Open multi-agent spatiotemporal cooperative capture method, system, device and medium for maneuvering targets

By constructing an online error upper bound and robust gain scheduling for the filtered target acceleration estimate, and decoupling the acceleration command into axial and normal components, the problems of measurement noise and target maneuver uncertainty in multi-agent cooperative acquisition are solved, and stable cooperative acquisition and terminal line-of-sight tracking are achieved.

CN123114016APending Publication Date: 2026-11-17BEIHANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611368659.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing cooperative acquisition methods struggle to guarantee the cooperative accuracy of multi-agent acquisition time and the performance of terminal line-of-sight tracking under conditions of measurement noise and target maneuver uncertainty, and guidance gain scheduling is not robust enough.

Method used

By constructing an online error upper bound for the filtered target acceleration estimate and using this error upper bound to synchronously schedule the robust gains of the line-of-sight axis and line-of-sight normal, the acceleration command is decoupled into axial and normal components, which control the closing velocity and line-of-sight direction respectively, thus generating a three-dimensional acceleration command.

Benefits of technology

It maintains stable cooperative acquisition performance under conditions of measurement noise and target maneuvering, enables multiple agents to arrive at near-simultaneous speeds and meet terminal space configuration requirements, and takes into account dynamic response speed and anti-interference capability, providing a unified and quantifiable guidance framework.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This application discloses an open multi-agent spatiotemporal cooperative capture method, system, device, and medium for maneuvering targets, relating to the field of multi-agent cooperative control technology. The method includes performing the following steps for any agent: generating a filtered target acceleration estimate and its online computable upper error bound; determining the cooperative information between the agent and its active neighbors; decoupling the agent's desired acceleration command into a line-of-sight axial component and a line-of-sight normal component; constructing a feedforward compensation term; generating an axial acceleration scalar based on the feedforward compensation term, axial robust gain, and cooperative information; generating a normal acceleration vector based on the feedforward compensation term, normal robust gain, and the agent's desired line-of-sight direction; synthesizing the axial acceleration scalar and normal acceleration vector to generate and output the agent's three-dimensional acceleration command. This application's method provides a unified, quantifiable, and online-adjustable guidance framework for multi-agent cooperative capture.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of multi-agent cooperative control technology, and in particular to an open multi-agent spatiotemporal cooperative capture method, system, device and medium for maneuvering targets. Background Technology

[0002] Multi-agent cooperative capture refers to the technology of using multiple agents (such as drones, unmanned surface vessels, and mobile robots) to approach and capture targets from different spatial directions. It has significant application needs in areas such as airspace security, unmanned system countermeasures, and the capture of moving targets. Compared to single-agent capture, multi-agent cooperative capture can improve mission reliability and achieve predetermined terminal spatial configurations.

[0003] Existing cooperative acquisition methods typically coordinate remaining flight time, estimated absolute acquisition time, or line-of-sight geometry to enable multiple agents to reach the target simultaneously. However, in practical engineering, the acceleration of a maneuvering target is usually not directly and accurately obtained, but can only be estimated from noisy position and velocity measurements. If the noisy estimate is directly used for guidance compensation, it can easily cause fluctuations in acceleration commands; if only the estimate is filtered without providing a calculable upper bound for the filtered error, it is difficult to select robust gain online based on the current uncertainty.

[0004] Therefore, how to construct an online-calculated filter observation error bound under the condition of coexistence of measurement noise and target maneuver uncertainty, and realize adaptive scheduling of guidance gain based on the error bound, so as to simultaneously ensure the accuracy of acquisition time coordination and terminal line-of-sight tracking performance, remains an unsolved technical problem. Summary of the Invention

[0005] The purpose of this application is to provide an open multi-agent spatiotemporal cooperative acquisition method, system, device and medium for maneuvering targets, to solve the technical problem that existing cooperative acquisition methods are unable to simultaneously guarantee the performance of multi-agent acquisition time coordination and terminal line-of-sight tracking under measurement noise and target maneuver uncertainty.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides an open multi-agent spatiotemporal cooperative capture method for maneuvering targets, wherein the following steps are performed for any agent: Based on the measurement information of the target by the intelligent agent, a filtered target acceleration estimate and its online calculable upper bound for error are generated; Based on the estimated absolute capture time of the agent's active neighbors, the estimated absolute capture time difference between the agent and its active neighbors is determined as cooperative information; The desired acceleration command of the intelligent agent is decoupled into the line-of-sight axis component and the line-of-sight normal component; A feedforward compensation term is constructed based on the estimated acceleration of the filtered target; and the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component are synchronously scheduled according to the upper bound of the error. An axial acceleration scalar is generated based on the feedforward compensation term, the axial robust gain, and the cooperative information; a normal acceleration vector is generated based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction. The axial acceleration scalar and normal acceleration vector are synthesized to generate and output the three-dimensional acceleration command of the intelligent agent.

[0007] Secondly, this application provides an open multi-agent spatiotemporal cooperative capture system for maneuvering targets, comprising: The perception and estimation module is used to generate a filtered target acceleration estimate and its online calculable upper bound based on the measurement information of the target by the intelligent agent; The communication information management module is used to determine the estimated absolute capture time difference between the agent and its active neighbors as cooperative information based on the estimated absolute capture time of the agent's active neighbors; The control decoupling and scheduling module is used to decouple the agent's desired acceleration command into a line-of-sight axial component and a line-of-sight normal component; construct a feedforward compensation term based on the filtered target acceleration estimate; and synchronously schedule the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component according to the error upper bound; generate an axial acceleration scalar based on the feedforward compensation term, the axial robust gain, and the cooperative information; and generate a normal acceleration vector based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction. The instruction synthesis module is used to synthesize the axial acceleration scalar and the normal acceleration vector, generate the three-dimensional acceleration command of the intelligent agent, and output it.

[0008] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the open multi-agent spatiotemporal cooperative capture method for maneuvering targets as described above.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the open multi-agent spatiotemporal cooperative capture method for maneuvering targets described in any one of the above-mentioned methods.

[0010] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides an open multi-agent spatiotemporal cooperative acquisition method, system, device, and medium for maneuvering targets. By constructing an online error upper bound for the filtered target acceleration estimate and using this error upper bound to synchronously schedule robust gains for the line-of-sight axis and line-of-sight normal, the guidance law can perceive the magnitude of current uncertainty in real time and adaptively adjust the strength of robust action, maintaining stable cooperative acquisition performance under measurement noise and target maneuvering conditions. Simultaneously, by decoupling the acceleration command into axial and normal control, the line-of-sight axial component is responsible for adjusting the closing velocity and the expected absolute acquisition time to achieve near-simultaneous arrival of multiple agents, while the line-of-sight normal component is responsible for tracking the desired line-of-sight direction to meet the terminal spatial configuration requirements. The two components share the same filtered target acceleration estimate and the same error upper bound, resulting in a simple structure and low computational cost. Furthermore, the use of feedforward projection compensation of the filtered target acceleration estimate and robust feedback gain scheduled by the error upper bound provides dual protection, ensuring rapid response during rapid target maneuvering and robust stability under model mismatch, balancing dynamic response speed and anti-interference capability. This provides a unified, quantifiable, and online adjustable guidance framework for multi-agent cooperative acquisition. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating an open multi-agent spatiotemporal cooperative capture method for maneuvering targets provided in an embodiment of this application; Figure 2 This is a schematic diagram of multi-agent cooperative target capture provided in an embodiment of this application; Figure 3 This is a schematic diagram of a multi-agent cooperative encirclement trajectory provided in an embodiment of this application; Figure 4a Convergence curves for the estimated absolute capture time of the six agents provided in the embodiments of this application; Figure 4b A schematic diagram illustrating the synchronous decay process of remaining flight time provided in an embodiment of this application; Figure 4c A schematic diagram illustrating the convergence process of the cooperative error of the six intelligent agents provided in this embodiment of the application; Figure 5a Convergence curves for the line-of-sight angle errors of the six intelligent agents provided in the embodiments of this application; Figure 5b The response curve of the azimuth angle of the flight path provided in the embodiments of this application; Figure 5c The response curve of the trajectory pitch angle provided in the embodiments of this application; Figure 6a Provided for the embodiments of this application x Acceleration response curve in the axial direction; Figure 6b Provided for the embodiments of this application y Acceleration response curve in the axial direction; Figure 6c Provided for the embodiments of this application z Acceleration response curve in the axial direction; Figure 7 Acceleration estimation norm error curves provided for embodiments of this application; Figure 8a The adaptive gain coefficient variation curve of the time coordination term provided in the embodiments of this application; Figure 8b The adaptive gain coefficient variation curve for the spatial configuration term provided in the embodiments of this application; Figure 9a The three-dimensional acceleration command response curve provided in the embodiments of this application; Figure 9b Norm curves of normal acceleration provided for embodiments of this application; Figure 9c Norm curve of total control acceleration provided for embodiments of this application; Figure 10 A timing diagram of member states provided for embodiments of this application; Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] This application provides an open multi-agent spatiotemporal cooperative capture method for maneuvering targets, which can be applied to multi-agent systems such as UAV swarms, unmanned surface vessel formations, or mobile robot networks.

[0016] like Figure 1As shown, this method performs the following steps S1 to S6 on any agent within the task area during each control cycle: Step S1: Generate a filtered target acceleration estimate and its online calculable upper bound based on the measurement information of the target by the intelligent agent; Specifically, each agent acquires the target's position and velocity measurements using its onboard sensors (such as radar, visual sensors, GPS / inertial navigation systems, etc.). Simultaneously, it receives guidance messages broadcast by other agents within its communication range via onboard communication equipment. These other agents are the active neighbors at the current moment. The guidance messages include, but are not limited to, agent ID, timestamp, relative distance, estimated absolute acquisition time, and local terminal approach requirements.

[0017] The intelligent agent uses the acquired target position and velocity measurements to construct a saturated observer to suppress measurement spikes. It then applies a first-order low-pass filter to the raw acceleration observations output by the observer to reduce noise and calculates the upper bound of the error in the filtered acceleration estimate online. This upper bound reflects in real time the maximum possible deviation between the current filtered target acceleration estimate and the true target acceleration.

[0018] Step S2: Based on the estimated absolute capture time of the agent's active neighbors, determine the estimated absolute capture time difference between the agent and its active neighbors as cooperative information; Specifically, the agent calculates the difference between its own estimated absolute acquisition time and that of each active neighbor based on the received estimated absolute acquisition time of each active neighbor, forming a local cooperative variable. This cooperative information is used for subsequent cooperative control of the acquisition time of the axial channel.

[0019] Step S3: Decouple the desired acceleration command of the agent into the line-of-sight axis component and the line-of-sight normal component.

[0020] Specifically, the agent decomposes the desired acceleration command to be generated into a component along the line of sight (line of sight component) and a component perpendicular to the line of sight (line of sight normal component) based on the current line of sight direction. The line of sight component is used to adjust the closing speed and predict the absolute acquisition time, while the line of sight normal component is used to control the line of sight direction to track the desired line of sight direction.

[0021] Step S4: Construct a feedforward compensation term based on the estimated acceleration of the filtered target; and synchronously schedule the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component according to the upper bound of the error.

[0022] Specifically, the agent projects the filtered target acceleration estimate onto the line-of-sight axis and the line-of-sight normal, respectively, and constructs axial and normal feedforward compensation terms to offset the impact of target maneuvering on guidance accuracy in advance. Simultaneously, the upper bound of the filtered observation error is projected onto the line-of-sight axis and the line-of-sight normal, respectively, to obtain the axial uncertainty bound and the normal uncertainty bound, and axial robust gain and normal robust gain are set to monotonically increase with the uncertainty bounds, respectively.

[0023] Step S5: Generate an axial acceleration scalar based on the feedforward compensation term, the axial robust gain, and the cooperative information; generate a normal acceleration vector based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction.

[0024] Specifically, in the line-of-sight axial channel, the agent calculates the local terminal approximation margin based on its own cooperative information with active neighbors. It then superimposes the axial compensation term, axial robustness gain, and local terminal approximation margin to generate an axial acceleration scalar, which is used to adjust the closing velocity and the estimated absolute acquisition time to achieve near-simultaneous arrival of multiple agents. In the line-of-sight normal channel, it acquires the desired line-of-sight direction, calculates the line-of-sight angle error between the current line-of-sight direction and the desired line-of-sight direction, constructs a normal sliding mode variable, and superimposes the normal compensation term, line-of-sight geometric coupling term, and the rate of change of the desired line-of-sight direction. This is combined with the normal sliding mode variable to generate a normal acceleration vector, used to track the desired line-of-sight direction to meet the terminal spatial configuration requirements.

[0025] Step S6: Synthesize the axial acceleration scalar and the normal acceleration vector to generate and output the three-dimensional acceleration command of the intelligent agent.

[0026] Specifically, the intelligent agent projects the axial acceleration scalar along the line of sight and the normal acceleration vector in the line of sight normal plane. The two are combined to obtain the desired acceleration vector command in three-dimensional space, which is then output to the underlying flight controller or motion control system and converted into thrust, attitude angle, rudder deflection angle or speed control commands.

[0027] This application constructs an online error upper bound for the filtered target acceleration estimate and uses this error upper bound to synchronously schedule robust gains for the line-of-sight axis and line-of-sight normal, enabling the guidance law to perceive the magnitude of current uncertainty in real time and adaptively adjust the strength of robust action, maintaining stable cooperative acquisition performance under measurement noise and target maneuvering conditions. Simultaneously, by decoupling acceleration commands into axial and normal controls, the axial component of the line-of-sight is responsible for adjusting the closing velocity and the expected absolute acquisition time to achieve near-simultaneous arrival of multiple agents, while the normal component of the line-of-sight is responsible for tracking the desired line-of-sight direction to meet the terminal spatial configuration requirements. The two channels share the same filtered target acceleration estimate and the same error upper bound, resulting in a simple structure and low computational cost. Furthermore, the use of feedforward projection compensation of the filtered target acceleration estimate and robust feedback gain scheduled by the error upper bound provides dual protection, ensuring rapid response during rapid target maneuvers and robust stability under model mismatch, balancing dynamic response speed and anti-interference capability. This provides a unified, quantifiable, and online adjustable guidance framework for multi-agent cooperative acquisition.

[0028] Preferably, for any agent, the following sequence is followed in each control cycle: (1) Preset the observer parameters, filter parameters, control parameters, acquisition radius, communication timeout threshold and acceptance threshold according to the target's physical maneuverability and sensor performance; (2) Initialize the target position, velocity and acceleration observations, initialize the active / candidate state, initialize the upper bound of the filtered observation error and the expected absolute acquisition time; (3) Collect or estimate the local status and target measurement information, and update the saturation observer, acceleration low-pass filter and filter observation error boundary generator; (4) Calculate the line-of-sight motion variables, including the line-of-sight vector, relative distance, line-of-sight unit vector, closing velocity, normal plane projection matrix, and tangential relative velocity; (5) If the absolute value of the closing velocity is not greater than the minimum closing velocity threshold, then single agent closing control is executed and the acquisition time coordination channel is not activated temporarily; otherwise, coordinated guidance continues to be executed. (6) If a member switching event such as message timeout, agent task completion or agent join request is detected, then the failure agent removal, candidate agent pre-synchronization and threshold admission are performed, and the communication graph is reconstructed; (7) If the agent is an active agent, the collaborative variables are calculated from the active neighbor information; if the agent is an unaccepted candidate agent and there are active neighbors, the pre-synchronization variables are calculated. (8) Calculate the terminal approximation requirement, exchange terminal approximation requirement values ​​with active neighbors, obtain the local maximum value, and calculate the local terminal approximation margin; (9) Calculate the scalar acceleration along the line of sight and the normal acceleration vector along the line of sight; (10) Synthesize three-dimensional acceleration commands and convert them into thrust, attitude angle, rudder deflection angle or speed control commands according to the specific platform dynamics; (11) Apply platform-permitted amplitude and rate of change limits to axial, normal and total acceleration commands; (12) When the relative distance is less than or equal to the capture radius, the capture is determined to be complete, cooperative guidance is stopped and the terminal mode is switched.

[0029] In one exemplary embodiment of this application, a preferred method for generating the filtered target acceleration estimate and its online error upper bound is provided, specifically including the following steps A1 to A5: Step A1: Acquisition of Measurement Information Let the first The measured values ​​of the target position and velocity obtained by each agent are denoted as follows: and It satisfies: , in, and These represent the target's true position and velocity, respectively. and These are position measurement noise and velocity measurement noise, respectively, which include sensor noise and transient pulses, among other combined disturbances.

[0030] Let the target's acceleration be... and accelerometer satisfy: , in, and This is a conservative upper bound selected based on the target's physical maneuverability.

[0031] Step A2: Construction of the saturation observer Define location residuals Velocity residual ,in and These are the position and velocity observations of the observer, respectively. They are then processed through a threshold of... and vector saturation function The saturation position residual is obtained. saturation velocity residual : , The saturation function is defined as: in, This represents the original measurement error signal to be processed (i.e., the input vector). or ); Indicates saturation threshold or Threshold , It is higher than the normal measurement residual, used to maintain linear correction under normal conditions, and to limit the amount of correction when abnormal peaks occur, so as to avoid drastic changes in the observer state due to measurement outliers.

[0032] Step A3: Construct a saturation observer: in, , , These are the rates of change of the observed values ​​of target position, velocity, and acceleration over time, respectively. These are the original acceleration observations; , , , , For observer gain, , , , , Observer bandwidth parameter and The preferred range for the observer bandwidth is 3 to 10 rad / s, with a typical value of 5 rad / s.

[0033] Step A4: First-order low-pass filtering To avoid directly using noisy raw acceleration observations, a first-order low-pass filter is set: in, These are the raw acceleration observations (including noise). This is the estimated acceleration value of the filtered target (used for guidance compensation). The derivative of the filtered target acceleration estimate. This represents the filter bandwidth (a constant). It should be higher than the guidance loop bandwidth to balance smoothness and response speed, with a preferred range of 5~20 rad / s and a typical value of 10 rad / s.

[0034] Step A5: Calculation of Filtered Observation Error Boundaries Define the observation error matrix And select the observer parameters to make the observer error matrix Let Hurwitz be the matrix; take positive definite matrix. , satisfy .

[0035] Based on the upper bound of the target jerk Measure the upper bound of the disturbance and the upper bound of the initial observation error, and calculate the original acceleration observation error bound online: in, This serves as the upper bound for the online calculation of the original acceleration observation error; The condition coefficients, ; For error decay rate, ; This is the upper bound of the initial integrated observation error; This represents the upper bound of the steady-state acceleration observation error. and Let represent the minimum and maximum eigenvalues ​​of the matrix, respectively. Further construct a filter observation error bound generator consistent with the acceleration filter: in, The upper bound of the filtered acceleration estimation error is given. for The derivative of the initial time. The value satisfies Therefore, the filtering target acceleration estimation error satisfies... .

[0036] In subsequent control law design, this upper bound of error... Robust gain scheduling will be used for both the line-of-sight axis and the line-of-sight normal.

[0037] To ensure that the online calculation of the upper bound of the filtered observation error has clear operability and numerical stability, the positive definite matrix... and Select according to the following steps: Step A51: First determine Solution Pre-select a symmetric positive definite matrix Then solve the Lyapunov equations. The unique symmetric positive definite solution matrix is ​​obtained. Due to the observer error matrix The solution is configured as a Hurwitz matrix, and therefore must exist and be unique. Preferably, the... If we take it as the identity matrix I, the weighting of the errors of each state is consistent, and the calculation is the simplest; or we can take it as a diagonal matrix. ,in , , All are greater than 0, and the convergence rates of position, velocity and acceleration observation errors are assigned respectively, with typical values ​​ranging from 0.1 to 10.

[0038] Step A52: Observer bandwidth parameterization configuration The three poles of the observer are uniformly arranged on the real axis. Among them This represents the observer bandwidth, measured in rad / s. The error matrix is ​​then... The standard friend matrix form is: The characteristic polynomial of this matrix is ​​(s+ω0). 3 That is, all three poles are in At this point, the observer's dynamic response is monotonic and has no overshoot, facilitating engineering debugging. Observer bandwidth. The preferred range is 3 to 10 rad / s, with a typical value of 5 rad / s.

[0039] Step A53: Example of specific numerical construction Taking the observer bandwidth ω0 = 5 rad / s as an example, construct it according to the following steps. and : Will Substituting into the standard friendly matrix form, we get: Pick (i.e., a 3×3 identity matrix): make It is a symmetric positive definite matrix, and its general form is: The above and Substituting into Lyapunov equations Solve (either by manual derivation or by direct calculation using a standard numerical linear algebra library) to obtain a unique positive definite solution: Verification according to the Sylvester criterion Positive definiteness: First-order principal minor 5 > 0; Second-order principal minor: 5 × 1 - (-0.5) 2 =4.75>0; Third-order principal form (determinant) >0; All three principal minors are greater than zero, therefore It is a positive definite matrix.

[0040] Back-generation verification is also satisfied The result has been confirmed to be correct.

[0041] In practice, the already determined and selected As input, standard numerical linear algebra libraries (such as MATLAB's `lyap` function or Python's `scipy.linalg.solve_lyapunov` function) are invoked to solve the problem offline or online. .because The Hurwitz matrix is ​​guaranteed to be used, and the solution process is unconditionally stable. Observer bandwidth. Generally, the bandwidth is set to 3 to 5 times that of the guidance loop to achieve a balance between rapidly tracking target maneuvers and suppressing measurement noise amplification; if the target maneuver is more violent (e.g., peak acceleration exceeds 20 m / s²), the bandwidth is adjusted accordingly. 2 (), can be appropriately increased to 8-10 rad / s, corresponding to The matrix elements increase proportionally.

[0042] Steps A1 to A5 suppress measurement spikes using a saturation observer and provide an online calculable upper bound for error using a filter error bound generator, providing a unified quantitative basis for subsequent online scheduling of axial and normal gains.

[0043] In one exemplary embodiment of this application, a preferred method for dynamically joining and leaving agents in an open multi-agent system is provided, specifically including the following steps B1 to B5: Step B1: Expect absolute capture time interaction Each agent periodically broadcasts a guidance message containing its ID, timestamp, relative distance, estimated absolute acquisition time, and local terminal approach requirements. : in, Send the agent number to the message; For message timestamp; For intelligent agents The relative distance to the target; To estimate the absolute capture time (i.e., the current moment) Plus remaining flight time , ); To approximate the needs of local terminals.

[0044] Step B2: Member Validity Judgment During member switching time For each candidate agent, perform the following validity check: (1) Message Received and the timestamp has not exceeded the communication timeout threshold (The preferred range is 0.5 to 5 seconds, with a typical value of 1 second). (2) The relative distance is greater than the capture radius (Based on the size of the intelligent agent platform and the safety distance setting).

[0045] Only candidate agents that meet all of the above conditions are considered valid agents.

[0046] Step B3: Candidate Agent Pre-synchronization Mechanism For valid candidate agents that have not yet entered the set of active agents Define the set of active neighbors from which information can be directly received. The candidate intelligent agent It only receives communication information from active neighbors in one direction and calculates the maximum difference between its own estimated absolute acquisition time and the estimated absolute acquisition times of all active neighbors: in, Candidate intelligent agents Acquisition time mismatch error; , as candidate intelligent agents A set of active neighbors that can directly receive information. Representing candidate intelligent agents A collection of active neighbors Represents a set of active agents; Candidate intelligent agents The estimated absolute capture time; Candidate intelligent agents The estimated absolute capture time; The maximum capture time mismatch threshold that is allowed for admission; This indicates traversing the set of active neighbors; Representing candidate intelligent agents With active neighbors The expected absolute capture time difference between them; Before the acceptance threshold is met, candidate agents only receive information for pre-synchronization, and their information is not fed back to the consistency control loop of active agents, thereby avoiding unsynchronized candidate agents from interfering with the stable system.

[0047] Step B4: Admission Criteria and Member Update When candidate agents An agent can be accepted as an active agent only if all of the following conditions are met: B41) is a valid intelligent agent (message received and not timed out, relative distance is greater than the capture radius); B42) There exists a non-empty set of active neighbors; B43) Acquisition time mismatch error Less than the preset acceptance threshold (The preferred range is 0.5 to 5 seconds, with a typical value of 2 seconds).

[0048] At each member switch, update the set of active agents as follows: in, For the moment before the switch, For the time after the switch, The set of active agents before the switch; The set of active agents after the switch; A set of active agents that have failed, timed out, completed their tasks, or voluntarily left the team; The set of candidate agents that meet the acceptance criteria.

[0049] Step B5: Communication Graph Reconstruction After a member switching event occurs, the active neighbor set, communication weights, and Laplace matrix of each agent are regenerated. Specifically, let the number of active agents be N, and their communication relationships be represented by an undirected weighted graph. It means that, among them This is the set of vertices (corresponding to each agent). Let be the set of edges. It is an adjacency matrix. Adjacency matrix elements For non-negative communication weights, the first... i The active neighbor set of each agent is If the intelligent agent i and j If there is a communication link >0, otherwise =0.

[0050] Laplace matrix Defined as: For the updated active agent, construct capture-time locally consistent variables. : in, , For the first iThe estimated absolute capture time for each agent; , For the first j The estimated absolute capture time for each agent; For the current moment, and The first i The first intelligent agent and the first j The remaining flight time of each agent. For candidate agents that are not accepted, the same form is used, but only their active neighbor set is used. Pre-synchronize one-way information in the process.

[0051] Steps B1 to B5 implement a one-way pre-synchronization and threshold admission mechanism for candidate agents, enabling seamless switching when agents dynamically join and leave the queue in the open system, thus avoiding impact on the already stable cooperative loop.

[0052] In one exemplary embodiment of this application, a preferred method for decoupling acceleration commands and synchronizing dual-channel gain is provided, specifically including the following steps C1 to C5: Step C1: Symbol Definition Let the first i The position, velocity, and acceleration commands for each agent are denoted as follows: , and The position, velocity, and unknown acceleration of the maneuvering target are denoted as follows: , and Define the following variables (see Table 1 for details): Table 1 defines variables

[0053] Step C2: Decoupling of acceleration commands The first i Acceleration commands for individual agents Decomposed into the axial component and the normal component of the line of sight: in, It is a scalar of line-of-sight axial acceleration, used to adjust the closing motion and capture time; The acceleration vector is the line-of-sight normal vector (located in the line-of-sight normal plane) and satisfies the orthogonal constraint. ; The unit direction is within the plane of the line of sight.

[0054] Step C3: Projection of Filtered Observation Error Boundaries Upper bound of the filtering observation error Projecting these onto the line-of-sight axis and the line-of-sight normal respectively, we obtain the axial uncertainty bound. and normal uncertainty boundary : in, This is the axial control coefficient. ; For axial numerical margin, ; For normal numerical margin, ; and To prevent the denominator from being zero and excessive gain, the preferred range is 0.01 to 0.1, with a typical value of 0.02.

[0055] Step C4: Axial Robust Gain Scheduling Define the uncertainty bound along the axis Monotonically increasing axial robust gain : in, The axial robust scheduling parameter; the axial robust gain , Increases and monotonically increases, with It decreases and monotonically decreases. Because... Includes positive margin The gain is always positive; when When the gain is large, it increases quadratically to enhance robustness.

[0056] Step C5: Normal Robust Gain Scheduling This step involves normal robustness gain. Starting from this point, we further construct the normal convergence gain for the normal control law. The relationship between the two is that of an intermediate variable and a final control parameter.

[0057] To construct the normal control law more precisely, this embodiment uses the normal robust gain... The normal convergence gain is further derived in the middle. .

[0058] Define the uncertainty bound along the normal direction Monotonically increasing normal robust gain: in, The normal error bound is used to schedule the gain (preferred range 0.5 to 2, typical value is 1); The allowable normalized normal error threshold, ; Furthermore, the normal convergence gain is constructed as : in, For the normal basis convergence parameters, (Preferred range 0.5–2, typical value 1). When the normal direction is uncertain... When it increases, Automatically increases, thus making Increase. This gain directly affects the saturation reaching law, thereby effectively improving the reaching law's anti-jamming capability and robustness under strong target maneuvering.

[0059] Steps C1 to C5 implement the same filter observation error boundary for gain scheduling of both axial and normal channels, enabling the control gain to automatically adjust with the uncertainty of the target maneuver, thus achieving "on-demand robust" adaptive control.

[0060] In one exemplary embodiment of this application, a preferred method for constructing a feedforward compensation term is provided, specifically including the following steps D1 to D2: The filtered target acceleration estimate Projecting onto the line-of-sight axis and line-of-sight normal respectively, constructing axial compensation terms and normal compensation terms: Step D1: Axial compensation item That is, the projection of the filtered target acceleration estimate onto the line-of-sight axis, used to compensate for the impact of target maneuvering on the dynamics of the closing velocity.

[0061] Step D2: Normal Compensation Item That is, the filtered target acceleration estimate is projected onto the line-of-sight normal, and is used to compensate for the impact of target maneuvering on the dynamics of the line-of-sight direction.

[0062] The physical function of the feedforward compensation term is to offset the impact of target maneuvering on guidance accuracy in advance, avoiding the lag response and acceleration command fluctuations caused by relying solely on feedback control. Since the feedforward term uses the estimated value after filtering and observer processing, rather than the original noisy measurement value, it can balance compensation accuracy and control smoothness.

[0063] The above axial compensation item The axial control law will be directly applied to step E4; normal compensation term. This forms the basis for the feedforward compensation of the normal control law in step F4, while the dynamics in step F3... Both are acceleration terms, representing different levels of target motion compensation.

[0064] In one exemplary embodiment of this application, a preferred method for generating an axial acceleration scalar is provided, specifically including the following steps E1 to E4: Step E1: Definition of Remaining Flight Time When the speed of closing the gaze When (minimum closing velocity threshold) is defined i Remaining flight time for each agent: in, >0, a sufficiently small positive number, is used to ensure the remaining flight time definition is valid; the specific value is set according to the actual closing speed in the terminal phase of the mission scenario. When At this time, the capture time coordination channel is not activated, and single-agent closed control is executed.

[0065] It should be noted that, It should be set to a positive number, and its value must be greater than the maximum possible closing velocity (a conservative upper bound) during the terminal phase (i.e., when the agent approaches the target to near the capture radius). This is to ensure the remaining flight time... The value is defined and positive throughout the entire cooperative capture process to avoid problems caused by insufficient closing speed at the terminal stage. It increases dramatically or loses its physical meaning.

[0066] Step E2: Axial Dynamics The axially controllable form is obtained from closed-loop velocity dynamics: in, , Here, The term to be compensated is formed by the tangential motion and the target axial acceleration. This refers to the axial control coefficient. Define the item to be compensated Decomposition form: Among them, the first item This is the line-of-sight geometric coupling term (centrifugal effect caused by line-of-sight rotation); Second item The term to be compensated is caused by the axial component of the target acceleration.

[0067] Step E3: Calculation of local terminal approximation margin Define terminal approximation requirements : in, Based on consistency gain, ; This is the axial robust gain.

[0068] The local neighborhood maximum terminal approach requirement is : Local terminal approximation margin : in, To ensure a margin for the remaining flight time to continue to decrease (empirical value of 0.1~0.5, typical value of 0.2).

[0069] Sign logic determination for coordinated acceleration and deceleration: In actual control, the specific acceleration or deceleration behavior is determined by locally consistent variables. The sign of the variable is determined (this variable is substituted into the co-occurrence term in the subsequent axial acceleration command). ): when >0: indicates that the first i The estimated absolute capture time of an agent is later than the neighborhood weighted average (i.e., relatively lagging), so the axial acceleration needs to be increased to catch up with active neighbors; when <0: indicates the first i If an agent's predicted absolute capture time is earlier than the neighborhood weighted average (i.e., relatively ahead), the axial acceleration needs to be reduced (or even negative acceleration applied) to wait for active neighbors; when When =0: indicates that the first i Each agent achieves complete time synchronization with its neighborhood and coordinates the correction of terms. Automatically resets to zero. At this point, the agent does not perform coordinated adjustments for acceleration or deceleration, relying solely on the feedforward compensation term and terminal approximation term in the axial acceleration command. Maintain the current close proximity and continue to move closer to the target.

[0070] Step E4: Axial Control Law The final line-of-sight axial acceleration scalar design is as follows: : Among them, the first item This is a feedforward compensation term (filtered target acceleration axial projection), used to compensate for the effect of target maneuvering on the closing velocity; Second item This is a line-of-sight geometric coupling compensation term, used to compensate for the effect of centrifugal force caused by line-of-sight rotation on the closing velocity; Third item For the terminal approach and capture time coordination term, For axial robustness gain, This is a local terminal approach margin used to ensure that the remaining flight time continues to decrease (terminal approach). Used to achieve coordinated convergence of capture time among multiple agents.

[0071] The gain scheduling characteristic of this control law: as axial uncertainty increases, the axial robust gain... Automatically increase to enhance robust synergy; when uncertainty decreases, Automatically reduce unnecessary control energy consumption to achieve "on-demand robust" adaptive control.

[0072] Implementation steps E1 to E4 are passed To ensure the remaining flight time continues to decrease, by To achieve convergence of capture time among multiple agents, the two agents are decoupled within the same axial control law.

[0073] In one exemplary embodiment of this application, a preferred method for generating a normal acceleration vector is provided, specifically including the following steps F1 to F4: Step F1: Desired line of sight Get the i The desired line of sight of an agent The desired line-of-sight direction can be a pre-defined fixed direction or a time-varying direction. When the direction changes, the external task planning module provides a time function of the desired line-of-sight azimuth and pitch angles. in, and These are the desired azimuth and pitch angles, respectively (given by the external mission planning module or generated according to a preset time function).

[0074] For a fixed direction, , It is a constant, at this time It does not change over time.

[0075] The desired line of sight To ensure that the smooth time function conforms to the maximum maneuverability of the agent platform, its rate of change is limited by the agent's maximum available overload, so as to ensure that the generated acceleration command is within the feasible domain of the physical actuator.

[0076] Step F2: Line of Sight Angle Error Define the current line of sight. and the desired line of sight The line-of-sight error between : make The tangential unit direction from the current line of sight to the desired line of sight: , Current line of sight and the desired line of sight When the directions are opposite ( The denominator is zero. At this point, we can maintain the previous moment's... The perpendicular unit vector is used as Because the system will immediately deviate from this singular configuration, once it deviates, the denominator will automatically return to non-zero, and the formula can be calculated normally.

[0077] In actual flight control software, a very small positive threshold ε (e.g., ε=1e) can be set. -6 ), when | When | < ε, the pre-stored orthogonal vector is directly called, thus avoiding singularities in numerical calculations.

[0078] Define the convergence gain of the line-of-sight angle (Preferred range 0.5~2, typical value 1), construct the desired tangential velocity corresponding to the desired line-of-sight direction. : Further construct the normal sliding mode variable: This sliding mode variable reflects both the line-of-sight angle error and the tangential velocity tracking error. Its convergence means that the current line-of-sight direction is approaching the desired line-of-sight direction.

[0079] This sliding mode variable also reflects the line-of-sight angle error (via...). ) and tangential velocity tracking error (via ).when When the error approaches zero, both errors converge to zero, meaning that the current line of sight approaches the desired line of sight, and the desired tangential velocity is accurately tracked.

[0080] Step F3: Effective Dynamics of the Normal Channel In the normal plane of the line of sight, the normal sliding mode variable The effective kinetics are: Among them, the first item This is the line-of-sight geometric coupling term (the tangential velocity change caused by line-of-sight rotation); Second item For the target maneuver term (the projection of the target acceleration change onto the normal plane); Third item This is the compensation term for the desired rate of change of tangential velocity (the projection of the desired change in line of sight direction onto the normal plane). The normal acceleration vector to be designed (control input); It is the unit vector of the line-of-sight normal acceleration vector.

[0081] Combine the three items other than the control input into a composite quantity: The dynamics can then be simplified as: in, It is a known / calculable composite quantity of the normal channel, which includes three parts: geometric coupling, target maneuver, and desired line-of-sight change.

[0082] It should be noted that, It is the target acceleration The derivative (jerk), the observer output in step A3. It can be obtained through numerical differentiation, or directly using... ; Depend on Further differentiation yields: in (When approaching the target, the closing velocity is positive and the distance decreases).

[0083] Step F4: Normal Control Law The final line-of-sight normal acceleration vector design is as follows : in, The term refers to the saturation reaching law used in this embodiment, which is used to ensure the normal sliding mode variable. Finite-time convergence; The normal convergence gain is given by step C5.

[0084] To prevent chattering that may occur in sliding mode control, the fourth item can be replaced with the following in actual engineering implementation: in, The allowable normalized normal error threshold ( The preferred range is 0.01~0.1, with a typical value of 0.05, to avoid chattering in sliding mode control. Specifically, the saturation function is defined as: When the right side of the equation When it is not zero, take Its vector value; when the right end When it is zero, take As can be seen from the construction method, the normal acceleration vector always lies within the line-of-sight normal plane, satisfying the orthogonal constraint. .

[0085] It should be noted that the singularity handling in step F2 targets special geometric points in the line-of-sight configuration, representing anomaly avoidance at the path planning level; while the saturation function in step F4 addresses the inherent chattering of sliding mode control, representing smoothing at the control level. These two processes complement each other, jointly ensuring the algorithm's numerical robustness under extreme conditions.

[0086] Steps F1 to F4 were implemented to achieve accurate tracking of the time-varying desired line-of-sight direction through sliding mode variable construction and robust approach term design. The numerical robustness of the algorithm under extreme conditions was ensured through saturation functions and singularity handling.

[0087] In one exemplary embodiment of this application, the aforementioned axial acceleration scalar is... and normal acceleration vector Synthesize to obtain the first i Three-dimensional acceleration commands for an intelligent agent : in, For the axial component of the line of sight, For the line-of-sight normal component.

[0088] The three-dimensional acceleration command The output is sent to the underlying flight controller or motion control system and converted into thrust, attitude angle, rudder deflection angle or speed control commands according to the specific platform dynamics.

[0089] Preferably, for the axial acceleration scalar Normal acceleration vector And the final synthesized three-dimensional acceleration command Apply platform-permitted amplitude and rate-of-change limits to ensure control commands remain within the feasible range of the physical actuators. Specifically: , , , , Among them, each upper limit value ( , , () represents the maximum amplitude allowed by the executing agency. , , The maximum rate of change allowed by the actuator is determined by the specific physical capabilities of the platform.

[0090] When relative distance When (capture radius) is reached, the first... i Once the agent completes capture, it ceases cooperative guidance and switches to terminal security, escaping, docking, or mission end mode.

[0091] This application also provides an application scenario that utilizes the aforementioned open multi-agent spatiotemporal cooperative capture method for maneuvering targets, using six potential agents and one maneuvering target as an example for detailed explanation. Specifically, the simulation conditions are set as follows: 1. Target motion parameters The target's initial position and velocity are as follows: The target employs a bounded smooth maneuver, with acceleration varying according to a sinusoidal law: The peak acceleration is approximately 15 m / s². 2 .

[0092] 2. Initial configuration of the agent Initially (t=0s), agents 1, 2, and 3 are active; agents 4, 5, and 6 become available at t=5s, 10s, and 15s respectively, and enter the active agent set only after meeting the admission conditions described in step B4; agent 2 leaves the active agent set at t=12s. (Capture radius) =5m.

[0093] 3. Main control and observation parameters In this embodiment, the values ​​of various parameters involved in each step are shown in Table 2 below: Table 2 Values ​​of various parameters

[0094] 4. Time-varying expected line of sight The desired line-of-sight direction of each agent is taken in a time-varying form. Let the desired line-of-sight azimuth angle be... and pitch angle They are respectively: The reference values ​​are taken in the following order: Phase parameters: Desired line-of-sight unit vector Synthesized from the formula in step F1: With the above parameter configuration, simulation was performed according to the methods described in the above steps, and the results are as follows: (1) Acquisition time coordination At the initial moment (t=0s), the maximum difference in the estimated absolute capture time of active agents 1, 2, and 3 is approximately 10.20s. After the coordinated adjustment of the axial control law described in step E, the estimated absolute capture time of each active agent converges to a small neighborhood at the final moment, and the difference in the final capture time is approximately 0.08s.

[0095] This demonstrates that the method described in this application can still achieve high-precision near-simultaneous capture even under conditions of dynamic addition and deletion of members.

[0096] (2) Terminal eye-tracking The initial line-of-sight angle error was greater than 34°. After adjustment by the normal control law described in step F, the terminal line-of-sight angle error was reduced to within 0.2°. The actual line-of-sight azimuth and elevation angles can effectively track their respective time-varying expected values.

[0097] This demonstrates that the method described in this application can achieve high-precision terminal line-of-sight direction control, meeting the configuration requirements for multiple agents to approach the target from different spatial directions.

[0098] (3) Target acceleration estimation The raw acceleration observations output by the saturation observer in step A contain significant fluctuations, while the estimated values ​​after low-pass filtering are smoother. The true filtering estimation error always lies within the upper bound of the online generated error. Within.

[0099] This indicates that the upper bound of the filtered observation error is... It can serve as an effective quantitative basis for axial and normal gain scheduling in step C.

[0100] (4) Robust gain scheduling Axial robust gain defined in step C4 And the normal error bound scheduling gain defined in step C5 The value is relatively large in the initial transient phase of the observer, and increases with the upper bound of the filtered observation error. It automatically decreases due to decay.

[0101] This demonstrates that the "on-demand robust" gain scheduling mechanism proposed in this application is effective, enhancing robustness when uncertainty is high and reducing control energy consumption when uncertainty decreases.

[0102] (5) Boundedness of acceleration command Under the aforementioned amplitude and rate of change constraints, the axial acceleration scalar Normal acceleration vector And the synthesized three-dimensional acceleration command All remain within the feasible range of physical actuators.

[0103] This indicates that the method described in this application has good engineering feasibility and can be directly deployed on actual drones, unmanned surface vessels, and other platforms.

[0104] (6) Open system compatibility Under the candidate agent pre-synchronization mechanism described in step B3, agents 4, 5, and 6, after becoming available, first undergo unidirectional pre-synchronization to reach the acceptance threshold. After 2 seconds, agents were added to the active agent set sequentially without oscillation. After agent 2 left at 12 seconds, the capture time coordination and final capture of the remaining active agents were not disrupted.

[0105] This indicates that the member management and pre-synchronization mechanism described in step B effectively decouples the impact of topological mutations on the cooperative control loop, enabling plug-and-play functionality of agents in the open system.

[0106] The above preferred embodiments demonstrate that, under complex conditions such as dynamic addition and removal of members, bounded target maneuvers, and coexistence of position / velocity measurement noise, the method of this application can achieve near-simultaneous acquisition of maneuvering targets by multiple agents (terminal acquisition time difference of approximately 0.08 s) and high-precision time-varying terminal line-of-sight tracking (terminal line-of-sight angle error less than 0.2°). Simultaneously, it achieves "on-demand robust" adaptive control through an online error bound-driven gain scheduling mechanism, ensuring that acceleration commands are always bounded and smooth, satisfying engineering execution mechanism constraints. Furthermore, the candidate agent pre-synchronization and threshold admission mechanisms effectively support the non-disruptive dynamic addition and deletion of agents, and topological mutations do not disrupt existing cooperative loops. Based on the above verification results, the method of this application exhibits good robustness, adaptability, and engineering practicality, and is suitable for typical application scenarios such as airspace security, unmanned system countermeasures, and moving target encirclement.

[0107] The open multi-agent spatiotemporal cooperative capture method, system, device, and medium for maneuvering targets provided in this application can be widely applied in military and civilian fields such as UAV swarm cooperative combat, unmanned surface vessel formation encirclement, and mobile robot network search. This method is independent of the target's maneuver type, changes in the communication graph, and the specific form of the desired line-of-sight direction, possessing broad applicability and significant industrial practical value and economic benefits.

[0108] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the protection scope of this application.

[0109] like Figure 2 As shown in the figure, this application provides a geometric schematic diagram of multi-agent cooperative capture of a maneuvering target. The figure illustrates a single agent. i The relative motion relationship between the target and the agent, and the overall physical process of multi-agent cooperative capture. Among these, the agents... i (No. i (Individual intelligent agents, such as drones, unmanned surface vessels, or mobile robots, are represented by blue dots; the blue curve represents the intelligent agent.) i The capture trajectory and target (the tracked moving target) (represented by red dots), the red curve represents the trajectory of the target, and the intelligent agent. i acceleration vector Velocity vector acceleration vector of a maneuvering target Velocity vector . Figure 2 The following relative motion amounts are marked: For intelligent agents i The line-of-sight vector pointing towards the target; For intelligent agents i The relative velocity vector with respect to the target satisfies ; The desired line-of-sight direction is a time-varying unit vector of the desired line-of-sight, given externally, used to guide the agent to approach the target from a specified spatial direction. Furthermore, Figure 2 The inertial coordinate system is marked on the right side. As a global reference coordinate system, all position, velocity, and acceleration vectors are described in this coordinate system. Figure 2The overall physical process of the method in this application is intuitively demonstrated, which is consistent with the logical flow of steps A to G mentioned above: the target continues to move; the agent obtains the target's motion state estimate in real time through a saturated observer and a low-pass filter; the agent calculates the line-of-sight vector and relative velocity; the acceleration command is decomposed into axial and normal components; the axial channel adjusts the approach speed and achieves time coordination, while the normal channel tracks the desired line-of-sight direction; multiple agents share the estimated absolute capture time through the communication information management module, and finally achieve simultaneous arrival at the target position to complete the coordinated capture. Figure 2 The other two dashed circles represent other cooperative intelligent agents, each approaching the target from different spatial directions using the same method, forming a multi-party encirclement.

[0110] Figure 3 This application provides a simulation trajectory diagram of six intelligent agents collaboratively capturing a maneuvering target in three-dimensional space, illustrating the six agents in a three-dimensional inertial coordinate system (3D). x , y , z The coordinated encirclement and capture process of a moving target in units of m, corresponding to Figure 2 Simulation implementation of the schematic diagram. In the diagram, the black asterisk marks the capture point, representing the final convergence point of all agents; the black curve represents the target's trajectory, starting from the initial position marked by the square and continuing to the position marked by the black asterisk; the colored curves (blue, red, yellow, purple, green, and light blue) correspond to the flight trajectories of agents 1 to 6, respectively. Each of the six agents starts from a different initial position, and their flight trajectories are different. Figure 3 The following six key simulation phenomena of the method in three-dimensional space are intuitively demonstrated: (1) The six agents are completely dispersed from their starting points and distributed in different positions in three-dimensional space; (2) After the coordinated adjustment of the axial control law described in step E, the trajectories of all agents eventually converge to the same star position; (3) The target itself continues to move, which verifies the effective tracking of the target's maneuvering state by the saturation observer and filter estimation in step A; (4) The maneuvering degree of each agent is inconsistent, which reflects the ability of the method in this application to automatically allocate maneuvering amount. Figure 3 To provide simulation results, the flight trajectories of all entities' positions over time are directly output. The path lengths of the agents vary significantly, but the cooperative algorithm utilizes the local consistency variables in step E. Approaching demand from end users The speed of each agent is automatically adjusted to ensure that although the flight distances are different, the time to reach the capture point is as consistent as possible, thus achieving simultaneous capture. Figure 3 The effectiveness of the proposed method was verified under conditions of dynamic member changes, bounded target maneuvers, and three-dimensional space.

[0111] Figure 4a The estimated absolute capture time for the six agents provided in the embodiments of this application The convergence curve is the core result graph for verifying the performance of multi-agent "simultaneous capture". The horizontal axis represents the simulation time. t (Unit: seconds), with the vertical axis representing the first digit. i The estimated absolute capture time of each agent (unit: s), where The remaining flight time is defined in step E1. The six colored curves correspond to agents 1 through 6, and the inset in the lower right corner shows a magnified view of the later stage (21.6–22.2 s). Initially, the estimated absolute capture times of each agent differed greatly (Agent 1 approximately 27 s, Agent 2 approximately 17 s, and the rest ranging from 17 to 27 s). During the simulation, the axial control law described in step E came into play. Agents with longer estimated times reduced their approach speeds, adjusting their times downwards, while agents with shorter estimated times accelerated their pursuit, adjusting their times upwards. All curves gradually converged. In the later stages of the simulation, all curves almost overlapped, and the estimated absolute capture times uniformly converged to approximately 22 s, indicating that the six agents achieved time coordination. Figure 4a and Figure 3 The physical relationship of the 3D trajectory graph is as follows: the 3D trajectory graph shows all agents flying towards the same capture point in space. This graph shows the algorithm coordinating the speed of each agent in time to ensure that they arrive at the capture point at the same time.

[0112] Figure 4b and Figure 4c For the same set of simulated twin graphs, each shows the remaining flight time. The synchronous decay process and cooperative error (i.e., the local consistency variable of the capture time). The convergence process. Figure 4b (Above figure) The horizontal axis represents the simulation time. t (Unit: seconds), with the vertical axis representing the remaining flight time. (Unit: seconds). Initially, the remaining flight time of each agent is inconsistent; under the action of the axial control law described in step E, the curves quickly converge and tend to be consistent; as the simulation progresses, the overall synchronous downward decay occurs, until... t When the time approaches 22s, all of them approach 0, indicating that all agents are evolving toward reaching the goal at the same time. Figure 4c (See the image below) The horizontal axis represents the simulation time. t (Unit: seconds), with the vertical axis representing cooperative error. (i.e., the first) i The weighted sum of the estimated absolute capture time differences between each agent and all active neighbors (defined in step B5), the ideal perfectly cooperative state is: =0. The initial time error is large; with the cooperative term in step E4 The effect is that all curves converge rapidly towards the 0 axis; the jumps that occur in the middle correspond to the member switching events of agents 4 to 6 joining the active agent set in sequence (handled by step B4); in the later stages of the simulation, all curves approach 0, proving that time coordination has been achieved. Figure 4b This demonstrates the synchronous decay trend of the absolute remaining flight time. Figure 4c The time difference between agents was measured by the quantitative error dimension, which together verified the effectiveness of the proposed method in time coordination.

[0113] Figure 5a The line-of-sight errors of the six intelligent agents provided in the embodiments of this application The convergence curve. The horizontal axis represents the simulation time. t (Unit: seconds), with the vertical axis representing the line-of-sight angle error. (unit: deg), where The angle between the actual line of sight and the desired line of sight as defined in step F2, ideally is... =0. Initially, the line-of-sight (LOS) error of all agents is relatively large (approaching 45° at its maximum), indicating a significant deviation between the initial line-of-sight direction and the desired encirclement configuration direction. During the simulation, under the action of the normal control law described in step F, all curves rapidly and monotonically decrease: from tens of degrees to a few degrees in the first 10 seconds, and then all approach 0° around 20 seconds. The physical significance of the LOS error converging to zero is that each agent continuously adjusts its flight attitude and trajectory to ensure that the actual line-of-sight direction tracks the desired line-of-sight direction of the coordinated encirclement. The spatial configuration constraints are satisfied. This figure and... Figure 3 The three-dimensional trajectory diagram and Figure 4 (time coordination curve) together verify the two core constraints of the coordinated encirclement: the time constraint is realized by the axial control law described in step E, and the spatial configuration constraint is realized by the normal control law described in step F.

[0114] Figure 5b and Figure 5c The track azimuth angle provided for the embodiments of this application and trajectory pitch angle The response curve is similar to the aforementioned Figure 3 Figure 4 Figure 5a These simulation examples belong to the same group and are used to verify the tracking and control performance of the method of this invention on the trajectory angle (horizontal heading and vertical direction) of the intelligent agent in three-dimensional space. In the figure, the solid colored lines represent the actual trajectory angles of each intelligent agent, and the dashed lines represent the expected trajectory angle reference values. Different colors correspond to intelligent agents 1 to 6, and the delayed appearance of some curves corresponds to dynamic time-sharing network entry.

[0115] track azimuth ( Figure 5b The yaw angle (in degrees) represents the heading angle of the agent's velocity vector in the horizontal plane, corresponding to the desired line-of-sight azimuth angle in step F1. The horizontal plane component reflects the agent's orientation relative to the target within the horizontal plane. Initially, the actual track angles of agents 1, 2, and 3 (blue, red, and yellow) deviate significantly from the desired dashed line. As the simulation progresses, under the normal control law described in step F, the actual track angles rapidly converge towards the dashed line (desired angle), and later slightly change towards the reference value, completing the horizontal plane orientation alignment. After agents 4, 5, and 6 (purple, green, and light blue) dynamically enter the network midway through the simulation, their track angles quickly adjust and directly converge to their respective desired encirclement azimuth angles, and then remain basically constant, maintaining the encirclement configuration. The physical significance of the track azimuth angles converging to the desired value is that each agent's heading is adjusted to the correct position, forming a spatial encirclement formation around the target.

[0116] Track pitch angle ( Figure 5c The angle (in degrees) between the agent's velocity vector and the horizontal plane corresponds to the desired line-of-sight pitch angle in step F1. The vertical component represents upward ascent, and negative values ​​represent downward descent. The actual pitch angle of the initially joining agents (blue, red, yellow) rapidly approaches the desired angle corresponding to the dashed line, tracking the reference command; later, it follows the desired angle with a gradual change, completing attitude adjustment in the altitude direction. Time-sharing agents (purple, green, light blue), after joining the network, also rapidly follow their own desired dashed line with their actual pitch angle, achieving pitch direction tracking control. The slow change in angle later is due to the continuous movement of the target; the three-dimensional encirclement configuration requires continuous fine-tuning of the UAV's altitude, which is normal.

[0117] from Figure 5b and Figure 5c It can be seen that, Control the horizontal heading. By controlling the vertical height direction, the two images together complete the trajectory angle tracking in three-dimensional space, ensuring that each agent forms an encirclement formation around the target in three-dimensional space. The correspondence between the above simulation results and the aforementioned specific implementation method is as follows: the expected trajectory azimuth and pitch angles are given by the external mission planning module in step F1. and Confirmed; the rapid tracking of the actual track angle to the desired value reflects the normal acceleration vector in step F4. Expected line of sight for time-varying conditions The tracking effect was good; the time-division access agents were still able to quickly complete track angle adjustment after joining the network, verifying the effectiveness of the candidate agent pre-synchronization mechanism described in step B3 and the admission conditions described in step B4. Figure 5a (Sight angle error) The relationship between the convergence curves is: Figure 5aVerifying spatial configuration constraints from the perspective of line-of-sight angle error, this set of figures directly demonstrates the tracking effect of the agent's own track angles (azimuth and pitch) on the desired reference command. The two complement each other and jointly verify the effectiveness of the method of the present invention in three-dimensional spatial enclosed configuration control from different perspectives.

[0118] Figure 6a , Figure 6b , Figure 6c The acceleration response curves in the three coordinate axes provided in this application embodiment correspond to respectively x , y , z Acceleration in the axial direction (unit: m / s²) 2 This is used to verify the noise immunity of the saturation observer and low-pass filter described in step A. In the three sub-figures, the solid black line represents the ideal reference acceleration command output by the guidance system (i.e., the three-dimensional acceleration command synthesized in step G). (The theoretical value), the green dashed line represents the original acceleration observations contaminated by noise. (Simulated sensor measurement noise), the red dashed line represents the acceleration estimate after the first-order low-pass filtering described in step A4 (i.e., the filtered target acceleration estimate). The black solid line is smooth and continuous, representing the theoretically expected acceleration command calculated by the upper-level cooperative guidance law; the green dashed line fluctuates violently, and directly using the measured value would cause control oscillations; the red curve, after filtering and estimation, effectively removes high-frequency noise, is smooth, and closely approximates the black ideal reference curve, verifying the effectiveness of the saturated observer and first-order low-pass filter described in step A. The comparison of the above three curves verifies that, in the presence of acceleration measurement noise, the filtered estimation output can closely approximate the true reference acceleration, providing a reliable feedback signal for the guidance and control system, and also indirectly verifies the upper bound of the filtering observation error in step C. The rationality of using it as a basis for gain scheduling.

[0119] Figure 7 The acceleration estimation norm error curve provided in this application embodiment is used to quantitatively evaluate the estimation accuracy of the saturated observer and the first-order low-pass filter described in step A. The horizontal axis represents the simulation time. t (unit: seconds), the vertical axis represents the norm of the acceleration estimation error (unit: m / s²). 2 The solid blue line in the graph represents the true acceleration value. Compared with the filtered target acceleration estimate norm error between The represents the actual estimation error; the orange dashed line represents the theoretical error upper bound calculated online by the filtered observation error bound generator in step A5. The actual estimation error was initially large, then decreased rapidly, and subsequently oscillated slightly around a low value (due to sensor noise), but remained at a very low level overall. The theoretical upper bound of the error gradually decreased over time, then leveled off. A key phenomenon is that the solid blue line remained below the dashed orange line throughout the entire process, meaning the actual estimation error was consistently less than the theoretically derived upper bound, proving that the filtered observation error bound described in step A5 is conservative and effective. This figure and... Figures 6a-6c The connection is as follows: Figure 6 qualitatively shows the degree of fit between the true value, the noisy observation value, and the filtered estimated value on the triaxial acceleration curve. This figure compresses the three-dimensional error into a norm for quantitative evaluation, and at the same time verifies the effectiveness of the theoretical error bound, providing a quantitative basis for axial and normal gain scheduling based on the error bound in step C.

[0120] Figure 8a and Figure 8b The adaptive gain coefficient variation curve provided in the embodiments of this application. Figure 8a Adaptive gain for time-coordinated terms (i.e., the axial robust gain defined in step C4). Figure 8b Adaptive gain for spatial configuration terms (i.e., the normal error bound scheduling gain defined in step C5), with the vertical axis on the order of 10. -2 The horizontal axis represents simulation time. t (Unit: seconds), the six colored curves correspond to agents 1 to 6 respectively. The initial error is relatively large, so the gain is automatically set to a larger value to speed up convergence; as the simulation progresses, the cooperative error gradually decreases, and the gain adaptively decreases. Figure 8a The time coordination gain approaches zero in the later stage, indicating that the control effect of the time coordination term automatically weakens when the expected absolute capture time of each agent is consistent. Figure 8b The spatial configuration gain initially decreases rapidly before converging to a small non-zero steady-state value. The base gain is retained to ensure the steady-state accuracy of the spatial configuration constraints. The gain curve of the time-sharing agent also undergoes adaptive adjustment from the moment of access. Both types of gains are based on the same upper bound of the filtered observation error generated in step A5. The scheduling process embodies the on-demand robust adaptive control concept of step C. When the error is large, it automatically adopts a large gain to accelerate convergence, and when the error is small, it automatically reduces the gain to suppress oscillations, thus balancing response speed and control smoothness.

[0121] Figure 9a , Figure 9b and Figure 9c The three-dimensional acceleration command response curves provided in this application embodiment are used to demonstrate the characteristics of the actual control quantity output by the method of this application changing over time. The three sub-graphs correspond to the normal control input, respectively. (i.e., the line-of-sight normal acceleration vector generated in step F4) (amplitude), normal acceleration norm and the norm of total control acceleration (i.e., the three-dimensional acceleration command synthesized in step G) (The modulus), the unit is m / s 2 Analysis of the curve changes shows that the initial stage has a large amplitude, corresponding to the agents performing turning maneuvers to quickly form an encirclement formation. After entering the steady-state encirclement stage, the normal input fluctuates slightly around zero, indicating that the formation remains stable. The acceleration at the end of the simulation increases again, corresponding to the terminal maneuver to converge towards the target. The time-sharing agents also experience the process of "large maneuver → decay → steady state → terminal increase". The amplitude of the control input was not amplified without limit throughout the process, satisfying the actuator constraints and verifying the effectiveness of the amplitude and rate of change constraints in step G. Figures 9a-9c The physical meaning can be divided into three stages: the initial stage involves rapid establishment of the encirclement geometry through high-acceleration maneuvers; the middle stage, a steady-state stage, requires only a small amount of control input to maintain the formation, resulting in low control costs; and the final stage involves a renewed increase in control input to complete the encirclement. The entire process demonstrates that the control input amplitude is bounded and there is no control divergence, proving that the method in this application not only achieves error convergence but also ensures reasonable control input without overload.

[0122] Figure 10 This is a timing diagram of the member states provided in an embodiment of this application. The vertical axis represents the six agents (numbered 1 to 6), and the horizontal axis represents the simulation time. t (Unit: s) is used to show the dynamic entry and exit timing of each agent during the simulation process. Horizontal lines in the figure represent the agent's position within the set of active agents. Within the system, agents 1, 2, and 3 participate in the capture mission; an upward ladder represents dynamic entry into the net, while a downward line represents exiting the system. t =0 is initially in the set of active agents; agent 4 is in t Access is phased in approximately 3s, 5.5s, and 7s; Agent 5 in... t Approximately 5s, 6.5s, and 8s are added sequentially; Agent 6 is in... t Agents are added gradually at ≈7s, 9s, and 11s; all agents exit uniformly in the later stages of the simulation. This timing simulation depicts a typical scenario of an open multi-agent system, corresponding to step B as follows: agents 4, 5, and 6 are connected in stages to the candidate agent pre-synchronization mechanism described in step B3; each agent is formally accepted as an active agent after meeting all the conditions described in step B4; after each member switch, step B5 reconstructs the communication graph and updates the Laplace matrix to ensure the local consistency variables of the remaining active agents. It can achieve a smooth transition, which is consistent with Figure 4. The curve exhibits only a small jump at the moment of member switching and then quickly recovers and converges. Figure 10Under the complete set of dynamic network access timing conditions shown, the aforementioned Figure 3 As shown in Figure 9, all performance indicators, including time coordination, spatial configuration, adaptive gain, filtered observation, and bounded control input, have been verified to function normally, fully demonstrating the adaptability and robustness of the proposed method for time-sharing access and dynamic exit of intelligent agents.

[0123] The above Figures 2 to 10 Together, they constitute the complete chain of the method in this application, from principle to simulation verification: Figure 2 This is a schematic diagram illustrating the principle, showing the geometric relationship between velocity, acceleration, and line-of-sight vector, as well as the overall physical process of multi-agent collaborative capture; Figure 10 To configure the experiment, provide timing conditions for the dynamic changes of open system members; Figure 3 Figure 4 illustrates the spatial convergence effect by showing the trajectory in three-dimensional space; Figure 4 also shows the temporal coordination performance. convergence, Synchronous decay, (approaching zero); Figure 5a Exhibition space enclosed configuration tracking ( (approaching zero); Figure 5b and Figure 5c Figure 6 shows the kinematic tracking of velocity magnitude and direction; Figure 6 shows the triaxial acceleration response and filtering effect. Figure 7 The acceleration estimation error and its theoretical bound are quantitatively evaluated; Figure 8 shows the evolution law of adaptive gain, explaining the adaptive adjustment mechanism of the control system from a mechanistic perspective; Figure 9 shows the boundedness and rationality of the control input. The above simulation results collectively verify all performance indicators of the proposed method under conditions of dynamic member changes, bounded target maneuvering, and measurement noise, including time coordination, spatial configuration tracking, adaptive gain scheduling, noise-resistant filtering estimation, and bounded control output. This fully demonstrates that the proposed method possesses good robustness, adaptability, and engineering feasibility, and is suitable for typical application scenarios such as airspace security, unmanned system countermeasures, and moving target encirclement.

[0124] Based on the same inventive concept, this application also provides a system for implementing the above-described open multi-agent spatiotemporal cooperative capture method for maneuvering targets. The solution provided by this system is similar to the implementation described in the above method; therefore, specific limitations in one or more system embodiments provided below can be found in the limitations of the method described above, and will not be repeated here.

[0125] In one exemplary embodiment, an open multi-agent spatiotemporal cooperative capture system for maneuvering targets is provided, comprising: The perception and estimation module is used to generate a filtered target acceleration estimate and its online calculable upper bound based on the measurement information of the target by the intelligent agent; The communication information management module is used to determine the estimated absolute capture time difference between the agent and its active neighbors as coordination information based on the estimated absolute capture time of the agent's active neighbors, and to perform dynamic member management of agents in the open system. Specifically, the communication information management module receives guidance messages from each active neighbor and performs member validity judgment (message timeout detection, relative distance determination); performs one-way pre-synchronization for candidate agents that have not entered the active agent set, and calculates the mismatch error between their estimated absolute capture time and that of the active neighbors; performs member joining after the admission threshold is met, and handles member exit events such as timeout, task completion, or voluntary departure; reconstructs the communication graph after a member switch occurs, and calculates the capture time local consistency variable. .

[0126] The control decoupling and scheduling module is used to decouple the agent's desired acceleration command into a line-of-sight axial component and a line-of-sight normal component; construct a feedforward compensation term based on the filtered target acceleration estimate; and synchronously schedule the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component according to the upper bound of the error; generate an axial acceleration scalar based on the feedforward compensation term, the axial robust gain, and the cooperative information; and generate a normal acceleration vector based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction.

[0127] The instruction synthesis module is used to synthesize the axial acceleration scalar and the normal acceleration vector, generate the three-dimensional acceleration command of the intelligent agent, and output it.

[0128] The data flow between modules is as follows: the output of the perception and estimation module (filtered target acceleration estimate) Upper bound of error ) and the output of the communication information management module (local consistency variables) Active Neighbors Collection The inputs are shared by both the axial acceleration and the control decoupling and scheduling module; the output of this module (axial acceleration scalar) is also shared. Normal acceleration vector The input is sent to the instruction synthesis module; after synthesis, the three-dimensional acceleration command is output. Down to the lowest-level executive bodies.

[0129] As an optional implementation, the perception and estimation module includes a target state perception unit (such as radar or a visual sensor) and a cooperative state communication unit (such as an airborne data link transceiver). The target state perception unit is used to: obtain the target's position. and speed The original measurement information (including sensor noise and transient pulses) is obtained and output to the target state observer to provide the original data source for subsequent saturation observation and filtering estimation. The cooperative state communication unit is used to: receive guidance messages broadcast by active neighbors and extract the serial number from the messages. timestamp Relative distance Expected absolute capture time Approaching demand in local terminals It also sends guidance messages to the communication network for member validity assessment, candidate agent pre-synchronization, and local consistency variables. The calculation provides information on active neighbor collaboration.

[0130] As an optional implementation, the control decoupling and scheduling module includes a decoupling unit, a feedforward construction unit, a gain scheduling unit, an axial control unit, and a normal control unit.

[0131] The decoupling unit is used to decompose the desired acceleration command of the agent into two orthogonal components along the line of sight (axial direction) and perpendicular to the line of sight (normal direction), ensuring that the two are mathematically independent, thus providing a basis for subsequent dual-channel independent control; The feedforward construction unit is used to project the filtered target acceleration estimate onto the line-of-sight axis and normal axis respectively, and construct axial and normal feedforward compensation terms to offset the impact of target maneuvering on guidance accuracy in advance, and avoid the lag response caused by pure feedback control. The gain scheduling unit is used to map the upper bound of the filter observation error to the uncertainty bounds of the axial and normal directions, respectively, and calculates the axial robust gain and normal convergence gain online accordingly, so as to achieve "on-demand robustness" to enhance robustness when the uncertainty is large and reduce control energy consumption when the uncertainty is small. The axial control unit is used to integrate feedforward compensation terms, axial robust gain and local consistency variables to generate an axial acceleration scalar, while ensuring that the remaining flight time continues to decrease and that the expected absolute capture time of each agent tends to be consistent. The normal control unit is used to integrate the feedforward compensation term, normal convergence gain and desired line of sight to generate a normal acceleration vector, enabling the agent to accurately track the time-varying desired line of sight. The data flow between the units is as follows: the decoupling unit provides the decomposition framework; the feedforward construction unit provides compensation terms to the axial and normal control units; the gain scheduling unit provides scheduling gain to the axial and normal control units; the axial control unit outputs the axial acceleration scalar, and the normal control unit outputs the normal acceleration vector. The two outputs are sent to the instruction synthesis module.

[0132] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 11 As shown, the computer device includes a processor, memory, input / output interfaces, and a communication interface. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the aforementioned open multi-agent spatiotemporal cooperative capture method for maneuvering targets.

[0133] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0134] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0135] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0136] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0138] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0139] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0140] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An open multi-agent spatiotemporal cooperative capture method for a maneuvering target, characterized in that, For any agent, perform the following steps: Based on the measurement information of the target by the intelligent agent, a filtered target acceleration estimate and its online calculable upper bound for error are generated; Based on the estimated absolute capture time of the agent's active neighbors, the estimated absolute capture time difference between the agent and its active neighbors is determined as cooperative information; The desired acceleration command of the intelligent agent is decoupled into the line-of-sight axis component and the line-of-sight normal component; A feedforward compensation term is constructed based on the estimated acceleration of the filtered target; and the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component are synchronously scheduled according to the upper bound of the error. An axial acceleration scalar is generated based on the feedforward compensation term, the axial robust gain, and the cooperative information; a normal acceleration vector is generated based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction. The axial acceleration scalar and normal acceleration vector are synthesized to generate and output the three-dimensional acceleration command of the intelligent agent.

2. The open multi-agent spatio-temporal cooperative capturing method for maneuvering targets according to claim 1, characterized in that, Generate filtered target acceleration estimates and their online-calculated upper bounds, specifically including: A saturation observer is constructed by using a vector saturation function to perform amplitude limiting correction on the position residual and velocity residual of the target, respectively. The raw acceleration observations output by the saturation observer are subjected to a first-order low-pass filter to obtain the filtered target acceleration estimate. The upper bound of the error is calculated in real time according to the upper bound of the target's jerk, the upper bound of the measurement disturbance, and the initial error of the saturation observer, following a pre-set exponential decay function.

3. The open multi-agent spatio-temporal cooperative capturing method for maneuvering targets according to claim 1, characterized in that, Based on the upper bound of the error, the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component are synchronously scheduled, specifically including: The upper bound of the error is projected onto the line of sight axis and the line of sight normal, respectively, to obtain the axial uncertainty bound and the normal uncertainty bound; The axial robustness gain is defined as monotonically increasing with the axial uncertainty bound, and the normal robustness gain is defined as monotonically increasing with the normal uncertainty bound.

4. The open multi-agent spatiotemporal cooperative capture method for maneuvering targets according to claim 1, characterized in that, Also includes: When an agent switching event is detected, agents that have failed, timed out, completed their tasks, or actively left the team are removed from the active agent set, and candidate agents that meet the acceptance criteria are added to the active agent set. The candidate agents that meet the acceptance criteria are those that, before being accepted, only receive communication information from active neighbors in one direction, and the maximum value of the difference between the candidate agent's estimated absolute acquisition time and the estimated absolute acquisition time of all active neighbors is less than a preset acceptance threshold. After the agent switching event, the active neighbor set, communication weight, and Laplace matrix of each agent are regenerated.

5. The open multi-agent spatiotemporal cooperative capture method for maneuvering targets according to claim 1, characterized in that, Constructing the feedforward compensation term specifically includes: constructing an axial compensation term using the axial projection of the filtered target acceleration estimate, and constructing a normal compensation term using the normal projection of the filtered target acceleration estimate.

6. The open multi-agent spatiotemporal cooperative capture method for maneuvering targets according to claim 5, characterized in that, The generation of the axial acceleration scalar includes: Calculate the local terminal approximation margin based on the collaborative information between the agent and its active neighbors; The axial compensation term, the axial robust gain, and the local terminal approximation margin are superimposed to generate the axial acceleration scalar.

7. The open multi-agent spatiotemporal cooperative capture method for maneuvering targets according to claim 5, characterized in that, Generating the normal acceleration vector specifically includes: The desired gaze direction of the agent is determined, wherein the desired gaze direction is a time-varying desired gaze direction, and the gaze angle error between the current gaze direction and the time-varying desired gaze direction is calculated; Based on the line-of-sight angle error and the preset line-of-sight angle convergence gain, construct the normal sliding mode variable; Determine the line-of-sight geometric coupling term based on the current line-of-sight direction and the time-varying desired line-of-sight direction; The normal compensation term, the line-of-sight geometric coupling term, and the rate of change of the time-varying desired line-of-sight direction are superimposed. The normal convergence gain is determined based on the normal robust gain. Then, the normal convergence gain is combined with the normal sliding mode variable to generate the normal acceleration vector.

8. An open multi-agent spatiotemporal cooperative capture system for maneuvering targets, characterized in that, include: The perception and estimation module is used to generate a filtered target acceleration estimate and its online calculable upper bound based on the measurement information of the target by the intelligent agent; The communication information management module is used to determine the estimated absolute capture time difference between the agent and its active neighbors as cooperative information based on the estimated absolute capture time of the agent's active neighbors; The control decoupling and scheduling module is used to decouple the agent's desired acceleration command into a line-of-sight axial component and a line-of-sight normal component; construct a feedforward compensation term based on the filtered target acceleration estimate; and synchronously schedule the axial robust gain of the line-of-sight axial component and the normal robust gain of the line-of-sight normal component according to the error upper bound; generate an axial acceleration scalar based on the feedforward compensation term, the axial robust gain, and the cooperative information; and generate a normal acceleration vector based on the feedforward compensation term, the normal robust gain, and the agent's desired line-of-sight direction. The instruction synthesis module is used to synthesize the axial acceleration scalar and the normal acceleration vector, generate the three-dimensional acceleration command of the intelligent agent, and output it.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the open multi-agent spatiotemporal cooperative capture method for maneuvering targets according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the open multi-agent spatiotemporal cooperative capture method for maneuvering targets as described in any one of claims 1-7.