "rigid network topology + distance controller" belongs to a fully distributed formation control algorithm, because each formation member only needs to rely on its own measurement sensor to determine the relative positions of a few neighbor members to ensure the stability of the formation. However, this method has two disadvantages: (1) The construction of the rigid graph network topology is slow, and different edge combinations need to be continuously tried until the rigid matrix is ​​full rank; (2) The distance-based formation controller does not meet the global stability requirements in theory. Although it can meet the formation stability when the number of formation members is small and the disturbance is within a certain range, it cannot cope with complex situations and multiple formation formation switches. In response to problem (1), the present invention proposes to use a directed persistence graph generation method based on tetrahedron decomposition to directly generate a persistence graph (a directed form of a rigid graph) according to the approximate shape of the formation, so as to quickly determine the following relationship between formation members; in response to problem (2), an improved projection-based three-dimensional formation controller is used to replace all non-leader-based distance controllers. Through analysis and simulation experiments, it can be seen that the improved formation controller can ensure that all members have a unique equilibrium position in the formation after reaching a steady state (using a distance-based controller, each node has two equilibrium positions), that is, the improved formation controller can ensure almost global stability. Therefore, multiple formation formations can be switched by configuring multiple sets of hyperparameters. In addition, considering the possible member failures and edge interference problems in the process of maintaining a multi-agent formation, in order to ensure that the remaining normal agents are not disturbed in forming a formation, a matching formation stability repair algorithm is proposed based on the solution of the above problems. The main means is to use local reconstruction of the directed persistence graph, delete faulty members and interference edges, and ensure that the network persistence remains unchanged by adding as few other edges as possible, thereby maintaining formation stability.

[0110] Embodiment 3

[0111] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for constructing a multi-agent three-dimensional motion control model as described in the first embodiment above or the method for controlling a multi-agent distributed three-dimensional formation as described in the second embodiment above is implemented.

[0112] The relevant technical solution is the same as the previous embodiment and will not be described again here.

[0113] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed multi-agent three-dimensional motion control model construction method, characterized in that: Used to achieve target formations, including: According to the three-dimensional target formation of the multi-agent system to be controlled, a three-dimensional directed persistent network communication topology graph with the least edges is constructed, wherein the node types include the first leader, the second leader, the third leader and the follower; According to the node type of each agent, a three-dimensional motion control model is constructed, in which the input of the agent is the relative position information of all the successor neighbor agents of the agent relative to the agent, and the output is the motion speed of the agent; wherein the three-dimensional motion control model of each follower is a three-dimensional formation control model based on projection, and the construction method is: Based on the relative position vectors of the three successor neighbor intelligent agent nodes relative to the follower, two mutually orthogonal unit direction vectors located in the plane formed by the three successor neighbor intelligent agent nodes and a unit direction vector perpendicular to the plane are determined; one of the two mutually orthogonal unit direction vectors is parallel to the line connecting any two successor neighbors; the projections of the relative position vectors of any intelligent agent node among the three successor neighbor intelligent agent nodes on the three mutually orthogonal unit vectors tend to correspond to expected projection values ​​as three projection constraints, and the three-dimensional motion control model of the follower is obtained through weighted summation; the successor neighbor intelligent agent node is the end node of the directed edge corresponding to the follower, and the successor neighbor intelligent agent node and the expected projection value are both determined according to the three-dimensional directed persistent network communication topology graph with the least edges.

2. The method for constructing a multi-agent three-dimensional motion control model according to claim 1, characterized in that: The three-dimensional directed persistent network communication topology diagram is constructed as follows: Based on the basic communication topology conditions, the target formation is tetrahedron-divided to obtain an undirected tetrahedron-divided graph, and the intelligent node as the first leader, the intelligent node as the second leader, the intelligent node as the third leader and the intelligent node as the follower are determined from the tetrahedron-divided graph. According to the predecessor-successor neighbor relationship, some node connections in the tetrahedron-divided graph are directed and redundant edges are deleted to obtain a three-dimensional directed persistent network communication topology with the least edges; wherein the basic communication topology conditions are: the undirected graph corresponding to the topology has rigidity, and the directed graph corresponding to the topology has persistence.

3. The method for constructing a multi-agent three-dimensional motion control model according to claim 2, characterized in that: The three-dimensional directed persistent network communication topology diagram is constructed in the following manner: The target formation is partitioned by a tetrahedron partitioning algorithm to obtain an undirected tetrahedron partitioning graph G = (V, E), and a node set P and a directed edge set Q are established, which are initialized to an empty set; the node set V = {v1, v2, v3, …, v n } represents the set of agents in the multi-agent system to be controlled, and the edge set E = {e1, e2, e3, …, e m } represents all communication links between intelligent nodes, and the directed edge e=(v i ,v j ) represents the agent node v i With agent node v j One-way communication and v i Get v j Relative position information, v i Called v j The predecessor neighbor agent node, v j Called v i The successor neighbor agent node of According to the preset intelligent agent node v1 as the first leader, the tetrahedron structure including v1 is selected from the tetrahedron partition graph G = (V, E) as the initial tetrahedron, and any two intelligent agent nodes v2 and v3 other than the intelligent agent node v1 are selected from the initial tetrahedron as the second leader and the third leader respectively, and the intelligent agent node v4 other than v1, v2, and v3 in the initial tetrahedron is used as a follower; the intelligent agent nodes v1, v2, v3, and v4 corresponding to the four vertices of the initial tetrahedron are used as initial nodes, added to the node set P, and a directed edge e is constructed. 21 =(v2,v1),e 31 =(v3,v1),e 32 =(v3,v2),e 41 =(v4,v1),e 42 =(v4,v2) and e 43 =(v4,v3), added to the directed edge set Q; Search for an agent node v from the remaining node set V\P that has an edge with any three nodes in the current node set P i , connect the three edges with v i Direct the predecessor neighbor agent node, add the directed edge set Q, and convert v i Add the current node set P; repeat this step until the remaining node set V\P is empty, delete the non-directed connections, and obtain the network communication topology diagram G'=(P,Q).

4. The method for constructing a multi-agent three-dimensional motion control model according to claim 1, characterized in that: The three-dimensional motion control model of the second leader is a distance-based gradient control model, and the distance constraint it adopts is that the second leader is only subject to the distance constraint of the first leader; the three-dimensional motion control model of the third leader is a distance-based gradient control model, and the distance constraint it adopts is that the third leader is subject to the dual distance constraints of the first leader and the second leader.

5. The method for constructing a multi-agent three-dimensional motion control model according to claim 4, characterized in that: The three-dimensional motion control model of the second leader is specifically: In the formula, represents the movement speed of the second leader, x1 and x2 are the current position vectors of the first leader and the second leader respectively, x1* and x2* are the target position vectors of the first leader and the second leader in the target formation respectively, d represents the expected distance between the first leader and the second leader, the symbol ||x|| represents the length of the vector x, and c is a parameter used to control the formation speed of the target formation; The three-dimensional motion control model of the third leader is specifically: In the formula, represents the movement speed of the third leader, x3 is the current position vector of the third leader, x3* is the target position vector of the third leader in the target formation, d 13 , d 23 They represent the expected distances between the third leader and the first leader and the second leader in the target formation respectively.

6. The method for constructing a multi-agent three-dimensional motion control model according to claim 1, characterized in that: The three-dimensional motion control model of the follower is as follows: In the formula, Indicates follower v i The speed of movement; x j , x k and x m For followers v i The three successor neighbor agent nodes v j 、v k and v m The current position vector of the agent; ε is a sufficiently small positive constant; the symbol × represents the vector product multiplication in three-dimensional space; h1, h2 and h3 are all intermediate variables, and the intermediate variable h1 represents the agent node v j 、v k The unit relative direction vector between them, h2 represents the relative direction vector with respect to the agent node v j 、v k and v m The three form a unit direction vector that is orthogonal to the plane. h3 represents a unit vector that is orthogonal to h1 and h2. ij =x i -x j , indicating v i and v j The relative position of the superscript symbol T indicates transposition, x i For followers v i The current position vector, x j Indicates v j The current position vector of x i * , x j * , x k * , x m * v i 、v j 、v k and v m The target position vector in the target formation; s1, s2, s3 are the projection parameters in the target formation, determined according to the target formation; c is the parameter used to control the formation speed of the target formation.

7. A multi-agent distributed three-dimensional formation control method, characterized in that: The three-dimensional motion control model of each intelligent agent constructed by the multi-agent three-dimensional motion control model construction method as described in any one of claims 1 to 6 is used to adjust the movement speed of the corresponding intelligent agent, so that the formation formed by all intelligent agents in space is finally consistent with the target formation.

8. A multi-agent distributed three-dimensional formation control method according to claim 7, characterized in that: When an agent failure occurs or communication between agents is disturbed, the method further comprises: The faulty agent and the two agents whose communication edges are disturbed are regarded as faulty agents. In the order of the first leader, the second leader, the third leader, and the follower, the nodes corresponding to the faulty agents are removed from the three-dimensional directed persistent network communication topology with the least edges through the local network reconstruction algorithm, and the directed edges are re-added to ensure that the node removal does not weaken the stability and anti-interference of the formation, so as to achieve the adjustment of the three-dimensional directed persistent network communication topology. According to the adjusted three-dimensional directed persistent network communication topology graph, the projection parameters or distance parameters in the three-dimensional motion control model of each intelligent body are adjusted.

9. A multi-agent distributed three-dimensional formation control method according to claim 8, characterized in that: If the faulty intelligent agent is the first leader, the adjustment method of the three-dimensional directed persistent network communication topology graph is as follows: all directed edges connected to the intelligent agent node are deleted from the three-dimensional directed persistent network communication topology graph, the current second leader node is changed to a new first leader node, the third leader node is changed to a new second leader node, the follower node in the initial tetrahedron is changed to a new third leader node, and all follower predecessor neighbor intelligent agent nodes of the intelligent agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node; If the faulty agent is the second leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node remains unchanged, the current third leader node is changed to the new second leader node, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node; If the faulty agent is the third leader, the three-dimensional directed persistent network communication topology graph is adjusted as follows: the directed edge connected to the agent node is deleted from the three-dimensional directed persistent network communication topology graph, the current first leader node and the second leader node remain unchanged, the follower node in the initial tetrahedron is changed to the new third leader node, and all the follower predecessor neighbor agent nodes of the agent node are added with a non-repeating edge pointing to the new first leader node, the second leader node or the third leader node; If the faulty agent is a follower, the network communication topology is adjusted as follows: all predecessor neighbor agent nodes of the agent node in the original network communication topology constitute a set V1, and the successor neighbor agent nodes constitute a set V2. Each node of V1 is traversed in turn, and a non-neighbor agent node of the node in the original network communication topology is searched from V2. A directed edge starting from the V1 node is established, and all directed edges connected to the agent node are deleted.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a multi-agent three-dimensional motion control model construction method as described in any one of claims 1 to 6 or a multi-agent distributed three-dimensional formation control method as described in any one of claims 7 to 9.

Citation Information

Patent Citations

  • A Self-Organizing Cooperative Control Method for Multi-UAV Formation Based on Rigid Graph Theory

    CN107632614B

  • Formation control method based on complex Laplace matrix

    CN112558613B