Intelligent agent formation tracking and control methods and related products

Abstract

This invention discloses an intelligent agent formation tracking control method and related products. In the first stage, the unknown parameters of unknown interference are fully evaluated. In the second stage, even if the conditions for updating their own position information to other intelligent agents become stringent, the time interval between updates will not approach zero. Intelligent agent formation can still achieve formation tracking even under limited communication frequencies and unknown interference.

Description

TECHNICAL FIELD

[0001] The present application relates to an agent formation tracking control method and related products. BACKGROUND

[0002] The structure of a single unmanned aerial vehicle (UAV) is very simple, but multiple UAVs can accomplish many very complex tasks in cooperation, so UAV swarms have received extensive attention in recent years. Among them, formation control of UAV swarms is an important research content in the field of UAV swarms, and can play a key role in many military, national defense and livelihood fields such as agricultural irrigation, environmental detection, transportation logistics, surveillance coverage and intelligence collection.

[0003] However, many traditional UAV swarm formation control methods use centralized control methods, that is, there is a control center that can communicate with all UAVs in the system and control each UAV using the obtained data. Therefore, the centralized control method often needs a large amount of real-time communication and high computing performance as support. The distributed control method does not have a control center, and each UAV only needs to achieve its own control target according to its own information and the information available in the neighborhood, and the entire UAV swarm can exhibit corresponding collective behavior to achieve the global target. Because the distributed control method has the advantages of scalability, robustness, and reduced upgrade cost, it has become a research hotspot in recent years.

[0004] In distributed formation control, communication is a key indicator. In existing research on distributed formation control at home and abroad, many achievements have put forward higher requirements for communication between UAVs. For example, some research achievements require that the communication topology of the UAV swarm must be an undirected graph, that is, two UAVs must be able to communicate in both directions. Obviously, this communication method is not applicable when the UAV is executing a specific task with sensitive information. For example, some research achievements require continuous communication between UAV swarms, that is, a UAV must obtain real-time data from other UAVs to achieve the control target. However, this type of research achievement cannot be applied to scenarios where communication resources are extremely limited due to enemy interference in the battlefield. Therefore, how to reduce the number of communications between UAV swarms, ensure that the communication frequency is within a certain range, and relax the two-way communication condition to a one-way communication condition for the communication of the UAV swarm is an important research content.

[0005] In addition, the environmental disturbance that the UAVs may face in practice should also be considered. Relevant research shows that, unlike the adaptive control of a single UAV, the disturbance existing in the UAV cluster is coupled with the one-way communication between the UAVs, which may cause the UAV cluster to diverge and cause significant losses. However, this problem has been rarely studied. At the same time, in order to solve the impact of disturbance, the final tracking control effect of many research results can only achieve bounded tracking, and cannot achieve asymptotic tracking. Therefore, the problem of distributed adaptive consensus asymptotic tracking formation control under the disturbance environment has the value of in-depth research.

[0006] The above problems also apply to intelligent agents such as unmanned ships.

[0007] CN114879743A "Unmanned cluster distributed time-varying optimization control method and system considering disturbance", reduces the impact of disturbance information, designs a distributed time-varying optimization controller, and can realize zero-error tracking of the optimal trajectory of the unmanned cluster system, while meeting the global performance index optimization. SUMMARY

[0008] The present application provides an intelligent agent formation tracking control method and related products.

[0009] The present application provides the following technical solution: an intelligent agent formation tracking control method, the communication topology of the intelligent agent formation can be represented as a directed graph containing at least one directed spanning tree, the root node of the at least one directed spanning tree corresponds to an intelligent agent that knows the preset trajectory of a virtual leader, and the rest of the intelligent agents at least know part of the parameters of the preset trajectory of the virtual leader, the control method for any intelligent agent in the intelligent agent formation includes:

[0010] In the first phase, the following operations are performed:

[0011] According to the updated position information of the intelligent agent itself and its neighbor intelligent agents, part of the parameters of the preset trajectory of the virtual leader, and the preset formation parameters, the expected trajectory of the intelligent agent itself and its neighbor intelligent agents from the current position information update time to the next position update time of the intelligent agent itself and its neighbor intelligent agents is predicted;

[0012] According to the comparison result of the real-time position information of the intelligent agent itself and the expected trajectory of the intelligent agent itself, the position information of the intelligent agent itself is updated to the intelligent agents that can receive the communication information of the intelligent agent itself;

[0013] The motion trajectory of the intelligent agent itself, the control input of the intelligent agent itself, and the known function representing the unknown disturbance are respectively first-order low-pass filtered and then sampled at fixed time intervals to obtain the filtered sampling sequence of the motion trajectory of the intelligent agent itself, the filtered sampling sequence of the control input of the intelligent agent itself, and the filtered sampling sequence of the known function representing the unknown disturbance of the intelligent agent itself, wherein the unknown disturbance is described by a group of known functions and a group of unknown parameters.

[0014] evaluating the formation tracking error of the ego and its neighbors as a whole according to the real-time position information of the ego and its neighbors, the expected trajectory of each of the ego and its neighbors, the preset formation parameters, and the position information of the virtual leader, wherein the parameter of the position information of the virtual leader is removed if the position information of the virtual leader is unknown, obtaining an estimated value of the unknown parameter representing the unknown disturbance according to the control input of the ego, the filtered sampling sequence of the motion trajectory of the ego, the filtered sampling sequence of the control input of the ego, and the filtered sampling sequence of the known function representing the unknown disturbance of the ego, and estimating the unknown parameter representing the unknown disturbance according to the motion trajectory of the ego, and determining the control input of the ego according to the obtained formation tracking error, the estimated value of the unknown parameter representing the unknown disturbance, the position information of the ego, the formation parameters of the ego, and the known part of the preset trajectory of the virtual leader;

[0015] the first stage ends when a preset condition is met, and the preset condition is used to evaluate the accuracy of the evaluation of the unknown parameter representing the unknown disturbance;

[0016] the following operations are performed in a second stage after the first stage:

[0017] predicting the expected trajectory of the ego and its neighbors from the current position update time to the next position update time of the ego and its neighbors according to the updated position information of the ego and its neighbors, the known part of the preset trajectory of the virtual leader, and the preset formation parameters;

[0018] updating the position information of the ego to the agents capable of receiving the communication information of the ego according to the comparison result of the real-time position information of the ego and the expected trajectory of the ego, wherein the judgment standard of whether to update the position information of the ego to the agents capable of receiving the communication information of the ego in the second stage is more stringent than that in the first stage;

[0019] evaluating the formation tracking error of the ego and its neighbors as a whole according to the real-time position information of the ego and its neighbors, the expected trajectory of each of the ego and its neighbors, the preset formation parameters, and the position information of the virtual leader, wherein the parameter of the position information of the virtual leader is removed if the position information of the virtual leader is unknown, obtaining an estimated value of the unknown parameter representing the unknown disturbance according to the control input of the ego, the filtered sampling sequence of the motion trajectory of the ego at the end of the first stage, the filtered sampling sequence of the control input of the ego at the end of the first stage, and the filtered sampling sequence of the known function representing the unknown disturbance of the ego at the end of the first stage, and estimating the unknown parameter representing the unknown disturbance according to the motion trajectory of the ego, and determining the control input of the ego according to the obtained formation tracking error, the estimated value of the unknown parameter representing the unknown disturbance, the position information of the ego, the formation parameters of the ego, and the known part of the preset trajectory of the virtual leader.

[0020] The application provides the following technical solutions: a control device of an intelligent agent, the intelligent agent being an intelligent agent in an intelligent agent formation, a communication topology of the intelligent agent formation being represented as a directed graph containing at least one directed spanning tree, root nodes of the at least one directed spanning tree corresponding to intelligent agents known preset trajectories of virtual leaders, and the rest of the intelligent agents at least known part parameters of the preset trajectories of the virtual leaders, and the control device comprising:

[0021] a first control module, configured to perform the following operations in a first stage:

[0022] predicting an expected trajectory of the intelligent agent and its neighbor intelligent agents from a current position information updating time to a next position information updating time of the intelligent agent and its neighbor intelligent agents according to updated position information of the intelligent agent and its neighbor intelligent agents, part parameters of the preset trajectories of the virtual leaders, and preset formation parameters;

[0023] updating the position information of the intelligent agent to intelligent agents capable of receiving communication information of the intelligent agent according to a comparison result of real-time position information of the intelligent agent and the expected trajectory of the intelligent agent;

[0024] respectively performing first-order low-pass filtering on a motion trajectory of the intelligent agent, a control input of the intelligent agent and a known function representing unknown interference, and then sampling at a fixed time interval to obtain a filtered sampling sequence of the motion trajectory of the intelligent agent, a filtered sampling sequence of the control input of the intelligent agent and a filtered sampling sequence of the known function representing the unknown interference of the intelligent agent, wherein the unknown interference is described by a group of known functions and a group of unknown parameters;

[0025] evaluating a formation tracking error of the intelligent agent and its neighbor intelligent agents as a whole according to real-time position information of the intelligent agent and its neighbor intelligent agents, respective expected trajectories of the intelligent agent and its neighbor intelligent agents, preset formation parameters and position information of the virtual leader, wherein the parameter of the position information of the virtual leader is removed if the position information of the virtual leader is unknown, estimating unknown parameters representing the unknown interference of the intelligent agent according to the control input of the intelligent agent, the filtered sampling sequence of the motion trajectory of the intelligent agent, the filtered sampling sequence of the control input of the intelligent agent, the filtered sampling sequence of the known function representing the unknown interference of the intelligent agent and the motion trajectory of the intelligent agent, obtaining an estimated value of the unknown parameters representing the unknown interference, and determining the control input of the intelligent agent according to the obtained formation tracking error, the estimated value of the unknown parameters representing the unknown interference, the position information of the intelligent agent, the formation parameters of the intelligent agent and the known part parameters of the preset trajectories of the virtual leaders;

[0026] the first stage ends when a preset condition is met, and the preset condition is used to evaluate the accuracy of the evaluation of the unknown parameters representing the unknown interference;

[0027] a second control module, configured to perform the following operations in a second stage after the first stage:

[0028] Based on the updated location information of itself and its neighboring intelligent agents, some parameters of the virtual leader's preset trajectory, and preset formation parameters, predict the expected trajectory of itself and its neighboring intelligent agents from the current location information update time to the next location update time of itself and its neighboring intelligent agents.

[0029] Based on the comparison between its real-time location information and its expected trajectory, it updates its location information to the intelligent agent that can receive its communication information. The judgment criteria for whether to update its location information to the intelligent agent that can receive its communication information in the second stage are more stringent than those in the first stage.

[0030] The overall formation tracking error of itself and its neighboring agents is evaluated based on their real-time position information, expected trajectories, preset formation parameters, and the position information of the virtual leader. If the position information of the virtual leader is unknown, the parameter is removed. The unknown parameters expressing the unknown interference are estimated based on its own control input, the filtered sampling sequence of its own motion trajectory obtained at the end of the first stage, the filtered sampling sequence of its own control input obtained at the end of the first stage, the filtered sampling sequence of its own known function expressing the unknown interference obtained at the end of the first stage, and its own motion trajectory. The estimated values ​​of the unknown parameters expressing the unknown interference are obtained. Based on the obtained formation tracking error, the estimated values ​​of the unknown parameters expressing the unknown interference, its own position information, its own formation parameters, and some parameters of the known preset trajectory of the virtual leader, its own control input is determined.

[0031] This invention provides the following technical solution: a control device for intelligent agents, wherein the intelligent agents are intelligent agents in an intelligent agent formation, the communication topology of the intelligent agent formation can be represented as a directed graph containing at least one directed spanning tree, the intelligent agent corresponding to the root node of at least one directed spanning tree knows the preset trajectory of the virtual leader, and the other intelligent agents know at least some parameters of the preset trajectory of the virtual leader, the control device includes: a memory and a processor, the memory stores a program, and the processor runs the program to execute the aforementioned method.

[0032] The present invention provides the following technical solution: a computer program product that, when running on a processor, can execute the aforementioned intelligent agent formation tracking and control method.

[0033] Even if communication between agents is not bidirectional, and even if the virtual leader's preset trajectory is only fully disclosed to a limited number of agents, the agent formation can still achieve formation tracking under limited communication frequency and unknown interference. Attached Figure Description

[0034] Figure 1is a flow chart of the design process of the intelligent agent formation tracking control method of the present application.

[0035] Figure 2 is a directed topological graph of the intelligent agent formation communication connection in the embodiment of the present application.

[0036] Figure 3 is a position-time curve of the intelligent agent formation and a tracking error-time curve of the intelligent agent formation in the embodiment of the present application, wherein the vertical direction of the left side of the dotted line is the first stage, and the right side is the second stage.

[0037] Figure 4 is a top view of the intelligent agent formation motion trajectory in the 3D space in the embodiment of the present application, and t is time.

[0038] Figure 5 is the time of each intelligent agent sending communication data on each coordinate axis in the embodiment of the present application, and t is time.

[0039] Figure 6 is a structural schematic diagram of the control device of the intelligent agent in the embodiment of the present application.

[0040] Figure 7 is a structural schematic diagram of the control device of the intelligent agent in the embodiment of the present application. DETAILED DESCRIPTION

[0041] The present application will be further described below in combination with specific embodiments, but the protection scope of the present application is not limited thereto.

[0042] The design idea of the technical scheme of the present application is introduced below. Fig. 1 is a flow chart of the design process of the intelligent agent formation tracking control method of the present application. The intelligent agent formation takes a UAV cluster as an example. First, a kinematic model, a formation and a communication topology model of the UAV cluster need to be established; second, a position observer needs to be designed for each UAV; then, in the first stage, an event-triggered communication rule is designed for each UAV to reduce the communication times between the UAVs, and a first-order low-pass filter and a data collection rule as well as a distributed controller and an adaptive law are designed; in the second stage, an event-triggered communication rule as well as a distributed controller and an adaptive law are designed for each UAV. In actual UAV operation, the first stage is in the front, and the second stage is in the back.

[0043] The specific design process of the UAV cluster is shown below.

[0044] Firstly, a kinematic model, a formation and a communication topology model of the UAV cluster are established.

[0045] The kinematic model of the UAV cluster is established, specifically as follows:

[0046] ;

[0047] Where i = 1, 2, ..., N, represents the drone number. This represents the coordinates of the i-th drone in the cluster within the Earth coordinate system Oxyz. The black dot above the letter p indicates the derivative with respect to time. It is the control input received by the i-th UAV, through design The speed of the drone can be directly controlled. These are the disturbances experienced by the drone in the x, y, and z directions of the environment, and it is assumed that their values ​​can be represented by a set of known functions and unknown parameters.

[0048] ;

[0049] in, They are A known smooth nonlinear function of dimension, It is an unknown constant vector.

[0050] This invention discusses a leader-follower drone swarm formation model. That is, the drone swarm needs to track a virtual leader in a specific formation. All drones in the swarm are followers. The virtual leader is the target being tracked and is also the reference trajectory mentioned later. Not all followers can dynamically acquire the virtual leader's position information; only a subset of followers in the swarm (at least including the root node below) can acquire the virtual leader's position information. The virtual leader's position information is pre-defined. The dynamic representation of the virtual leader in the Earth coordinate system Oxyz is as follows:

[0051] ;

[0052] in, It represents the virtual leader's coordinates in the Earth coordinate system Oxyz, and f x f y f z It is a constant, g x g y g z It is a constant not greater than 0. When a drone swarm performs a mission, the dynamics of the reference trajectory are pre-designed by engineers according to the mission requirements; therefore, f... x f y f z g x g y g z It is also known and is not affected by the environment.

[0053] Establish formations for drone swarm systems.

[0054] The follower in the UAV cluster needs to track the reference trajectory of the virtual leader in a specific formation, and the known formation parameters are designed in advance for each UAV . Wherein, is a constant. Finally, for the UAV i, the achieved formation effect should be:

[0055] .

[0056] A communication topology model is established for the UAV cluster system.

[0057] The communication topology relationship between UAVs can be described by a directed graph containing N nodes , wherein is a node set, is a set of edges between different UAVs. It means that the UAV j can obtain the information of the UAV i, and the UAV i is called a neighbor of the UAV j. The neighbor set of the UAV i is defined as . It should be noted that the present application considers the case of communication topology as a directed graph, so that the UAV j can obtain the information of the UAV i, which does not mean that the UAV i can obtain the information of the UAV j. At the same time, the case of (i, i) is not considered, so . According to the above communication topology relationship, the adjacency matrix is defined, wherein if , then , otherwise . The adjacency matrix is a description of the communication topology between the UAVs.

[0058] The in-degree matrix of the communication topology is defined as a diagonal matrix, and the diagonal elements are . The in-degree matrix describes how many position information of UAVs can be directly received by each UAV i.

[0059] Then the Laplacian matrix of the directed graph is defined as . The Laplacian matrix integrates the communication topology and the in-degree information of the UAV cluster.

[0060] A directed path from the UAV i to the UAV j is a sequence of consecutive edges . If a directed graph has a node, and this node has a directed path to any other node in the directed graph, then the directed graph contains a directed spanning tree, and this node is called the root node. In the present application, it is assumed that the communication topology of the UAV cluster is a directed graph containing a directed spanning tree, and at least one root node can obtain the real-time information of the virtual leader , and the root node is This means that drone i can directly obtain the virtual leader's real-time location information without relying on other drones. This indicates that drone i cannot obtain the real-time location information of the virtual leader. And denote the matrix... Matrix B describes whether drones have the authority to directly obtain the real-time location information of the virtual leader.

[0061] Define column vectors based on the communication topology diagram. and diagonal matrix And the matrix can be calculated. Used for subsequent design and analysis. Column vector q, diagonal matrix P, and matrix Q are intermediate quantities used for controller design. All three contain information about the communication topology, in-degree, and permissions of the UAV swarm.

[0062] Design event-triggered communication rules to reduce the communication resource consumption of the drone swarm and ensure that the interval between each trigger is greater than a certain normal number; design filters and collect data, and design the drone swarm controller and adaptive law based on the collected data to estimate unknown parameters in drone interference, thereby realizing distributed consensus formation tracking of the virtual leader by the drone swarm.

[0063] Design a position observer. For example, a position observer subroutine or process would run on each drone.

[0064] The location observer can be affected by data transmitted during communications by neighboring drones. satisfy It is the time sequence of drone i transmitting its current x-axis direction information to neighboring drones containing drone i (the time when drone i updates its own x-axis position information). It is the initial moment. Representing the first drone Next, transmit the current x-axis position data of drone i to other neighboring drones containing drone i. The moment when drone i receives location data from its neighboring drones. Only at any moment Update it, and its value At any moment It will remain unchanged. Let be the neighbor set of drone i. Similarly, the above operations can be performed in the y and z axes as well.

[0065] remember satisfy It is the time sequence of drone i transmitting its current y-axis direction information to the outside world. It is the initial moment. represents the time when UAV i transmits its current position data for the first time For UAV i, the position data it receives from its neighbors will only be updated at time and its value will remain unchanged at time

[0066] is the sequence of time when UAV i transmits its current z-axis direction information for the first time is the initial time, represents the time when UAV i transmits its current position data for the first time For UAV i, the position data it receives from its neighbors will only be updated at time and its value will remain unchanged at time

[0067] For UAV i, it needs to design a position observer for itself and all its neighbors, the communication between x-axes is shown as follows:

[0068] .

[0069] The first row of equations describes the dynamic differential equation of the position observer in each trigger interval, and the second row of equations describes the initial value of the position observer in each trigger interval.

[0070] The communication between y-axes is shown as follows:

[0071] .

[0072] The first row of equations describes the dynamic differential equation of the position observer in each trigger interval, and the second row of equations describes the initial value of the position observer in each trigger interval.

[0073] The communication between z-axes is shown as follows:

[0074] .

[0075] The first row of equations describes the dynamic differential equation of the position observer in each trigger interval, and the second row of equations describes the initial value of the position observer in each trigger interval.

[0076] ​​​​​​​​​​The design of the position observer can prepare for the subsequent implementation of the asymptotic tracking formation control of UAV swarm.

[0077] For a single UAV, the input of the position observer running on it is the position information at each triggering time , and the output of the position observer is The position observer describes the expectation of the real-time coordinates of UAV i and its neighbor UAVs when they are tracking the virtual leader according to the expected formation spatial layout in this time interval

[0078] In the first stage, the event-triggered communication rule is designed.

[0079] In order to reduce the number of communications, in the x-axis, we set the current position value of UAV i If the difference between the current position value of UAV i and the value of the position observer is too large, UAV i will only transmit the information at the current time to the neighbor UAVs containing UAV i.

[0080] The event-triggered communication rule in the x-axis is set as follows:

[0081] ;

[0082] Wherein, is an adjustable parameter, which is greater than 0, and is a set value in actual operation. The larger it is, the fewer the number of communications is, but the worse the control performance is, The smaller it is, the more the number of communications is, but the better the control performance is.

[0083] The event-triggered communication rule in the y-axis is set as follows:

[0084] ;

[0085] Wherein, is a preset adjustable parameter, which is greater than 0, and is a set value in actual operation.

[0086] The event-triggered communication rule in the z-axis is set as follows:

[0087] ;

[0088] Wherein, is an adjustable parameter, which is greater than 0, and is a set value in actual operation.

[0089] The design of the event-triggered communication rule can effectively reduce the number of communications between UAVs.

[0090] ​​When the event-triggered communication rule is running, the input information is the position of the UAV in three directions and the value of the position observer in three directions, and the output information is the time of the next communication event in the corresponding direction. The running principle of the event communication rule is that only when the position information of the UAV i and the expected coordinate position difference of the UAV i and its neighbor UAVs in the ideal state according to the formation tracking of the virtual leader are large enough, the UAV i will send the current coordinate position of the UAV i to the UAV that can receive the information of the UAV i.

[0091] In the first stage, a first-order low-pass filter and a data collection method are designed.

[0092] In the x-axis direction, the real-time coordinates of the UAV, the control input, and the known function in the expression of the unknown disturbance are expressed as A first-order low-pass filter is designed respectively, and the filtered signal is denoted as The specific design of the filter is as follows:

[0093] ;

[0094] Where k is an arbitrary adjustable parameter, which is greater than 0, and is a constant in actual operation. The above filter constructs a differential equation in each direction and assigns an initial value to it.

[0095] In this stage, data collection is performed at a certain time interval according to the filtered signal

[0096] , a total of m times.

[0097] The parameter m can be set by oneself, and can be as large as possible to ensure that the amount of collected data is sufficient, and it is necessary to ensure that the following formula is satisfied:

[0098] .

[0099] That is, the matrix is row full rank, which means that the matrix is positive definite, so that a strict lower bound can be achieved between any two communication event triggers in the determined direction within the UAV cluster. The filtered signal refers to the filtered signal, that is, .

[0100] After m times of data collection, the matrix is calculated:

[0101] .

[0102] This matrix is an intermediate quantity obtained in the controller design, which is a positive definite matrix and has good mathematical properties.

[0103] Similarly, in the y-axis direction, the signal is filtered by a first-order filter, and the filtered signal is denoted as The specific design of the filter is as follows:

[0104] .

[0105] At this stage, data collection is performed at a certain time interval based on the filtered signal:

[0106] Data is collected together with the x-axis, a total of m times.

[0107] In principle, whether the time of collecting data in the x-direction and the y-direction is the same does not affect the final result, as long as the previous equation holds, and the following equation must be satisfied: .

[0108] After m times of data collection, the matrix

[0109] is calculated.

[0110] In the z-axis direction, the signal is filtered by a first-order low-pass filter, and the filtered signal is denoted as The specific design of the filter is as follows:

[0111] .

[0112] The value of k is as small as possible for better effect, generally around 0.05, other values greater than 0 are also acceptable. This filter avoids acquiring the derivative value of the UAV position , as this value is difficult to obtain in practice. After adopting the filter design, the derivative value of the UAV position can be calculated by

[0113] ,

[0114] to avoid directly obtaining the derivative value of the UAV position.

[0115] At this stage, data collection is performed at a certain time interval based on the filtered signal: Data is collected together with the x-axis, a total of m times.

[0116] And the following equation must be satisfied: .

[0117] After m times of data collection, the following matrix is calculated:

[0118] .

[0119] The above data collection rule is designed to ensure that there is a strict lower bound for the trigger interval of each UAV, which is not achieved by other existing methods.

[0120] In the above steps, the real-time coordinates of the UAVs, the control inputs, and the known functions in the expression of the unknown disturbance are low-pass filtered, and then sampled at fixed time intervals according to the filtered signals. As long as the matrix is row full rank, the system can converge.

[0121] As long as the matrix is row full rank, the system can converge. As long as enough data is collected, this matrix can be row full rank. Generally, only in the case of very special forms of disturbance, more data cannot be row full rank, and such special cases are difficult to occur in actual engineering. If the matrix is not row full rank, the time of the first stage is extended until the matrix is row full rank.

[0122] In the first stage, the controller and the adaptive law of the UAV cluster in this stage are designed.

[0123] In the x-axis direction, the local consensus error based on event-triggered communication is first defined as follows:

[0124] ;

[0125] The physical meaning of the local consensus error is to represent the gap between the UAV and its neighbor UAVs in achieving position consensus. The smaller the value, the more consistent the positions of the UAV and its neighbor UAVs.

[0126] Then, the controller for each UAV i is designed as follows:

[0127] ;

[0128] wherein is a parameter estimator for estimating unknown parameters, and the adaptive law is designed as follows:

[0129] .

[0130] c is a preset positive coefficient. The physical meaning of the output value of the two parameter estimators is to estimate the unknown parameters present in the disturbance.

[0131] The above two parameter estimators ensure that the entire UAV cluster can also asymptotically achieve formation tracking control in the presence of disturbance, and ensure that there is a strict lower bound for the trigger interval. This cannot be achieved by other existing methods. The input information of the parameter estimator is the value of the controller at the current time and the value of the filtered signal collected , and the output information is the parameter estimator .

[0132] The parameter estimator needs to note that this stage is still in the first stage, so for the following expression: Only the collected data is summed up, and there is no need to sum up m times of data.

[0133] In the y-axis direction, first define the local consistency error based on event-triggered communication as follows:

[0134] .

[0135] Then design the controller for each UAV as follows:

[0136] ;

[0137] where, is the parameter estimator for estimating unknown parameters, and its adaptive law is designed as follows:

[0138] .

[0139] c is a preset positive coefficient. The physical meaning of the output value of the two parameter estimators is to estimate the unknown parameters existing in the disturbance.

[0140] The parameter estimator needs to note that this stage is still in the first stage, so for the following expression: Only the collected data is summed up, and there is no need to sum up m times of data.

[0141] In the z-axis direction, first define the local consistency error based on event-triggered communication as follows:

[0142] .

[0143] Then design the controller for each UAV i as follows:

[0144] ;

[0145] where, is the parameter estimator for estimating unknown parameters, and its adaptive law is designed as follows:

[0146] .

[0147] c is a preset positive coefficient. The physical meaning of the output value of the two parameter estimators is to estimate the unknown parameters existing in the disturbance.

[0148] The parameter estimator needs to note that this stage is still in the first stage, so the following expression:

[0149] This only needs to sum all the data collected, and does not need to sum m times the data. It can also be considered that the first stage m represents the number of data points that have been obtained so far.

[0150] The adaptive law can effectively deal with unknown parameters in environmental interference, effectively solve the coupling problem between unknown parameters in interference and one-way communication between UAVs. It ensures that the entire UAV cluster can also asymptotically realize formation tracking control in the presence of interference, and ensures that the trigger interval has a strict lower bound.

[0151] After the first stage, in the second stage, the controller and adaptive law of the UAV cluster in this stage and the event-triggered communication rule are designed.

[0152] The controller of each UAV i in each direction is designed as follows:

[0153]

[0154]

[0155] ;

[0156] wherein, , and The parameter estimator for estimating unknown parameters is designed as follows:

[0157]

[0158]

[0159] .

[0160] It should be noted that this stage is in the second stage, so , and The three items need to sum all the m times data collected in the first stage.

[0161] And the event-triggered communication rule in the second stage is quite different from the first stage. The event-triggered threshold is changed from a constant to an exponential decay function at this moment, as follows:

[0162]

[0163]

[0164] ;

[0165] wherein, is a constant that can be adjusted, greater than 0, and in actual operation, is a constant set. The greater, the less the number of communications, but the worse the control performance, The smaller, the more the number of communications, but the better the control performance. The selection of the parameter needs to meet the following conditions:

[0166]

[0167]

[0168]

[0169] wherein, respectively represent the minimum eigenvalue and the maximum eigenvalue of the matrix.

[0170] indicates the decay rate of the event-triggered communication threshold, The smaller, the slower the decay of the event-triggered communication threshold, and the less the number of communications; The greater, the faster the decay of the event-triggered communication threshold, and the more the number of communications. But if the event-triggered communication threshold decays too fast, faster than the convergence speed of the UAV swarm to achieve the tracking task, then there will be no strict lower bound for the triggering interval of each UAV, and the triggering interval may tend to zero, which is not suitable in actual work. Therefore cannot be greater than a critical value, which is obtained through theoretical analysis. The second stage no longer collects data because the first stage has collected sufficient data. Because the data collected in the first stage makes the information of the UAV's evaluation of unknown disturbances more sufficient, by the second stage, even if there is an exponential decay term, the actual communication frequency will not tend to 0.

[0171] Using an exponential decay function as the threshold of the event-triggered communication rule can make the UAV swarm eventually achieve good results in asymptotic tracking formation control, while making the time interval between the adjacent two times of information transmission of each UAV strictly greater than a constant that does not change with time.

[0172] The event-triggered communication rule in the second stage is more stringent than in the first stage, and the way is not limited to this. For example, in the second stage, in the x-axis direction, when the absolute value of the difference between the position of itself and the position defined by the expected trajectory of itself is greater than a positive number smaller than h x , the agent that can receive its communication information updates its position information in that coordinate axis direction.

[0173] In summary, after the above steps are taken, the UAV cluster can track the preset reference signal in a specific formation in an environment containing interference, thereby completing task execution.

[0174] Based on the above design idea, the embodiment of the application substantially provides an agent formation tracking control method. From the perspective of a program, the execution subject of the agent formation tracking control method of the embodiment of the application can be a computer program. From the perspective of equipment, the execution subject of the control method can be a processor carrying these computer programs, or any hardware circuit such as a dedicated integrated circuit or a programmable logic device running the agent formation tracking control method.

[0175] The communication topology of the agent formation can be represented as a directed graph containing at least one directed spanning tree, the root node of the at least one directed spanning tree corresponding to an agent that knows the preset trajectory of the virtual leader, and the remaining agents knowing at least part of the parameters of the preset trajectory of the virtual leader. The control method for any agent in the agent formation includes:

[0176] In the first stage, the following operations are performed:

[0177] According to the updated position information of the agent itself and its neighbor agents, part of the parameters of the preset trajectory of the virtual leader, and the preset formation parameters, the expected trajectory of the agent itself and its neighbor agents from the current position information update time to the next position update time of the agent itself and its neighbor agents is predicted;

[0178] According to the comparison result of the real-time position information of the agent itself and the expected trajectory of the agent itself, the position information of the agent itself is updated to the agents that can receive the communication information of the agent itself;

[0179] The motion trajectory of the agent itself, the control input of the agent itself, and the known function representing the unknown disturbance are respectively first-order low-pass filtered and then sampled at fixed time intervals to obtain a filtered sampling sequence of the motion trajectory of the agent itself, a filtered sampling sequence of the control input of the agent itself, and a filtered sampling sequence of the known function representing the unknown disturbance of the agent itself, wherein the unknown disturbance is described by a set of known functions and a set of unknown parameters.

[0180] According to the real-time position information of the ego and its neighbor agents and their respective expected trajectories, the preset formation parameters and the position information of the virtual leader, evaluate the formation tracking error of the ego and its neighbor agents as a whole, wherein if the position information of the virtual leader is unknown, remove this parameter, according to the control input of the ego, the filtered sampling sequence of the motion trajectory of the ego, the filtered sampling sequence of the control input of the ego and the filtered sampling sequence of the known function expressing unknown disturbance of the ego, estimate the unknown parameters expressing unknown disturbance of the motion trajectory of the ego, obtain the estimated value of the unknown parameters expressing unknown disturbance, and determine the control input of the ego according to the obtained formation tracking error and the estimated value of the unknown parameters expressing unknown disturbance, the position information of the ego, the formation parameters of the ego and part of the known preset trajectory of the virtual leader.

[0181] The first stage ends when the preset condition is met, which is used to evaluate the accuracy of the unknown parameters expressing unknown disturbance.

[0182] The following operations are performed in the second stage after the first stage:

[0183] According to the updated position information of the ego and its neighbor agents, part of the preset trajectory of the virtual leader and the preset formation parameters, predict the expected trajectories of the ego and its neighbor agents from the current position information update time to the next position update time of the ego and its neighbor agents.

[0184] According to the comparison result of the real-time position information of the ego and its expected trajectory, update the position information of the ego to the agents that can receive its communication information, wherein the judgment standard of whether to update the position information of the ego to the agents that can receive its communication information in the second stage is more stringent than that in the first stage.

[0185] According to the real-time position information of the ego and its neighbor agents and their respective expected trajectories, the preset formation parameters and the position information of the virtual leader, evaluate the formation tracking error of the ego and its neighbor agents as a whole, wherein if the position information of the virtual leader is unknown, remove this parameter, according to the control input of the ego, the filtered sampling sequence of the motion trajectory of the ego at the end of the first stage, the filtered sampling sequence of the control input of the ego at the end of the first stage and the filtered sampling sequence of the known function expressing unknown disturbance of the ego at the end of the first stage, estimate the unknown parameters expressing unknown disturbance of the motion trajectory of the ego, obtain the estimated value of the unknown parameters expressing unknown disturbance, and determine the control input of the ego according to the obtained formation tracking error and the estimated value of the unknown parameters expressing unknown disturbance, the position information of the ego, the formation parameters of the ego and part of the known preset trajectory of the virtual leader.

[0186] For example, taking the earth rectangular coordinate system Oxyz as a reference, in the first stage, when the absolute value of the difference between the self position and the position defined by the self expected trajectory is greater than a first preset value in any coordinate axis direction, the agent that can receive its communication information is updated with the position information of the agent in the coordinate axis direction;

[0187] Taking the earth rectangular coordinate system as a reference, in the second stage, when the absolute value of the difference between the self position and the position defined by the self expected trajectory is greater than an exponentially decaying value of a second preset value over time, the agent that can receive its communication information is updated with the position information of the agent in the coordinate axis direction.

[0188] Based on the same inventive concept, referring to Figure 6 The application also provides a control device of an agent, the agent being an agent in an agent formation, a communication topology of the agent formation being represented as a directed graph containing at least one directed spanning tree, a root node of the at least one directed spanning tree corresponding to an agent known preset trajectory of a virtual leader, and the rest of the agents at least known partial parameters of the preset trajectory of the virtual leader, the control device comprising:

[0189] The first control module is configured to perform the following operations in the first stage:

[0190] According to the updated position information of the agent and its neighbor agents, the partial parameters of the preset trajectory of the virtual leader, and the preset formation parameters, the expected trajectory of the agent and its neighbor agents from the current position information update time to the next position update time of the agent and its neighbor agents is predicted;

[0191] According to the comparison result of the real-time position information of the agent and the expected trajectory of the agent, the position information of the agent is updated to the agent that can receive its communication information;

[0192] The motion trajectory of the agent, the control input of the agent, and the known function representing the unknown disturbance are respectively first-order low-pass filtered and then sampled at fixed time intervals to obtain a filtered sampling sequence of the motion trajectory of the agent, a filtered sampling sequence of the control input of the agent, and a filtered sampling sequence of the known function representing the unknown disturbance of the agent, wherein the unknown disturbance is described by a set of known functions and a set of unknown parameters.

[0193] evaluate the formation tracking error of the ego and its neighbor agents as a whole according to real-time position information of the ego and its neighbor agents, respective expected trajectories, preset formation parameters, and position information of the virtual leader, wherein the parameter of the position information of the virtual leader is removed if the position information of the virtual leader is unknown, obtain an estimated value of the unknown parameter representing unknown disturbance according to a filtered sampling sequence of the motion trajectory of the ego, a filtered sampling sequence of the control input of the ego, a filtered sampling sequence of the known function representing unknown disturbance of the ego, and an unknown parameter representing unknown disturbance estimated by the motion trajectory of the ego, and determine the control input of the ego according to the obtained formation tracking error, the estimated value of the unknown parameter representing unknown disturbance, position information of the ego, formation parameters of the ego, and part of the preset trajectory of the known virtual leader;

[0194] the first stage ends when a preset condition is met, and the preset condition is used to evaluate the accuracy of the evaluation of the unknown parameter representing unknown disturbance;

[0195] a second control module, configured to perform the following operations in a second stage after the first stage:

[0196] predict the expected trajectories of the ego and its neighbor agents from a current position information update time to a next position update time of the ego and its neighbor agents according to updated position information of the ego and its neighbor agents, part of the preset trajectory of the virtual leader, and preset formation parameters;

[0197] update the position information of the ego to agents capable of receiving communication information of the ego according to a comparison result of real-time position information of the ego and the expected trajectory of the ego, wherein the judgment standard for whether to update the position information of the ego to agents capable of receiving communication information of the ego in the second stage is more stringent than that in the first stage;

[0198] evaluate the formation tracking error of the ego and its neighbor agents as a whole according to real-time position information of the ego and its neighbor agents, respective expected trajectories, preset formation parameters, and position information of the virtual leader, wherein the parameter of the position information of the virtual leader is removed if the position information of the virtual leader is unknown, obtain an estimated value of the unknown parameter representing unknown disturbance according to a filtered sampling sequence of the motion trajectory of the ego, a filtered sampling sequence of the control input of the ego, a filtered sampling sequence of the known function representing unknown disturbance of the ego, and an unknown parameter representing unknown disturbance estimated by the motion trajectory of the ego, and determine the control input of the ego according to the obtained formation tracking error, the estimated value of the unknown parameter representing unknown disturbance, position information of the ego, formation parameters of the ego, and part of the preset trajectory of the known virtual leader.

[0199] The above modules can be implemented by software, hardware, or a combination of both.

[0200] refer to Figure 7 The present invention also provides a control device for intelligent agents, wherein the intelligent agents are intelligent agents in an intelligent agent formation, the communication topology of the intelligent agent formation can be represented as a directed graph containing at least one directed spanning tree, the intelligent agent corresponding to the root node of at least one directed spanning tree knows the preset trajectory of the virtual leader, and the other intelligent agents know at least some parameters of the preset trajectory of the virtual leader. The control device includes: a memory and a processor, the memory stores a program, and the processor runs the program to execute the aforementioned method.

[0201] The present invention also provides a computer program product that executes the aforementioned method when running on a processor.

[0202] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention is provided through specific embodiments.

[0203] Consider a scenario where there are 3 drones with the following dynamics:

[0204] .

[0205] A swarm of drones needs to perform a mission in a scenario with constant-speed, high-wind interference. We currently only know that the wind is constant in all directions, but the specific wind speed is unknown. Therefore, the interference term can be represented as:

[0206] .

[0207] That is, in the above scheme All are taken as one-dimensional 1, and the unknown parameters are also one-dimensional.

[0208] All three drones are currently on the ground, and their initial positions are as follows: It needs to be formed into an equilateral triangle (with pre-set formation parameters). Encircle a fixed target point at a certain height. ), that is, the parameters of the leader's reference trajectory are However, considering the sensitive nature of the target location information, it can only be obtained in real time by our No. 1 drone. Drones No. 2 and No. 3 are partner drones, used only for collaborative mission execution, and cannot directly acquire target location information. Furthermore, data communication between the three drones is not bidirectional; their communication topology is shown in the diagram below. Figure 2 As shown in the table below, other parameter settings involved in the technical solution are also shown in the table below.

[0209] Parameter Value Parameter Value 3 0.5 5 k 0.05 1 0.05 0.1s m 20 c 4 0

[0210] It should be noted that, The simulation is the wind speed, but the value is unknown for the UAV.

[0211] The above scenario is simulated by using MATLAB simulation software, the simulation step is 10 -4 s, and the total simulation time is 15s. After the first stage, the matrix is calculated, and the parameters are selected as shown in the following table.

[0212] Parameter Value 0.4934 0.4934 0.4934

[0213] The simulation effect is shown in Figures 3 to 5 . Figure 3 The reaction is the position change of three UAVs in x, y and z axes and the tracking error, and it can be seen that in the second stage, the tracking error of each UAV asymptotically converges to 0.

[0214] Figure 4 The reaction is the top view of the motion trajectory of the three UAVs in the 3D space, and it can be seen that the positions of the UAVs finally form a regular triangle and surround the target point [7, 7, 0] T .

[0215] Figure 5 The reaction is the communication transmission time of the UAV in x, y and z axes, and it can be seen that the communication amount is greatly reduced, which reflects the advantage of the present application in saving communication resources compared with other applications. The specific situation of the communication times is shown in the following table.

[0216] Direction\drone Drone 1 Drone 2 Drone 3 X-axis 68 56 71 Y-axis 47 46 52 Z-axis 41 34 38

[0217] In summary, the present application can be applied not only to UAV clusters containing various types of interference in system dynamics, but also to scenarios where UAVs cannot communicate bidirectionally. The distributed consensus tracking algorithm proposed in the present application can still achieve progressive tracking formation control of the reference trajectory by the entire UAV cluster even if only part of the UAVs can obtain the reference trajectory. At the same time, the event-triggered communication rule proposed in the present application can ensure that the time interval between the adjacent two triggers of each UAV is strictly greater than a normal number that does not change with time, greatly reducing the communication amount between the UAV cluster and saving the communication resources of the UAV cluster.

[0218] Each embodiment in the present application is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly explains the difference from other embodiments.

[0219] The scope of the present application is not limited to the above-described embodiments, and it will be apparent to those skilled in the art that various changes and modifications can be made thereto without departing from the scope and spirit of the present application. If these changes and modifications fall within the scope of the claims and their equivalents, they are also intended to be embraced by the present application.

Claims

1. An agent formation tracking control method, characterized in that, The communication topology of the intelligent agent formation can be represented as a directed graph containing at least one directed spanning tree, the root node of the at least one directed spanning tree corresponding to an intelligent agent known as a virtual leader, and the rest of the intelligent agents at least known as partial parameters of the preset trajectory of the virtual leader, and the control method for any intelligent agent in the intelligent agent formation comprises the following steps: In the first stage, the following operations are performed: According to the updated position information of the intelligent agent itself and its neighbor intelligent agents, the partial parameters of the preset trajectory of the virtual leader, and the preset formation parameters, the expected trajectory of the intelligent agent itself and its neighbor intelligent agents from the current position information update time to the next position update time of the intelligent agent itself and its neighbor intelligent agents is predicted; According to the comparison result of the real-time position information of the intelligent agent itself and the expected trajectory of the intelligent agent itself, the position information of the intelligent agent itself is updated to the intelligent agents that can receive the communication information of the intelligent agent itself; The motion trajectory of the intelligent agent itself, the control input of the intelligent agent itself, and the known function representing the unknown disturbance are respectively subjected to first-order low-pass filtering and then sampled at fixed time intervals to obtain the filtered sampling sequence of the motion trajectory of the intelligent agent itself, the filtered sampling sequence of the control input of the intelligent agent itself, and the filtered sampling sequence of the known function representing the unknown disturbance of the intelligent agent itself, wherein the unknown disturbance is described by a group of known functions and a group of unknown parameters; According to the real-time position information of the intelligent agent itself and its neighbor intelligent agents, the expected trajectory of each, the preset formation parameters, and the position information of the virtual leader, the overall formation tracking error of the intelligent agent itself and its neighbor intelligent agents is evaluated, wherein if the position information of the virtual leader is unknown, the parameter is removed, the unknown parameters representing the unknown disturbance are estimated based on the control input of the intelligent agent itself, the filtered sampling sequence of the motion trajectory of the intelligent agent itself, the filtered sampling sequence of the control input of the intelligent agent itself, and the filtered sampling sequence of the known function representing the unknown disturbance of the intelligent agent itself, and the motion trajectory of the intelligent agent itself, to obtain the estimated value of the unknown parameters representing the unknown disturbance, and the control input of the intelligent agent itself is determined based on the obtained formation tracking error and the estimated value of the unknown parameters representing the unknown disturbance, the position information of the intelligent agent itself, the formation parameters of the intelligent agent itself, and the known partial parameters of the preset trajectory of the virtual leader; The first stage ends when a preset condition is met, and the preset condition is used to evaluate the accuracy of the evaluation of the unknown parameters representing the unknown disturbance; In the second stage after the first stage, the following operations are performed: According to the updated position information of the intelligent agent itself and its neighbor intelligent agents, the partial parameters of the preset trajectory of the virtual leader, and the preset formation parameters, the expected trajectory of the intelligent agent itself and its neighbor intelligent agents from the current position information update time to the next position update time of the intelligent agent itself and its neighbor intelligent agents is predicted; According to the comparison result of the real-time position information of the intelligent agent itself and the expected trajectory of the intelligent agent itself, the position information of the intelligent agent itself is updated to the intelligent agents that can receive the communication information of the intelligent agent itself, wherein the judgment standard for whether the position information of the intelligent agent itself is updated to the intelligent agents that can receive the communication information of the intelligent agent itself in the second stage is more stringent than that in the first stage. According to real-time position information of the agent and its neighbor agents, and respective expected trajectories, preset formation parameters, and position information of a virtual leader, an overall formation tracking error of the agent and its neighbor agents is evaluated, wherein if the position information of the virtual leader is unknown, the parameter is removed, according to a control input of the agent, a filtered sampling sequence of a motion trajectory of the agent obtained at the end of the first stage, a filtered sampling sequence of the control input of the agent obtained at the end of the first stage, and a filtered sampling sequence of a known function representing unknown disturbance of the agent obtained at the end of the first stage, an unknown parameter representing unknown disturbance of the motion trajectory of the agent is estimated, and an estimated value of the unknown parameter representing unknown disturbance is obtained, and according to the obtained formation tracking error and the estimated value of the unknown parameter representing unknown disturbance, position information of the agent, formation parameters of the agent, and part of parameters of a preset trajectory of the virtual leader known, a control input of the agent is determined; With the rectangular coordinate system of the earth Oxyz as a reference, the preset conditions for the end of the first stage include: In the x-axis direction, the following is satisfied: ， wherein subscript i is the number of the agent, subscript x represents the x-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to To describe the data points of the filtered sampling sequence of the known function of the unknown disturbance in the x-axis direction, the position disturbance d i,x in the x direction is expressed as ; In the y-axis direction, the following is satisfied: ， wherein subscript i is the number of the agent, subscript y represents the y-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to To describe the data points of the filtered sampling sequence of the known function of the unknown disturbance in the y-axis direction, the position disturbance d i,y in the y direction is expressed as ; In the z-axis direction, the following is satisfied: ， wherein subscript i is the number of the agent, subscript z represents the z-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to To describe the data points of the filtered sampling sequence of the known function of the unknown disturbance in the z-axis direction, the position disturbance d i,z in the z direction is expressed as ; wherein, are respectively a known smooth nonlinear function of v, is an unknown constant vector, p i,x is a coordinate in the x-axis direction of itself, p i,y is a coordinate in the y-axis direction of itself, p i,z is a coordinate in the z-axis direction of itself; In the first stage and the second stage, the control input of the agent in the x-axis direction is represented as: ； wherein u i,x is the control input in x direction, subscript i is the agent number, subscript x represents the x-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the x-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the x direction, and the sum represents the estimation of the unknown parameters; In the first stage and the second stage, the control input of the agent in the y-axis direction is represented as: ； wherein u i,y is the control input in y direction, subscript i is the agent number, subscript y represents the y-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the y-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the y direction, and the sum represents the estimation of the unknown parameters; In the first stage and the second stage, the control input of the agent in the z-axis direction is represented as: ； where u i,z is the control input in z direction, subscript i is the agent number, subscript z represents the z-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the z-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the z direction, and the sum represents the estimation of the unknown parameters.

2. The method of claim 1, characterized in that, ； ; ; wherein N is the number of agents in the formation of agents, = 1 means that the agent numbered j is a neighbor of the agent numbered i, = 0 means that the agent numbered j is not a neighbor of the agent numbered i, represents that the agent i can directly obtain the real-time position information of the virtual leader without other agents, represents that the agent i cannot obtain the real-time position information of the virtual leader, , , are, in sequence, the coordinates along the x, y, z directions indicated by the expected trajectory of the agent numbered i, , , are, in sequence, the coordinates along the x, y, z directions indicated by the expected trajectory of the agent numbered j, , , are, in sequence, the formation parameters of the agent numbered i in the x, y, z directions , , are, in sequence, the formation parameters of the agent numbered j in the x, y, z directions, , , are, in sequence, the trajectories of the virtual leader in the x, y, z directions.

3. The method of claim 1, wherein, in the first stage, ； ； ； wherein, , , are, in sequence, the filtered sample sequence of the derivative of the trajectory of the agent i in the x, y, z directions with respect to time, , , are, in sequence, the filtered sample sequence of the control input of the agent i in the x, y, z directions, , , are, in sequence, the filtered sample sequence of the known function representing the unknown disturbance of the agent i in the x, y, z directions; wherein l is the number of each data point in the filtered sampling sequence of the motion trajectory, the filtered sampling sequence of the control input, and the filtered sampling sequence of the known function representing unknown disturbance, and m is the number of each data point in the filtered sampling sequence of the motion trajectory, the filtered sampling sequence of the control input, and the filtered sampling sequence of the known function representing unknown disturbance currently known.

4. The method of claim 1, wherein, in the second stage, ； ； ； wherein, , , are, in sequence, the filtered sample sequence of the derivative of the trajectory of the agent i in the x, y, z directions with respect to time, , , are, in sequence, the filtered sample sequence of the control input of the agent i in the x, y, z directions, , , are, in sequence, the filtered sample sequence of the known function representing the unknown disturbance of the agent i in the x, y, z directions; wherein l is the number of the m data points of the filtered sampling sequence of the motion trajectory collected in the first stage, and m is a fixed value.

5. The method of claim 1, wherein, With the rectangular coordinate system of the earth Oxyz as a reference, in the first stage, when the absolute value of the difference between the position of the agent and the position defined by the expected trajectory of the agent is greater than a first preset value in any coordinate axis direction, the position information of the agent in the coordinate axis direction is updated to the agent capable of receiving the communication information of the agent; With the rectangular coordinate system of the earth as a reference, in the second stage, when the absolute value of the difference between the position of the agent and the position defined by the expected trajectory of the agent is greater than an exponentially decaying value of a second preset value over time, the position information of the agent in the coordinate axis direction is updated to the agent capable of receiving the communication information of the agent.

6. A control device of an agent, characterized by, The agent is an agent in an agent formation, and a communication topology of the agent formation can be represented as a directed graph containing at least one directed spanning tree, a root node of the at least one directed spanning tree corresponds to an agent whose preset trajectory of a virtual leader is known, and the remaining agents at least know part of parameters of the preset trajectory of the virtual leader, and the control device comprises: a first control module, configured to perform the following operations in the first stage: predict the expected trajectory of the ego and its neighbor agents from the current position information update time to the next position update time of the ego and its neighbor agents according to the updated position information of the ego and its neighbor agents, part of the parameters of the preset trajectory of the virtual leader, and the preset formation parameters; update the position information of the ego to the agents that can receive the communication information of the ego according to the comparison result of the real-time position information of the ego and the expected trajectory of the ego; perform first-order low-pass filtering on the motion trajectory of the ego, the control input of the ego, and the known function representing unknown disturbance, and then sample at fixed time intervals to obtain the filtered sampling sequence of the motion trajectory of the ego, the filtered sampling sequence of the control input of the ego, and the filtered sampling sequence of the known function representing unknown disturbance of the ego, wherein the unknown disturbance is described by a set of known functions and a set of unknown parameters; evaluate the formation tracking error of the ego and its neighbor agents as a whole according to the real-time position information of the ego and its neighbor agents, the expected trajectory of each, the preset formation parameters, and the position information of the virtual leader, wherein if the position information of the virtual leader is unknown, the parameter is removed, estimate the unknown parameters representing unknown disturbance according to the control input of the ego, the filtered sampling sequence of the motion trajectory of the ego, the filtered sampling sequence of the control input of the ego, and the filtered sampling sequence of the known function representing unknown disturbance of the ego, and the motion trajectory of the ego, obtain the estimated value of the unknown parameters representing unknown disturbance, and determine the control input of the ego according to the obtained formation tracking error, the estimated value of the unknown parameters representing unknown disturbance, the position information of the ego, the formation parameters of the ego, and part of the known preset trajectory of the virtual leader; the first stage ends when the preset condition is met, and the preset condition is used to evaluate the accuracy of the evaluation of the unknown parameters representing unknown disturbance; the second control module is configured to perform the following operations in the second stage after the first stage: predict the expected trajectory of the ego and its neighbor agents from the current position information update time to the next position update time of the ego and its neighbor agents according to the updated position information of the ego and its neighbor agents, part of the parameters of the preset trajectory of the virtual leader, and the preset formation parameters; update the position information of the ego to the agents that can receive the communication information of the ego according to the comparison result of the real-time position information of the ego and the expected trajectory of the ego, wherein the judgment standard for whether the second stage updates the position information of the ego to the agents that can receive the communication information of the ego is more stringent than that of the first stage. According to the real-time position information of the agent and its neighbor agents, and the respective expected trajectories, the preset formation parameters, and the position information of the virtual leader, the overall formation tracking error of the agent and its neighbor agents is evaluated, wherein if the position information of the virtual leader is unknown, the parameter is removed, according to the control input of the agent, the filtered sample sequence of the motion trajectory of the agent obtained at the end of the first stage, the filtered sample sequence of the control input of the agent obtained at the end of the first stage, and the filtered sample sequence of the known function representing the unknown disturbance of the agent obtained at the end of the first stage, the unknown parameter representing the unknown disturbance of the motion trajectory of the agent is estimated, and the estimated value of the unknown parameter representing the unknown disturbance is obtained, and according to the obtained formation tracking error and the estimated value of the unknown parameter representing the unknown disturbance, the position information of the agent, the formation parameters of the agent, and part of the known preset trajectory of the virtual leader, the control input of the agent is determined. Wherein, taking the rectangular coordinate system Oxyz of the earth as the reference, the preset conditions for ending the first stage include: In the x-axis direction: ， wherein subscript i is the number of the agent, subscript x represents the x-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to To describe the data points of the filtered sampling sequence of the known function of the unknown disturbance in the x-axis direction, the position disturbance d i,x in the x direction has the expression ; In the y-axis direction: ， wherein subscript i is the number of the agent, subscript y represents the y-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to For the description of the data points of the filtered sampling sequence of the known function of the unknown disturbance in the y-axis direction, the position disturbance d i,y in the y direction is expressed as ; In the z-axis direction: ， wherein subscript i is the number of the agent, subscript z represents the z-axis direction, m is the number of data points of the filtered sampling sequence of the motion trajectory of the self, to For the description of the data points of the filtered sampling sequence of the known function of the unknown disturbance in the z-axis direction, the position disturbance d i,z in the z direction is expressed as ; wherein, are respectively a known smooth nonlinear function of v, is an unknown constant vector, p i,x is the x-axis direction coordinate of itself, p i,y is the y-axis direction coordinate of itself, p i,z is the z-axis direction coordinate of itself; In the first stage and the second stage, the control input of the agent in the x-direction is represented as: ； where u i,x is the control input in x direction, subscript i is the agent number, subscript x represents the x-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the x-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the x direction, and the sum represents the estimation of the unknown parameters; In the first stage and the second stage, the control input of the agent in the y-direction is represented as: ； wherein u i,y is the control input in y direction, subscript i is the number of agent, subscript y represents the y-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the y-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the y direction, and the sum represents the estimation of the unknown parameters; In the first stage and the second stage, the control input of the agent in the z-direction is represented as: ； where u i,z is the control input in z direction, subscript i is the agent number, subscript z represents the z-axis direction, c is a preset constant, is the evaluation of the overall formation tracking error of the agent and its neighbor agents, and are parameters describing the virtual leader trajectory, is the z-direction formation parameter, and are variables for estimating unknown parameters of unknown disturbances in the z direction, and the sum represents the estimation of the unknown parameters.

7. A control device of an agent, characterized by, The agent is an agent in an agent formation, and the communication topology of the agent formation can be represented as a directed graph containing at least one directed spanning tree, the root node of the at least one directed spanning tree corresponds to an agent whose preset trajectory of the virtual leader is known, and the remaining agents at least know part of the preset trajectory of the virtual leader. The control device comprises a memory and a processor, the memory stores a program, and the processor runs the program to execute the method according to any one of claims 1 to 5.

8. A computer program product, characterised in that, When running on the processor, it executes the method according to any one of claims 1 to 5.

Images