Three-dimensional heterogeneous agent collaborative control method for unmanned aerial vehicles, unmanned vehicles and underground soft tube robots

Abstract

The application discloses a three-dimensional heterogeneous intelligent agent cooperative control method for unmanned aerial vehicles, unmanned vehicles and underground hose robots, and belongs to the technical field of heterogeneous multi-intelligent agent cooperative control and robot autonomous control. A cross-layer support-morphology coupling unified protocol containing absolute target anchoring items, graph structure relative geometry items, exclusive coupling items and memory correction items is constructed in a host computer, protocol reference positions are generated for three types of heterogeneous intelligent agents, corresponding tracking controllers are designed for the unmanned aerial vehicles, the unmanned vehicles and the underground hose robots in the lower layer, stable tracking of the protocol reference positions is realized, and stable cooperative control of the three-layer heterogeneous intelligent agents is realized. The method can realize unified organization, stable formation, cross-layer support maintenance and underground flexible morphology maintenance of the three-layer heterogeneous intelligent agents, and is suitable for cooperative operation scenes in complex environments such as facility agriculture, underground inspection and disaster rescue.

Description

Technical Field

[0001] This invention belongs to the field of heterogeneous multi-agent cooperative control and robot autonomous control technology, specifically relating to a three-dimensional heterogeneous intelligent agent cooperative control method for air-ground-underground drones, unmanned vehicles, and underground flexible tube robots. More specifically, this invention relates to a hierarchical control architecture of an upper-layer unified protocol (host computer) + lower-layer heterogeneous actuators. The upper layer is used to uniformly generate reachable references that meet the requirements of absolute anchoring, relative geometry, cross-layer support relationships, and underground flexible tube morphology maintenance. The lower layer configures different controllers according to the dynamic differences of the three types of robots, realizing unified organization, hierarchical convergence, and stable formation of the three-layer heterogeneous system. Background Technology

[0002] With the increasing application of intelligent equipment such as drones, unmanned vehicles, and underground robots in facility agriculture, underground inspection, disaster relief, and three-dimensional collaborative operations, collaborative control of multiple types of robots has become an important issue in the field of multi-agent systems. Existing research mostly focuses on isomorphic formations or only involves two-layer collaboration: air-to-ground or ground-to-underground. It is difficult to simultaneously handle three-layer heterogeneous systems (air, ground, and underground) within a unified framework. Existing methods have the following main shortcomings: First, they emphasize relative geometry preservation and lack absolute target anchoring, easily leading to overall drift; second, they do not fully characterize the interlayer structural relationships between air-supported ground and ground-supported underground systems; third, they often simplify underground flexible robots to rigid points, ignoring the tension, compression, and bending constraints of flexible hose chains; and fourth, they struggle to account for the differences in dynamic characteristics, input continuity, and robust control among drones, unmanned vehicles, and underground flexible nodes. Therefore, there is an urgent need to construct a collaborative control method for three-layer heterogeneous systems (air-to-ground-underground) to achieve a unified expression and stable control of target anchoring, topological geometry preservation, cross-layer support coupling, and underground flexible morphological constraints. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a three-layer heterogeneous intelligent agent collaborative control method for UAVs, unmanned vehicles, and underground flexible robots, encompassing air, ground, and underground environments. This method does not simply piece together three types of controllers. Instead, it first establishes a unified protocol at the upper layer that can simultaneously act on all three types of objects. This protocol is used to uniformly generate protocol reference states that meet the requirements of absolute target anchoring, graph structure relative geometry preservation, cross-layer support relationships, and underground flexible morphology preservation. Then, at the lower layer, heterogeneous execution controllers are designed to address the dynamic differences of each of the three types of objects, thus forming an overall control framework of "unified organization - heterogeneous execution - hierarchical convergence."

[0004] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0005] A cooperative control method for three-dimensional heterogeneous intelligent agents for drones, unmanned vehicles, and underground flexible tube robots:

[0006] In the host computer, a cross-layer support-morphological coupling unified protocol is constructed, which simultaneously includes absolute target anchoring terms, graph structure relative geometry terms, dedicated coupling terms and memory correction terms, to serve as a reference position for generating protocols for three types of heterogeneous intelligent agents.

[0007] The UAV is controlled by an adaptive backstepping sliding mode controller constructed based on its protocol reference position and dynamic model. The unmanned vehicle is controlled by a continuously normalized kinematic controller constructed based on its protocol reference position and dynamic model. The underground flexible tube robot is controlled by an expanded state observer-assisted optimization iterative sliding mode controller constructed based on its protocol reference position, dynamic model, tension error, and bending error. This achieves stable collaborative control of three heterogeneous intelligent agents in the air, ground, and underground layers.

[0008] Furthermore, the cross-layer support-morphological coupling unified protocol, which includes absolute target anchoring terms, graph structure relative geometry terms, dedicated coupling terms, and memory correction terms, is as follows:

[0009]

[0010] in, Indicates the first The protocol reference location of each node. Indicates the first The protocol reference location of each node. Indicates the first The target location of each node Indicates the first The target location of each node For communication adjacency matrix elements, Indicates the anchor weight. Indicates the absolute target anchoring gain. This represents the relative geometric gain of the graph structure. Indicates the weight of the specific coupling term. Indicates the first Each node's dedicated coupling item, To correct the gain for memory, To unify the protocol's memory state;

[0011] The update law for the unified protocol's memory state is:

[0012]

[0013] in, This is the memory decay coefficient of the protocol layer. For protocol layer parameters, It is the hyperbolic tangent function.

[0014] Furthermore, in order to prevent the unified agreement from degenerating into a regular anchoring formation agreement, cross-layer support relationships are constructed between the air layer and the ground layer, as well as between the ground layer and the underground layer.

[0015] The cross-layer support relationship between the aerial layer and the ground layer is as follows:

[0016]

[0017] in, This represents the support error of the air layer relative to the ground layer. For the first The horizontal component of the drone protocol reference position For the first The horizontal component of the drone target position, the first The current center of gravity of the drone , No. The center of mass of the drone target , For the first The horizontal component of the autonomous vehicle protocol reference position For the first The horizontal component of the target position of the autonomous vehicle For the number of ground support nodes, For ground support assembly;

[0018] The cross-layer support relationship between the ground layer and the underground layer is as follows:

[0019]

[0020] in, This indicates the support error of the ground layer relative to the underground layer, the first... The current underground support center of gravity of the unmanned vehicle , No. The underground support center of mass for the unmanned vehicle target , For the first The horizontal component of the protocol reference position of each discrete node For the first The horizontal component of the target position of each discrete node The number of underground support nodes. It is an underground support assembly.

[0021] Furthermore, the adaptive backstepping sliding mode controller serves as the first... Control input of a drone ,in For the first The control input for a drone The output of the upper-layer unified protocol The second derivative of the reference position of the drone protocol, For position error feedback gain, For speed error feedback gain, Represents element-wise product. For the gain of the sliding mode robustness term, For boundary layer thickness, The position error of the UAV is a saturation function. Indicates the first The deviation of the current position of each drone from the protocol reference position. For the speed error of the drone, Indicates the first The velocity vector of the drone For sliding mode variables, For the first The equivalent damping estimate of a drone, and Satisfying the adaptive damping estimation law , For adaptive gain, Leakage coefficient, This indicates taking the absolute value of the component.

[0022] Furthermore, the continuously normalized kinematic controller is designed to determine the linear velocity of the unmanned vehicle. and angular velocity :

[0023]

[0024]

[0025] in, For longitudinal error feedback gain, For lateral error feedback gain, For heading error feedback gain, It is a saturation function. , For reference velocity modulus, For longitudinal error, For lateral error, For heading error, This is the reference angular velocity.

[0026] Furthermore, the extended state observer assists in optimizing the iterative sliding mode controller to construct the original control input for the underground hose robot. ,in, For the first Reference velocity feedforward for discrete underground nodes For position error feedback gain, For the stretching error feedback gain, For bending error feedback gain, For the gain of the sliding mode robustness term, For boundary layer thickness, For memory state feedback gain, For the extended state observer to estimate disturbances, the location error of the underground discrete nodes. Indicates the first The deviation of the actual position of each underground discrete node from the protocol reference position. For the first Stretching error at discrete nodes For the first Bending error of discrete nodes, underground sliding mode combination variables underground discrete node location error Indicates the first The deviation of the actual position of each underground discrete node from the protocol reference position. For the weight of the stretching error term, For the weight of the bending error term, For the weights of the memory states, The state of the underground discrete nodes is iteratively memorized.

[0027] Furthermore, the original control input Employ optimized iterative smooth updates:

[0028]

[0029] in, The weighting coefficients for the current control input. The weighting coefficients for controlling the input in the previous time step. The weighting coefficients for controlling the input in the first two time steps. To control the update cycle, For the first The control input for each discrete node.

[0030] Furthermore, the stretching error of the discrete nodes of the underground flexible hose robot is:

[0031]

[0032] in, For the first The position vectors of discrete nodes. For the first The position vectors of discrete nodes. Indicates the first The set of neighboring nodes of a discrete node in a hose robot It is a constant. For discrete nodes and The target distance, and , For the first The target location of each discrete node. For the first The target location of each discrete node.

[0033] Furthermore, the bending error of the discrete nodes of the underground flexible hose robot is:

[0034]

[0035] in, For the first The position vectors of discrete nodes. For the first The position vectors of discrete nodes. For the first The target location of each discrete node. For the first The target location of each discrete node.

[0036] Furthermore, the dynamic model of the UAV adopts a second-order dynamic model with velocity damping and external disturbance terms, the dynamic model of the unmanned vehicle adopts a nonholonomic kinematic model, and the dynamic model of the underground flexible tube robot adopts a discrete node dynamic model. The discrete nodes are embedded in a three-dimensional common coordinate system to realize a unified spatial expression of the three-layer heterogeneous intelligent agents.

[0037] Compared with the prior art, the present invention has the following beneficial effects, and these are also the innovative points that need to be clearly emphasized in the present invention.

[0038] First, this invention proposes a three-layer heterogeneous cooperative control architecture consisting of a unified upper-layer protocol and heterogeneous lower-layer actuators. The innovation lies not in designing three separate controllers for the three types of robots, but in first generating a unified protocol reference state that all three types of objects follow at the upper layer, and then implementing dynamic tracking separately at the lower layer. This structure enables the three-layer heterogeneous system to converge collaboratively within a unified organizational framework for the first time, rather than simply connecting multiple controllers in parallel.

[0039] Second, this invention explicitly introduces cross-layer support relationships between the air layer and the ground layer, and between the ground layer and the underground layer, into the unified protocol. Many existing heterogeneous formation methods only describe the geometric relationships between nodes on the same layer, while this invention incorporates the inter-layer centroid support relationship as a dedicated coupling term into the unified protocol. This makes the collaboration of the three-layer system not just an extension of ordinary formation control, but a truly three-layer collaborative control method with upper and lower support structures.

[0040] Third, this invention does not treat the underground flexible tube robot as an ordinary rigid point, but incorporates the tensile error and discrete bending error of the underground layer into a unified protocol and lower layer controller to achieve synchronous control of position convergence and flexible shape maintenance.

[0041] Fourth, this invention introduces protocol memory state in the upper layer protocol and iterative memory state and optimized smooth update mechanism in the lower layer controller, so that the system can not only react to instantaneous errors, but also use historical error information to improve the convergence quality in the middle and later stages, significantly reduce input oscillations, and improve the practical feasibility of engineering.

[0042] Fifth, this invention matches the second-order dynamics of UAVs, the nonholonomic constraints of unmanned vehicles, and the strongly disturbed environment of underground flexible nodes under a unified reference framework, taking into account the requirements of robustness, input continuity, and shape preservation. Therefore, it is more suitable for application in scenarios such as facility agriculture, underground inspection, emergency rescue, and three-dimensional collaborative operations in complex environments. Attached Figure Description

[0043] Figure 1 This is a flowchart of the three-layer heterogeneous intelligent agent collaborative control method of the present invention (air-ground-underground).

[0044] Figure 2 This is a state diagram of the formation construction process at t=0s in the simulation of heterogeneous intelligent agent formation cooperative control of this invention;

[0045] Figure 3 This is a state diagram of the formation construction process at t=5s in the simulation of heterogeneous intelligent agent formation cooperative control of this invention;

[0046] Figure 4 This is a state diagram of the formation construction process at t=10s in the simulation of heterogeneous intelligent agent formation cooperative control of the present invention;

[0047] Figure 5 This is a state diagram of the formation construction process at t=20s in the simulation of heterogeneous intelligent agent formation cooperative control of the present invention;

[0048] Figure 6 This is a state diagram of the formation construction process at t=30s in the simulation of heterogeneous intelligent agent formation cooperative control of this invention;

[0049] Figure 7 This is a three-dimensional spatial diagram of the formation in the simulation of heterogeneous intelligent agent formation cooperative control in this invention;

[0050] Figure 8 This is a graph showing the maximum morphological error in an embodiment of the present invention;

[0051] Figure 9 This is a graph showing the root mean square absolute error of the absolute position in an embodiment of the present invention.

[0052] Figure 10 This is a root mean square curve of the edge error in an embodiment of the present invention;

[0053] Figure 11 This is a root mean square curve of the tensile error of the underground flexible hose in an embodiment of the present invention;

[0054] Figure 12 This is a root mean square curve of the bending error of the underground flexible hose in an embodiment of the present invention;

[0055] Figure 13 This is a graph showing the maximum control input norm in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention; the scope of protection of this invention is not limited thereto.

[0057] like Figure 1 As shown, a collaborative control method for air-ground-underground three-dimensional heterogeneous intelligent agents for UAVs, unmanned vehicles, and underground flexible tube robots includes:

[0058] In this embodiment, the aerial drone, the ground unmanned vehicle, and the underground flexible tube robot are denoted as A, G, and S, respectively; let the drone index be... The index for autonomous vehicles is The discrete node index of the underground flexible tube robot is Let the number of aerial drone nodes be... The number of ground-based unmanned vehicle nodes is The number of discrete nodes of the underground flexible tube robot is The total number of nodes is:

[0059]

[0060] In this embodiment, the superscript " "Indicates the target or expected quantity, indicated by the superscript " , "Indicates the horizontal plane component, This represents the Euclidean norm.

[0061] Step 1: Establish dynamic models for drones, unmanned vehicles, and underground flexible tube robots;

[0062] Before establishing dynamic models for the three heterogeneous intelligent agents, this embodiment first collects state information and relevant model parameters of UAVs, unmanned vehicles, and underground flexible tube robots, and then uniformly transforms the collected results into a three-dimensional common coordinate system (O-XYZ). For UAVs, their position and velocity are obtained through an onboard IMU (Inertial Measurement Unit); for unmanned vehicles, their planar position, heading angle, linear velocity, and angular velocity are obtained through a wheel speed encoder, IMU, and LiDAR SLAM (Simultaneous Localization and Mapping) module; for underground flexible tube robots, the position changes of each discrete node are obtained through miniature IMUs deployed at discrete nodes of the flexible tube and wire displacement sensors. The collected data undergoes time synchronization, coordinate transformation, and filtering processing.

[0063] For the A drone, whose position vector and velocity vector are defined as follows:

[0064] ,

[0065] At the formation level, UAVs are abstracted into a second-order dynamic model with velocity damping and external disturbance terms:

[0066]

[0067]

[0068] in, Indicates the first The position vector of each drone. Indicates the first The velocity vector of the drone, ( , , ) indicates the first The location coordinates of the drone, ( , , ) indicates the first A drone in a common coordinate system , , Velocity components in the axial direction; For the first The rate of change of position of each drone For the first The rate of change of speed of the drone For the control input of the drone; The initial value of the damping matrix of the UAV can be obtained from the nominal parameters of the UAV or pre-flight test. In the actual control process, since the damping matrix is ​​difficult to obtain accurately, this invention treats it as an unknown bounded parameter and updates the equivalent damping estimate online through the subsequent adaptive damping estimation law to compensate for uncertainties such as air resistance and actuator damping. The sum of external disturbances and unmodeled dynamics can be considered as an unknown bounded disturbance, the effects of which are suppressed and compensated by the robust term in the subsequent adaptive backstepping sliding mode controller; in this embodiment It represents the set of real numbers.

[0069] For the An autonomous vehicle, whose planar position is defined as... The heading angle is Its nonholonomic kinematic model is as follows:

[0070]

[0071]

[0072]

[0073] in, For the first The linear velocity input of an autonomous vehicle, For the first The angular velocity input of the autonomous vehicle For the first An autonomous vehicle in a public coordinate system Rate of change of position in the axial direction For the first An autonomous vehicle in a public coordinate system Rate of change of position in the axial direction For the first The angular velocity of an autonomous vehicle.

[0074] For underground flexible hose robots, let them be... Composed of discrete nodes; the first The position vector of a discrete node is defined as:

[0075]

[0076] The discrete node dynamics model is as follows:

[0077]

[0078] in, For the first The position vectors of discrete nodes. For the first Discrete nodes in a common coordinate system Position in the axial direction For the first Discrete nodes in a common coordinate system Position in the axial direction For the first Rate of change of position of discrete nodes For the first Control input for a discrete node; The disturbance is a combination of underground medium contact, soil damping, local friction, etc., and can be regarded as an unknown bounded disturbance; its impact is estimated online by the subsequent ESO (Extended State Observer) and compensated in the control law.

[0079] To facilitate unification with the aerial drone layer and the ground unmanned vehicle layer into the same three-dimensional common coordinate system In this embodiment, the first Two-dimensional location of discrete nodes of underground hose Embedded in three-dimensional space, represented as:

[0080] ,

[0081] in, underground layer relative to the ground plane The fixed depth, with the negative sign indicating that the underground hose robot is located below ground level.

[0082] This 3D embedding is mainly used to unify the spatial representation and 3D formation display of three-layer heterogeneous intelligent agents; the planar motion control of the underground hose robot is still based on its 2D position. conduct.

[0083] Step 2: Define the tensile error and discrete bending error of the underground flexible hose robot;

[0084] To ensure that the underground flexible hose robot can maintain both positional accuracy and flexible structural form during the convergence process, this embodiment introduces tensile error and discrete bending error into the control.

[0085] Consider adjacent discrete nodes in the underground flexible hose robot. and The target distance is:

[0086]

[0087] in, For the first The target location of each discrete node. For the first The target position of each discrete node; the target position is determined by the underground operation task and the preset hose target configuration, and can be obtained by discretizing the target path or the target hose shape curve; it is not a real-time measurement, but a desired position used to calculate the target distance between adjacent nodes, tensile error and discrete bending error.

[0088] Definition of the first The stretching error of each discrete node is:

[0089]

[0090] in, For the first The position vectors of discrete nodes. Indicates the first The set of neighboring nodes of a discrete node in a hose robot; To prevent tiny constants with a denominator of zero; For the first The stretching error vector of a discrete node. The physical meaning of this formula is: the distance between adjacent discrete nodes. Greater than the target distance When, the coefficient in parentheses When the distance between discrete nodes approaches a positive value, the discrete nodes will tend to shrink during the control process; when the distance between adjacent discrete nodes is less than the target distance, the discrete nodes will tend to widen during the control process. Therefore, Used to limit the local elongation or compression of underground hose robots.

[0091] Definition of the first The discrete bending error of each discrete node is:

[0092]

[0093] in, Indicates the first The bending error vector of discrete nodes; the first term The current position of the hose robot is at the [number]th position. Discrete second-order differences at discrete nodes are used to characterize the current local curvature; the second term The target position of the hose robot is at the 1st The discrete second-order difference at each discrete node is used to characterize the local curvature of the target; the difference between the two... This refers to the deviation of the current bend from the target bend. This error term can suppress localized bends or non-smooth bends.

[0094] Step 3: Construct a cross-layer support-morphological coupling unified protocol that simultaneously includes absolute target anchoring terms, graph structure relative geometry terms, dedicated coupling terms, and memory correction terms, to generate protocol reference states for three types of heterogeneous intelligent agents;

[0095] The system defines a uniform protocol reference location for each node (referring to drones, unmanned vehicles, and underground flexible tube robots):

[0096]

[0097] This embodiment adopts the following unified protocol:

[0098]

[0099] in, , indicating the first The protocol reference position of each node (pre-defined); Indicates the first The protocol reference location of each node; , indicating the first The target location of each node (pre-defined); Indicates the first The target location of each node; For elements of the communication adjacency matrix, if node Available nodes The information, ,otherwise ; , representing the anchoring weight, is used to characterize the degree of direct traction of a node to the absolute target; , representing the absolute target anchoring gain; , representing the relative geometric gain of the graph structure; , representing the weight of the dedicated coupling term; , indicating the first Each node's dedicated coupling item; This provides a memory correction gain. This is to unify the protocol's memory state.

[0100] The first item in the unification agreement This is used to solve the overall drift problem, ensuring that all nodes are eventually pulled by the absolute target point; the second item Used to guarantee relative geometric relationships under communication graph constraints; third item Used to explicitly incorporate cross-layer support relationships and underground flexible morphological constraints; Item 4 Used to improve the dynamic quality during the protocol convergence process.

[0101] For drone nodes, define the dedicated coupling term as:

[0102] ,

[0103] in, To account for the support error between the air layer and the ground layer, zeros are padded at the end of the dedicated coupling term to embed the UAV node into the three-dimensional protocol space (the same method is used for unmanned vehicle nodes and underground flexible tube robot nodes). This is a collection of aerial drone nodes.

[0104] For the autonomous vehicle node, the dedicated coupling term is defined as:

[0105] ,

[0106] in, This represents the support error between the ground layer and the underground layer. This is a set of ground-based unmanned vehicle nodes.

[0107] For the underground flexible tube robot node, the dedicated coupling term is defined as:

[0108] ,

[0109] in: , which represents the stretching error weight in the protocol layer; , which is the bending error weight in the protocol layer; , for the first Individual discrete node stretching error; , for the first Individual discrete node bending error; This is a set of discrete nodes for an underground hose robot.

[0110] The above definition illustrates that although the specific coupling terms of the three types of nodes differ, they are all uniformly written into the same protocol equation. Therefore, the key innovation of this embodiment is not simply putting the three types of intelligent agents together, but constructing a structure of a unified protocol plus specific operators for different nodes at the protocol layer.

[0111] To improve the smoothness and dynamic quality of convergence in the later stages of the unified protocol, the update law for the unified protocol memory state is defined as follows:

[0112]

[0113] in, , is the memory decay coefficient of the protocol layer; , which are protocol layer parameters; It is a hyperbolic tangent function, which in this embodiment operates on a component basis to continuously and boundedly compress the error signal, thereby preventing the protocol layer output from being too sharp.

[0114] In this way, the upper-layer protocol uniformly outputs the reference positions of the three types of nodes. It is important to emphasize that this layer does not directly generate actuator control quantities, but only generates a unified reference that can be implemented by the lower-layer dynamics. This is precisely the fundamental difference between this invention and conventional formation control.

[0115] Step 4: Construct the cross-layer support relationship between the aerial layer and the ground layer, as well as between the ground layer and the underground layer. Here, the aerial layer is the drone, the ground layer is the unmanned vehicle, and the underground layer is the underground flexible hose robot.

[0116] To prevent the upper-layer protocol from degenerating into a regular anchoring formation protocol, this embodiment introduces a cross-layer support relationship.

[0117] Let the first The ground support set corresponding to each drone is: , define the first The current centroid of the drone is:

[0118]

[0119] The corresponding target centroid is:

[0120]

[0121] Define the support error of the air layer to the ground layer as:

[0122]

[0123] in, For the first The horizontal component of the drone protocol reference position For the first The horizontal component of the drone target position; For the first The horizontal component of the autonomous vehicle protocol reference position For the first The horizontal component of the target position of the autonomous vehicle For the number of ground support nodes, This represents the support error of the air layer relative to the ground layer. This formula indicates that it's not simply about getting the drone to a specific point, but rather about maintaining a predefined support geometry between it and its associated ground group.

[0124] Similarly, let the first... The underground support set corresponding to each unmanned vehicle is as follows: , define the first The current underground support center of mass for the unmanned vehicle is:

[0125]

[0126] The target underground support center of mass is:

[0127]

[0128] Define the support error of the ground layer to the underground layer as:

[0129]

[0130] in, For the first The horizontal component of the protocol reference position of each discrete node For the first The horizontal component of the target position of each discrete node This indicates the support error of the ground layer relative to the underground layer. This represents the number of nodes in the underground support set.

[0131] Step 5: Construct an adaptive backstepping sliding mode controller based on the second-order dynamics model of the UAV;

[0132] The upper-layer protocol obtains the drone's reference position. Reference speed and reference acceleration Then, the drone is tracked and controlled at a lower level.

[0133] First, define the drone's position error as:

[0134]

[0135] in, Indicates the first Current location of the drone Deviation relative to the protocol reference position.

[0136] Constructing a virtual desired velocity:

[0137]

[0138] in, , for the first The virtual expected speed of a drone; , for the first The first derivative of the reference position of the drone protocol, i.e., the reference velocity; , which is the position error feedback gain.

[0139] Define the speed error as:

[0140]

[0141] in, The actual speed of the drone The deviation between the actual expected speed and the actual speed.

[0142] Constructing the sliding surface:

[0143]

[0144] in, For sliding mode variables; , where is the coupling coefficient between position error and velocity error. This construction allows position error and velocity error to be integrated into a single robust control variable.

[0145] To compensate for the damping uncertainty of the UAV, an adaptive damping estimation law is designed:

[0146]

[0147] in, , for the first Equivalent damping estimate for a single UAV; , is the adaptive gain; , where is the leakage coefficient; Represents element-wise product; This indicates taking the absolute value of the component.

[0148] Based on the aforementioned position error, velocity error, sliding mode variables, and adaptive damping estimation law, an adaptive backstepping sliding mode controller for the UAV is constructed as the first... The control input for each drone is:

[0149]

[0150] in, For the first Control inputs for a drone; The output of the upper-layer unified protocol The second derivative of the reference position of the UAV protocol, i.e., the reference acceleration; For position error feedback gain; For speed error feedback gain; Represents element-wise product; For sliding mode robustness gain; Boundary layer thickness; It is a saturation function.

[0151] In this control law, The reference acceleration feedforward term is used to improve the trajectory tracking response speed; Used to reduce position tracking errors; Used to suppress velocity tracking errors; Used to compensate for the uncertainty of equivalent damping of drones; This is a sliding mode robust term used to suppress external disturbances and unmodeled dynamic effects, and to reduce control chattering through the boundary layer.

[0152] After the above steps, the UAV can stably track the upper unified reference even in the presence of damping uncertainties and external disturbances.

[0153] Step 6: Construct a continuously normalized kinematic controller based on the nonholonomic kinematic model;

[0154] Autonomous vehicles only track the planar reference portion of the upper-layer protocol. ,make The plane reference velocity is The reference velocity modulus is:

[0155]

[0156] Definition of the first The reference heading angle for the autonomous vehicle is:

[0157]

[0158] in, For the first The plane reference speed of the autonomous vehicle is directional components, For the first The plane reference speed of the autonomous vehicle is Component of direction; It is the arctangent function in the four quadrants.

[0159] Definition of the first The reference angular velocity of the autonomous vehicle is:

[0160]

[0161] To meet the nonholonomic constraints of autonomous vehicles, the tracking error between the actual position of the j-th autonomous vehicle and the protocol reference position given by the upper layer protocol is transformed into the vehicle coordinate system:

[0162]

[0163]

[0164]

[0165] in, This refers to longitudinal error; This refers to lateral error; For heading error; Reference location for driverless car protocols; This refers to the actual location of the driverless vehicle. This represents the actual heading angle of the unmanned vehicle.

[0166] Based on this, the linear velocity and angular velocity control laws for the autonomous vehicle are designed:

[0167]

[0168]

[0169] in, , which is the longitudinal error feedback gain; , which is the lateral error feedback gain; , is the heading error feedback gain; This is a saturation function used to limit the amplitude of the control input; the function and stipulate Used to avoid The values ​​at that time are unstable.

[0170] In linear velocity control law Indicates the main propulsion along the desired direction. Used to eliminate longitudinal errors; in angular velocity control laws It is a reference angular velocity feedforward. Used to eliminate heading errors Used to eliminate lateral errors and avoid numerical singularities.

[0171] Therefore, the autonomous vehicle control input in step six and It is not generated independently, but rather by the first... The horizontal component of the autonomous vehicle protocol reference position and its derivative Driven generation. By stably tracking the reference trajectory of the planar protocol, the autonomous vehicle indirectly executes the absolute target anchoring, graph structure geometry preservation, and ground layer to underground layer support relationships contained in the upper-layer unified protocol.

[0172] In this way, ground-based unmanned vehicles can achieve continuous, smooth, and singular tracking of the trajectory of the upper reference plane under nonholonomic constraints.

[0173] Step 7: Based on the dynamic model of discrete nodes of the underground hose robot, construct an ESO (Extended State Observer) to assist in the optimization of the iterative sliding mode controller;

[0174] The cross-layer support-morphological coupling unified protocol in step four is the first Reference location for the generation protocol of discrete nodes of underground hoses Since the underground flexible tube robot primarily moves within the underground plane, this step extracts the horizontal plane component of the protocol reference position as the lower-level tracking reference for the underground node, denoted as... Its first derivative with respect to time represents the... The reference velocity of each discrete node of the underground hose is denoted as... The above and All values ​​are calculated by the upper-level unified protocol in each control cycle and used as inputs to the extended state observer (ESO) to assist in the optimization of the iterative sliding mode controller.

[0175] First, the location error of the underground discrete node is defined as:

[0176]

[0177] in, For the first The deviation of the actual position of each underground discrete node from the protocol reference position.

[0178] To estimate the combined disturbances in the subsurface environment, an extended state observer (ESO) is constructed:

[0179]

[0180]

[0181] in: , for the first State estimates of discrete underground nodes; , which is the estimate of the comprehensive underground disturbance; , representing the ESO observer gain.

[0182] Because complex disturbances such as soil contact, local friction, and blockage are prone to occur in underground environments, the sliding mode robustness term must be set very large if disturbance estimation is not performed, which can easily lead to input jitter. The role of ESO is to estimate some unknown disturbances in advance and compensate for them in the control.

[0183] Introducing iterative memory state of underground discrete nodes And its dynamic update law is designed as follows:

[0184]

[0185] in, , for the first The iterative memory state of each underground discrete node is obtained online by calculating the position error, stretching error and discrete bending error through the above dynamic update law. The initial value can be set to zero. Its function is to record and smooth historical error information to improve the convergence stability of the underground hose robot's shape tracking process. , where is the memory decay coefficient of the discrete node; , where is the weight of the stretching error term; , where is the weight of the bending error term; , where is the boundary layer parameter.

[0186] This formula accumulates historical information on position error, tension error, and bending error, enabling the controller to not only react to instantaneous errors but also to have iterative correction capabilities.

[0187] Define the underground slipform combination variables as:

[0188]

[0189] in, This is used for memory state weights. Unlike ordinary sliding mode variables that only contain position errors, this embodiment integrates the three types of errors of the underground hose robot into a single sliding mode variable, making the underground controller truly responsible for the overall shape of the hose.

[0190] The initial control inputs for constructing the underground hose robot:

[0191]

[0192] in, , is the position error feedback gain; , is the gain for stretching error feedback; , is the bending error feedback gain; , which is the gain of the sliding mode robustness term; , where is the boundary layer thickness; , which is the memory state feedback gain.

[0193] In this formula, Used to follow protocol references Used to eliminate position errors Used to inhibit local elongation or compression. Used to suppress local angle or bending deviations. Used to enhance robustness against disturbances Used for iterative correction based on historical errors. Used to compensate for overall underground disturbances.

[0194] Finally, to avoid large fluctuations in the control input, the control input is updated smoothly through optimized iterative processes:

[0195]

[0196] in, , representing the weighting coefficient of the current control input; , which represents the weighting coefficient of the control input in the previous moment; , representing the weighting coefficients of the control input in the first two time steps; satisfying ; This is to control the update cycle.

[0197] This smooth update mechanism can significantly reduce the abrupt changes in underground control inputs and improve the stability of the flexible execution process.

[0198] Thus, the underground layer not only enables tracking of protocol references, but also... and The feedback enables synchronous constraints on the local length and curvature distribution of underground hoses.

[0199] Step 8: Construct a cascaded system consisting of an upper-level unified reference system and a lower-level heterogeneous tracking system to achieve stable and collaborative control of heterogeneous intelligent agents at three levels: air, ground, and underground.

[0200] After completing the upper-layer unified protocol design and the design of three types of lower-layer controllers, this invention further constructs a cascaded closed-loop cooperative control structure consisting of an upper-layer unified reference system and a lower-layer heterogeneous tracking system. The upper-layer unified reference system is used to uniformly generate the protocol reference states of the UAV, unmanned vehicle, and underground hose discrete nodes based on the target intelligent agent configuration, cross-layer support relationships, and underground hose morphological constraints; the lower-layer heterogeneous tracking system is used to stably track the upper-layer protocol reference states based on the respective dynamic characteristics of the three types of robots.

[0201] Specifically, within each control cycle, the current state information of discrete nodes—including the aerial UAV, ground-based unmanned vehicle, and underground hose—is first collected. This includes the UAV's position and velocity, the unmanned vehicle's position and heading angle, the underground hose node's position, and the relationships between adjacent nodes. Then, this state information, along with the target agent configuration, cross-layer support relationships, and underground hose morphological constraints, is input into the upper-layer unified reference system. The upper-layer unified reference system calculates and updates the protocol reference state of each type of node based on the cross-layer support-morphological coupling unified protocol.

[0202] For the For each UAV, the upper-level unified reference system outputs its three-dimensional protocol reference position, reference velocity, and reference acceleration, denoted as follows: , , The aforementioned reference values ​​serve as inputs to the UAV's adaptive backstepping sliding mode controller, used to calculate the UAV's control inputs. .

[0203] For the Since the autonomous vehicle only moves within the ground plane, the lower-level controller only reads the horizontal plane component of the upper-level protocol reference state. , ,in, Indicates the first The planar protocol reference position of an autonomous vehicle This represents the reference plane velocity. The continuously normalized kinematics controller of the unmanned vehicle calculates the linear velocity input based on the above reference value. and angular velocity input .

[0204] For the Each underground flexible tube has discrete nodes; the lower-level controller reads the underground horizontal reference position and reference velocity output by the upper-level unified reference system. , Based on the locations of adjacent underground nodes, tensile errors, discrete bending errors, and ESO disturbance estimation results, the control inputs of the underground nodes are calculated. .

[0205] Thus, the upper-level unified reference system and the lower-level heterogeneous tracking system form the following cascaded relationship: target agent configuration, cross-layer support relationship, morphological constraints → upper-level unified protocol → protocol reference state → lower-level heterogeneous controller → actual motion of the three types of robots → state feedback. The upper-level unified protocol does not directly output actuator control quantities, but rather outputs a protocol reference state that is commonly followed by the three types of robots. The lower-level heterogeneous controllers do not redesign the formation relationship, but rather track the protocol reference state based on their respective dynamic models. Although the UAV, unmanned vehicle, and underground flexible tube robot use different controllers, their reference states are all generated by the same upper-level unified protocol. Therefore, the three types of robots exhibit heterogeneous tracking at the execution layer and unified formation at the coordination layer.

[0206] Furthermore, after the lower-level controller executes, the actual states of the three types of robots are fed back to the upper-level unified reference system in the next control cycle to update the position error, cross-layer support error, underground hose tension error, and discrete bending error. Based on this, the upper-level unified reference system recalculates the protocol reference state, and the lower-level heterogeneous tracking system generates control input again, thus forming a closed-loop collaborative control process of state acquisition unified reference generation - heterogeneous control execution - state feedback correction.

[0207] Through the aforementioned cascaded closed-loop mechanism, the collaborative relationship between the aerial UAV layer, the ground-based unmanned vehicle layer, and the underground flexible hose robot layer is not achieved by simply connecting three independent controllers in parallel. Instead, it is uniformly organized by an upper-layer unified protocol, and then executed separately by lower-layer heterogeneous controllers. Since the unified protocol simultaneously includes absolute target anchoring terms, graph structure relative geometry terms, cross-layer support coupling terms, underground flexible hose morphology constraint terms, and protocol memory correction terms, each robot in the lower layer can maintain the three-layer target configuration, inter-layer support relationship, and underground flexible hose morphology while tracking its own reference trajectory, thus achieving stable collaborative control of the three-layer heterogeneous intelligent agents in the air, ground, and underground layers.

[0208] In a preferred embodiment, the stability of this invention can be described from three aspects: the stability of the upper-layer unified protocol, the stability of the lower-layer heterogeneous tracking, and the stability of the overall cascaded system. For ease of description, let the stacking vector of all protocol reference position errors be:

[0209]

[0210] in, The overall vector is formed by stacking the protocol reference positions of all agents in a fixed order. The overall target vector is formed by stacking the target locations of all intelligent agents in the same order. Specifically, both are arranged in the order of "UAV node - unmanned vehicle node - underground hose discrete node". Generated online by the upper-layer unified protocol. Determined by the preset target configuration, because and Since the nodes share the same order and coordinate dimensions, the overall protocol error can be directly obtained by subtraction; let the communication graph be... The corresponding Laplace matrix is The anchoring matrix is ,and Zhengding, among which , For the first The anchoring weight of each node represents the degree to which that node is directly influenced by the target location. If a node can directly obtain the target location information, then... ,otherwise Set up dedicated coupling terms It includes cross-layer support error, underground hose tension error and discrete bending error, which are bounded and locally continuous Lipschitz functions.

[0211] (1) Stability description of the upper-layer unified protocol

[0212] According to the unified protocol of this embodiment, the protocol error system can be written as follows:

[0213]

[0214] in, , which is the absolute target anchoring gain, used to adjust the strength of the node's convergence to the target position; , which is the relative geometric gain of the graph structure, used to adjust the strength of maintaining the relative positional relationship between nodes; , which is a dedicated coupling term gain used to adjust the impact of cross-layer support error and underground hose morphology error on the protocol; This is the protocol memory correction gain, used to adjust the smoothing effect of the memory state on the protocol convergence process. To ensure the unified protocol for upper-layer memory states, the following conditions must be met:

[0215]

[0216] Selecting Lyapunov candidate functions:

[0217]

[0218] in, , for design parameters; Differentiation yields:

[0219]

[0220] Substituting the protocol error system and the memory state update law, we have:

[0221]

[0222] because Positive definite, there exists a constant. , so that:

[0223]

[0224] Meanwhile, both the cross-layer support item and the underground flexible tube morphology item are constructed from support error, tensile error, and bending error. The cross-layer support item specifically refers to the support error item in the UAV node and the unmanned vehicle node. , The underground hose morphology term specifically refers to the morphological coupling term in the underground hose node. The above two items together constitute a dedicated coupling term. The directions of the cross-layer support term and the underground flexible conduit morphology term are consistent with the error correction direction, therefore satisfying:

[0225]

[0226] Using Young's inequality and We can obtain:

[0227]

[0228]

[0229] in, , , is an auxiliary positive parameter in the Young's inequality, used to define the cross terms and ensure that the negative definite term in the Lyapunov derivative is dominant; therefore, we have:

[0230]

[0231] right , , , , , Take values ​​such that the Lyapunov derivative is... and The coefficients before are all negative; specifically, this can be achieved by increasing... , To improve target anchoring and relative geometric constraint strength, by increasing Increase the intensity of memory state decay and select By applying Young's inequality to the cross terms, we can ensure that ,and:

[0232]

[0233]

[0234] in, , is the protocol error attenuation coefficient. , where is the protocol memory state decay coefficient, and both are determined by the control gain and auxiliary parameters, used to ensure the convergence of upper-layer protocol errors and memory states; then:

[0235]

[0236] Therefore, the error of the upper-layer unified protocol system can be understood. and memory state Bounded, and ,in, Describes a bounded function space, if a certain signal belongs to This indicates that the signal is bounded over the entire time interval, meaning its amplitude will not increase indefinitely; combined with the boundedness of the right-hand side of the error system, we can obtain... Bounded. According to Barbalat's lemma, the protocol error... It converges to zero or to a small neighborhood determined by a bounded perturbation. Therefore, the upper-level unified protocol can generate a bounded and asymptotically convergent protocol reference state.

[0237] (2) Description of the stability of the unmanned aerial vehicle layer

[0238] For the For a drone, the position tracking error is defined as:

[0239]

[0240] in, The actual location of the drone. The reference location generated for the upper-layer protocol.

[0241] Constructing a virtual desired velocity:

[0242]

[0243] And define the speed error:

[0244]

[0245] Construct sliding mode variables:

[0246]

[0247] Selecting Lyapunov candidate functions:

[0248]

[0249] in, The equivalent damping estimation error, The actual equivalent damping parameters of the UAV are usually unknown but bounded. For the first An equivalent damping estimate for the UAV is used to compensate for the damping uncertainty of the UAV.

[0250] Depend on Combining the second-order dynamic model of the UAV, the adaptive damping estimation law, and the adaptive backstepping sliding mode controller, the following is performed: Taking the derivative, we get:

[0251]

[0252] in, This is the sum of external disturbances and unmodeled dynamics of the UAV, which may include wind disturbances, airflow changes, load changes, sensor noise, and model simplification errors, etc. This term is usually difficult to determine precisely, and this invention treats it as an unknown bounded disturbance, i.e., a constant exists. , making And it is suppressed and compensated by the sliding mode robustness term; Let be a bounded constant introduced by the adaptive leakage term.

[0253] If the disturbance satisfies , and select ( Given a known or estimated upper bound for the external disturbances and unmodeled dynamics of the UAV (used to characterize the maximum range of the disturbance amplitude), then:

[0254]

[0255] in, , This introduces a small constant to the boundary layer. Therefore:

[0256]

[0257] Therefore, it can be seen that the position error, velocity error and damping estimation error of the UAV are all eventually bounded; when the external disturbance is small or the boundary layer parameters are sufficiently small, the UAV can asymptotically track the upper unified protocol reference.

[0258] (3) Stability description of unmanned vehicle layer

[0259] For the For an unmanned vehicle, its tracking error is transformed into the vehicle coordinate system to obtain the longitudinal error. lateral error and heading error The error dynamics model for autonomous vehicles is as follows:

[0260]

[0261]

[0262]

[0263] in, The reference linear velocity for the autonomous vehicle. The reference angular velocity for the autonomous vehicle is... Input the actual angular velocity of the autonomous vehicle. This refers to the longitudinal position error in the vehicle coordinate system. This represents the lateral position error in the vehicle coordinate system. The Lyapunov function is selected:

[0264]

[0265] in, This is the feedback gain for the heading error of the unmanned vehicle;

[0266] Differentiating the Lyapunov function yields:

[0267]

[0268] After differentiation, substituting into the continuous normalized kinematic control law of the autonomous vehicle:

[0269]

[0270]

[0271] We can obtain:

[0272]

[0273] Since for any real number All of them have:

[0274]

[0275] and This holds true within a small-angle neighborhood, therefore:

[0276]

[0277] Therefore, it can be concluded that the tracking error of the autonomous vehicle is bounded. Combining the LaSalle invariant principle, under the conditions that the reference speed is not permanently zero and input saturation does not violate reachability, , and It converges to zero or to the zero neighborhood. Therefore, the autonomous vehicle can stably track the upper-layer protocol reference trajectory while satisfying nonholonomic constraints.

[0278] (4) Stability description of underground flexible tube robot layer

[0279] For the For each underground discrete node, the position tracking error is defined as:

[0280]

[0281] Let the tensile error be Discrete bending error is The underground iterative memory state is The ESO integrated disturbance estimation error is Then construct the underground sliding mode combination variables:

[0282]

[0283] Selecting Lyapunov candidate functions:

[0284]

[0285] in, , where is the weight of the perturbation estimation error. For Taking the derivative and substituting it into the ESO-assisted optimization iterative sliding mode control law for the underground hose robot, we get:

[0286]

[0287] in, This refers to the bounded small quantity introduced by the boundary layer, the perturbation estimation residual, and the discrete update. This is the position error attenuation coefficient. This is the coefficient for attenuation of stretching error. Let be the bending error attenuation coefficient. Using Young's inequality, we have:

[0288]

[0289] in, This is an auxiliary positive parameter introduced in Young's inequality;

[0290] Meanwhile, the ESO error system, by appropriately selecting the observer gain , Time has boundaries, and in engineering, it is usually taken as... , Furthermore, the gain is adjusted based on the sampling period, measurement noise, and the rate of change of disturbance; too low a gain will result in a slow estimation response, while too high a gain may amplify the measurement noise. That is:

[0291]

[0292] in, The known or estimated upper bound for the overall disturbance of the underground hose robot;

[0293] Therefore, we can conclude that:

[0294]

[0295] Where, constant , This is a bounded perturbation term.

[0296] Therefore, it can be seen that the positional error, tensile error, bending error and iterative memory state of the underground hose node are all eventually bounded, and the final error neighborhood can be further compressed by increasing the feedback gain, improving the ESO estimation accuracy and reducing the boundary layer thickness.

[0297] Meanwhile, the control input for the underground flexible tube robot is:

[0298]

[0299] and , Therefore, the optimized iterative smooth update is a convex combination of control inputs, which does not violate the boundedness of the closed-loop system and can reduce input mutations and chattering during the execution of underground flexible nodes.

[0300] (5) Stability description of the overall cascaded system

[0301] The above analysis shows that the upper-layer unified protocol can generate a bounded and convergent protocol reference state; the lower-layer controllers of the UAV, unmanned vehicle, and underground hose robot can all achieve stable tracking of their respective reference states. Since the lower-layer tracking error system takes the bounded protocol reference and its derivative as input, and each type of lower-layer closed-loop error is uniformly and ultimately bounded, the stability conclusion of the cascaded system indicates that the overall closed-loop system of the air-ground-underground three-layer heterogeneous intelligent agent is uniformly and ultimately bounded. Furthermore, when external disturbances, observation errors, and boundary layer parameters tend to zero, the protocol error, position tracking error, relative edge error, underground hose tension error, and bending error can all tend to zero, thereby achieving stable cooperative control of the three-layer heterogeneous intelligent agents.

[0302] In a specific simulation embodiment, the system consists of 3 aerial drones, 4 ground unmanned vehicles, and 5 underground flexible tube robots. The total simulation time is set to 30 seconds, and the integration step size is set to 0.02 seconds. The fixed display depth of the underground layer is set to... m, ground level is located Plane, aerial layer located Space. All three types of intelligent agents are modeled and displayed in a unified three-dimensional coordinate system.

[0303] In this embodiment, the air-level target configuration is a triangular support structure, and its target positions are as follows:

[0304]

[0305] The ground-level target configuration is located in the intermediate plane, and its target positions are as follows:

[0306]

[0307] The underground target configuration is a discrete flexible hose chain with an undulating shape, and its target locations are as follows:

[0308]

[0309] The system is initially set to a dispersed state, where the initial positions of the aerial drones are as follows:

[0310]

[0311] The initial positions of the ground-based unmanned vehicles are as follows:

[0312]

[0313] The initial positions of the underground hose robot are as follows:

[0314]

[0315] The significant spatial deviation between the initial and target positions allows for the verification of the formation building capability, cross-layer support maintenance capability, and underground hose morphology maintenance capability of the proposed control method. The simulation script also explicitly provides the target agent configuration, initial conditions, and cross-layer support mapping relationship.

[0316] In terms of controller configuration, the upper layer adopts a unified cross-layer support-morphological coupling protocol, which includes absolute target anchoring terms, graph structure relative geometry terms, cross-layer support coupling terms, underground hose tension and bending morphological terms, and protocol memory correction terms. The lower layer is configured with controllers based on different robot dynamics characteristics: the UAV uses an adaptive backstepping sliding mode controller to handle second-order dynamics, damping uncertainties, and external disturbances; the unmanned vehicle uses a continuously normalized nonholonomic kinematics controller to satisfy linear velocity and angular velocity constraints and avoid heading control singularities; the underground hose robot uses an ESO-assisted optimization iterative sliding mode controller to compensate for underground integrated disturbances and simultaneously suppress position, tension, and bending errors. The simulation script correspondingly sets up three types of lower-layer controllers: the adaptive backstepping sliding mode controller, the continuously normalized kinematics controller, and the ESO-assisted optimization iterative sliding mode controller.

[0317] Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 The diagram illustrates the formation process. In each diagram, red nodes represent drone nodes, blue nodes represent unmanned vehicle nodes, and green nodes represent underground flexible tube robot nodes. Figure 2 It can be seen that, in At time s, the three layers of intelligent agents—airborne, ground-based, and underground—are in a decentralized state; as the control method progresses, the system... s、 s and The structure gradually forms a transitional configuration with interlayer support relationships. (See also...) Figure 3 , Figure 4 , Figure 5 ;arrive At time s, the three-layered intelligent agent basically converges to the vicinity of the target formation, see [reference]. Figure 6 The results demonstrate that the upper-layer unified protocol can simultaneously coordinate absolute target anchoring, relative geometry preservation, and cross-layer support relationships, enabling the three-layer heterogeneous agents to gradually form a target formation from an initial dispersed state.

[0318] Figure 7 The final 3D spatial layered formation result is shown. In the figure, red nodes represent UAV nodes, blue nodes represent unmanned vehicle nodes, and green nodes represent underground flexible tube robot nodes. Figure 7As can be seen, the aerial drone layer ultimately resides in the upper space, the ground-based unmanned vehicle layer remains stably positioned in the middle plane, and the underground flexible robot layer remains at a fixed depth underground, forming a pre-defined flexible chain shape. This result demonstrates that the present invention not only achieves position convergence in the conventional sense but also maintains the spatial hierarchy and support relationship between the aerial layer, ground layer, and underground layer, preventing inter-layer chaos or overall drift of the three intelligent agents during unified control.

[0319] Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 , Figure 13 The convergence curves for error, morphology, and control input are shown. Figure 8 The maximum morphological error in the model is used to describe the maximum deviation of each agent from the target configuration. Figure 9 The absolute position root mean square error is used to describe the overall convergence of all nodes relative to the target position; Figure 10 The root mean square error of the edges in the communication topology is used to describe the relative distance preservation effect of the edges. Figure 11 The root mean square error of underground hose tension is used to describe the elongation or compression deviation between adjacent underground nodes. Figure 12 The root mean square error of the underground hose bending error is used to describe the deviation between the local curvature of the hose chain and the target shape. Figure 13 The maximum control input norm is used to describe the boundedness of the control inputs of the three types of actuators. In the simulation, the above indicators gradually decay over time and eventually remain within a small range, indicating that the present invention can achieve convergence of overall formation error, preservation of topological edge constraints, preservation of underground hose shape, and stability of control input. The simulation script calculates the maximum shape error, root mean square error of absolute position, root mean square error of edge, root mean square error of underground hose tension, root mean square error of underground hose bending, and maximum control input norm.

[0320] In summary, this embodiment demonstrates that the upper-layer unified protocol + lower-layer heterogeneous actuator control architecture proposed in this invention can enable UAVs, unmanned vehicles, and underground flexible hose robots to gradually converge to the target three-layer configuration even under initially significantly dispersed conditions. Simultaneously, the system can maintain air-to-ground and ground-to-underground cross-layer support relationships and effectively suppress tensile and bending errors in the underground flexible hose chain. Therefore, this invention not only achieves stable formation control of three-layer heterogeneous intelligent agents but also enables collaborative control under underground flexible morphological constraints, verifying the feasibility of this method in complex three-dimensional collaborative operations.

[0321] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A three-dimensional heterogeneous intelligent agent cooperative control method for unmanned aerial vehicles, unmanned vehicles, and underground flexible tube robots, characterized by: In the host computer, a cross-layer support-morphological coupling unified protocol is constructed, which simultaneously includes absolute target anchoring terms, graph structure relative geometry terms, dedicated coupling terms and memory correction terms, to serve as a reference position for generating protocols for three types of heterogeneous intelligent agents. An adaptive backstepping sliding mode controller is constructed based on the UAV's protocol reference position and its dynamic model to control the UAV. The unmanned vehicle is controlled by a continuously normalized kinematic controller constructed based on its protocol reference position and dynamic model. The underground flexible tube robot is controlled by an expanded state observer-assisted optimization iterative sliding mode controller constructed based on its protocol reference position, dynamic model, tension error, and bending error. This achieves stable collaborative control of the three heterogeneous intelligent agents in the air, ground, and underground layers. The cross-layer support-morphological coupling unified protocol, which includes absolute target anchoring terms, graph structure relative geometry terms, dedicated coupling terms, and memory correction terms, is as follows: in, Indicates the first The protocol reference location of each node. Indicates the first The protocol reference location of each node. Indicates the first The target location of each node Indicates the first The target location of each node For communication adjacency matrix elements, Indicates the anchor weight. Indicates the absolute target anchoring gain. This represents the relative geometric gain of the graph structure. Indicates the weight of the specific coupling term. Indicates the first Each node's dedicated coupling item, To correct the gain for memory, To unify the protocol's memory state; The update law for the unified protocol's memory state is: in, The memory decay coefficient of the protocol layer. For protocol layer parameters, It is the hyperbolic tangent function; To prevent the unified agreement from degenerating into a common anchoring formation agreement, cross-layer support relationships are constructed between the air layer and the ground layer, as well as between the ground layer and the underground layer. The cross-layer support relationship between the aerial layer and the ground layer is as follows: in, This represents the support error of the air layer relative to the ground layer. For the first The horizontal component of the drone protocol reference position For the first The horizontal component of the drone target position, the first The current center of gravity of the drone , No. The centroid of the drone target , For the first The horizontal component of the autonomous vehicle protocol reference position For the first The horizontal component of the target position of the autonomous vehicle For the number of ground support nodes, For ground support assembly; The cross-layer support relationship between the ground layer and the underground layer is as follows: in, This indicates the support error of the ground layer relative to the underground layer, the first... The current underground support center of gravity of the unmanned vehicle , No. The underground support center of mass for the unmanned vehicle target , For the first The horizontal component of the protocol reference position of each discrete node For the first The horizontal component of the target position of each discrete node. The number of underground support nodes. For underground support assembly; The continuously normalized kinematics controller is designed to improve the linear velocity of the unmanned vehicle. and angular velocity : in, For longitudinal error feedback gain, For lateral error feedback gain, For heading error feedback gain, It is a saturation function. , For reference velocity modulus, For longitudinal error, For lateral error, For heading error, Used as reference angular velocity; The expanded state observer-assisted optimization iterative sliding mode controller constructs the original control input for the underground hose robot. ,in, For the first Reference velocity feedforward for discrete underground nodes For position error feedback gain, For the stretching error feedback gain, For bending error feedback gain, For the gain of the sliding mode robustness term, Boundary layer thickness, For memory state feedback gain, For the extended state observer to estimate disturbances, the location error of the underground discrete nodes. Indicates the first The deviation of the actual position of each underground discrete node from the protocol reference position. For the first Stretching error of a discrete node For the first Bending error of discrete nodes, underground sliding mode combination variables underground discrete node location error Indicates the first The deviation of the actual position of each underground discrete node from the protocol reference position. For the weight of the stretching error term, For the weight of the bending error term, For the weights of the memory states, Iterative memory of the state for underground discrete nodes; The original control input Employ optimized iterative smooth updates: in, The weighting coefficients for the current control input. The weighting coefficients for controlling the input in the previous time step. The weighting coefficients for controlling the input in the first two time steps. To control the update cycle, For the first The control input for each discrete node.

2. The three-dimensional heterogeneous intelligent agent cooperative control method according to claim 1, characterized in that, The adaptive backstepping sliding mode controller is used as the first Control input of a drone ,in For the first The control input for a drone The output of the upper-layer unified protocol The second derivative of the reference position of the drone protocol, For position error feedback gain, For speed error feedback gain, Represents element-wise product. For the gain of the sliding mode robustness term, Boundary layer thickness, The position error of the UAV is a saturation function. Indicates the first The deviation of the current position of each drone from the protocol reference position. For the speed error of the drone, Indicates the first The velocity vector of the drone For sliding mode variables, For the first The equivalent damping estimate of a drone, and Satisfying the adaptive damping estimation law , For adaptive gain, Leakage coefficient, This indicates taking the absolute value of the component.

3. The three-dimensional heterogeneous intelligent agent cooperative control method according to claim 1, characterized in that, The tensile error of the discrete nodes of the underground flexible tube robot is: in, For the first The position vectors of discrete nodes. For the first The position vectors of discrete nodes. Indicates the first The set of neighboring nodes of a discrete node in a hose robot It is a constant. For discrete nodes and The target distance, and , For the first The target location of each discrete node. For the first The target location of each discrete node.

4. The three-dimensional heterogeneous intelligent agent cooperative control method according to claim 3, characterized in that, The bending error of the discrete nodes of the underground hose robot is: in, For the first The position vectors of discrete nodes. For the first The position vectors of discrete nodes. For the first The target location of each discrete node. For the first The target location of each discrete node.

5. The three-dimensional heterogeneous intelligent agent cooperative control method according to claim 1, characterized in that, The dynamic model of the UAV adopts a second-order dynamic model with velocity damping and external disturbance terms, the dynamic model of the unmanned vehicle adopts a nonholonomic kinematic model, and the dynamic model of the underground flexible tube robot adopts a discrete node dynamic model. The discrete nodes are embedded in a three-dimensional common coordinate system to realize a unified spatial expression of the three-layer heterogeneous intelligent agents.

Images