Multi-machine cooperation hoisting method

Abstract

The invention discloses a multi-unmanned-aerial-vehicle cooperative hoisting method and system, and the method comprises the steps: firstly, enabling each unmanned aerial vehicle to operate an extended state observer, regarding internal and external disturbances (such as airflow, unknown load parameters, internal coupling force and the like) as a lumped disturbance, and carrying out the real-time estimation; and then, through a communication network operation consensus algorithm between the unmanned aerial vehicles, fusing the local disturbance estimation values of the unmanned aerial vehicles into a global consistent system-level disturbance estimation. And finally, each unmanned aerial vehicle actively counteracts the influence of disturbance by using the accurate disturbance estimation value in a feedforward compensation controller, and accurately tracks the respective target trajectory in combination with feedback control. According to the method, active observation of disturbance, distributed fusion and feed-forward compensation are combined, dependence on an accurate system dynamic model is eliminated, and trajectory tracking accuracy, stability and stability of the system in a complex environment are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a multi-UAV collaborative hoisting method. Background Technology

[0002] With the rapid development of drone technology, its application scenarios have expanded from simple aerial photography and reconnaissance to complex logistics transportation, collaborative operations, and other fields. Especially when handling loads that are too large or too heavy, using multiple drones for collaborative lifting has become a highly promising solution.

[0003] In the prior art, the concept of multi-drone cooperative transportation has emerged. For example, US Patent US20160378108A1 discloses a swarm configuration that allows multiple smaller drones to couple together to form a swarm for carrying larger or heavier items. This approach mainly focuses on the macroscopic processes of physical connection, navigation, and decoupling between drones, but it does not delve into the complex, coupled, and nonlinear dynamics problems that arise during cooperative flight, nor does it provide a robust control strategy that can effectively suppress various disturbances at the low-level control layer.

[0004] Achieving precise control of such multi-agent systems is extremely complex. Existing research has employed various control strategies, but all have certain limitations:

[0005] Classical and geometric controllers: Proportional-integral-derivative (PID) controllers and geometric controllers are widely used due to their simple structure. However, given the inherent high nonlinearity and strong coupling characteristics of multi-UAV-load systems, especially when load parameters are unknown or there are strong external disturbances such as gusts of wind, the performance of these controllers will drop sharply, making it difficult to guarantee system stability and trajectory tracking accuracy.

[0006] Distributed control architecture: To improve system scalability and stability, distributed control schemes have become mainstream. For example, the distributed strategy proposed by Chopra and Ghose uses sensors mounted on the payload to provide feedback information to each drone to maintain formation. While this type of approach improves the coordination between drones, its underlying control law is usually still a relatively simple PD controller, lacking the ability to perform optimal control in complex environments and handle multiple constraints (such as obstacle avoidance and motor saturation).

[0007] Model-based control: Model predictive control (MPC) has been applied in the field of UAV control due to its ability to optimize future trajectories online and explicitly handle system constraints. However, the performance of MPC is highly dependent on the accuracy of its internal predictive model. In multi-UAV collaborative lifting scenarios, establishing an accurate system dynamics model is a major challenge. Key parameters such as the mass, inertia, and center of mass of the load may be unknown, or even time-varying (e.g., the sloshing of internal liquids when transporting liquid loads). Furthermore, complex aerodynamic effects and external environmental disturbances (such as wind fields) are extremely difficult to model accurately. This strong dependence on model accuracy makes standard MPC methods very vulnerable to real-world uncertainties.

[0008] In summary, a fundamental technical contradiction exists in the existing technology: On the one hand, to achieve high-precision trajectory tracking and high stability (improved characteristics) in multi-UAV-payload systems, the control system requires an extremely accurate system dynamics model and tightly coupled, high-bandwidth control actions between UAVs. On the other hand, pursuing a high-precision model greatly increases system complexity (deteriorated characteristics), requires significant system identification work, and the model may still become mismatched due to unforeseen disturbances or load changes, which in turn reduces the system's reliability and stability. Furthermore, the tightly coupled control scheme is highly sensitive to communication delays, packet loss, or individual UAV failures, which also leads to a decrease in overall system reliability.

[0009] Existing methods force designers to compromise between performance and stability: either use a simplified model, sacrificing control performance and stability; or try to build a complex, high-fidelity model, which is impractical in engineering and reduces the system's adaptability to change. Summary of the Invention

[0010] The purpose of this invention is to provide a multi-machine collaborative hoisting method to solve the problems existing in the background art.

[0011] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0012] This invention relates to a multi-drone cooperative hoisting method, applied to a hoisting system consisting of at least two drones, a load connected to the drones via cables, and a communication network between the drones. The method includes the following steps: S100: Each drone acquires its own current state information, which includes at least position information and velocity information; S200: Each drone, based on its own current state information, its own control input, and disturbance estimation information received from at least one neighboring drone through the communication network, runs a distributed extended state observer to generate its own lumped disturbance estimate, wherein the state equation of the distributed extended state observer includes a consensus fusion term to make the lumped disturbance estimates generated by each drone tend to be consistent; S300: Each drone determines its own target trajectory according to a preset load target trajectory and the geometric configuration of the hoisting system; S400: Each drone, based on its own target trajectory, its current state information, and its lumped disturbance estimate, calculates and generates control commands to drive its own movement.

[0013] Specifically, for the UAV numbered i, the state equation of its distributed extended state observer is as follows: ,

[0014] in, The actual location of drone i Here is the estimated position of UAV i; Here is the speed estimate for drone i; For the acceleration control term of UAV i; Let be the lumped disturbance estimate for UAV i; , , The observer gain coefficient; This represents the consensus gain coefficient. Let i be the set of neighboring drones; The elements of the communication topology adjacency matrix represent the communication link status between UAV i and UAV j. Let be the lumped disturbance estimate received from neighboring drone j.

[0015] Furthermore, in step S400, the calculation of the control command includes: offsetting the influence of the lumped disturbance estimate by a disturbance feedforward compensation term; and driving the UAV to track the target trajectory based on the error between the target trajectory and the current state information by a feedback control term.

[0016] Specifically, for the UAV numbered i, its control commands are... The specific calculation formula is as follows: ,

[0017] in, For the quality of drone i; , , These are the target position, target velocity, and target acceleration of UAV i, respectively; The actual speed of drone i; and This is the gain matrix of the feedback controller; This is the lumped disturbance estimate generated by the distributed extended state observer.

[0018] Specifically, the lumped disturbance includes at least one of the following: external environmental disturbance, internal dynamic coupling of the hoisting system, and uncertainty in the UAV's own model.

[0019] Specifically, the observer gain coefficient , , It is determined by using the pole placement method based on the preset observer bandwidth.

[0020] Furthermore, the method further includes: after step S400, each UAV decomposes the control command into a desired resultant force and a desired attitude, and converts the desired resultant force and desired attitude into control signals for the UAV actuators through an inner-loop attitude controller.

[0021] Compared with the prior art, the advantages of this invention are:

[0022] This invention fundamentally solves the technical contradiction in multi-UAV collaborative hoisting where high-precision control relies on an accurate model and the resulting stability degradation. This is achieved through the synergistic effect of distributed extended state observers and feedforward compensation control. Its core advantages are: First, each UAV's local extended state observer treats all internal and external uncertainties (such as unknown load parameters, internal dynamic coupling, and external gusts) as a single "lumped disturbance" and performs real-time, high-bandwidth estimation. This eliminates the need for an accurate prior dynamic model, achieving a high degree of decoupling between the control law and the complex model, thus endowing the system with extremely strong stability to cope with sudden disturbances and time-varying characteristics. Second, the consensus fusion mechanism, as a key link in the collaboration, ensures that the entire UAV swarm can make decisions based on a unified and highly complete disturbance information, avoiding control conflicts caused by the limitations of a single observation and greatly improving the overall reliability and collaborative consistency of the system. The most crucial collaborative effect is that the calculation of control commands directly utilizes the fused disturbance estimate to form the feedforward compensation term. This design can proactively and in real-time offset the impact of disturbances, rather than passively waiting for errors to occur before feedback correction. This close collaboration of "estimation-fusion-compensation" enables the system to maintain excellent trajectory tracking accuracy and system stability even when faced with severe unknown disturbances, ultimately achieving high-performance collaborative hoisting operations without sacrificing stability. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer and more complete, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0026] This invention aims to solve the problem of achieving high-precision and high-stability collaborative lifting and transportation of multiple UAVs under conditions of significant model uncertainty and external disturbances, while avoiding the technical challenge of constructing complex and fragile high-fidelity system models. This invention addresses the technical contradiction that improving trajectory tracking stability leads to a decrease in system stability and reliability due to increased model complexity.

[0027] Example 1:

[0028] refer to Figure 1 The method disclosed in this invention is applied to a multi-UAV collaborative lifting system. A typical physical configuration of this system includes:

[0029] Drone Swarm: Composed of at least two (N≥2) drones. In this embodiment, a quadcopter drone is used as an example, but the invention is not limited to this model. Each drone is equipped with a core flight controller that performs low-level flight control tasks, an accompanying computer that runs advanced algorithms, a sensor suite for state awareness (such as IMU, RTK-GPS), and a wireless communication module for data interaction. Load: The target object to be transported. Its physical parameters, such as mass, inertia, and center of mass position, may be partially or completely unknown. Cable (or rigid connector): Used to connect the drones and the load, serving as the medium for force transmission. Communication Network: Built among the N drones to exchange key data such as state information, control commands, or estimates. This network can be modeled as an undirected or directed graph G=(V,E), where the vertex set V={1,2,…,N} represents the N drones, and the edge set E represents the communication links between the drones. This method is applied to a hoisting system consisting of at least two drones, a load connected to the drones by cables, and a communication network between the drones. The detailed description of the system architecture above forms the basis for the physical and information interaction in implementing the method of this invention.

[0030] The present invention provides a multi-machine cooperative hoisting method, the core of which lies in endowing a relatively simple feedback controller with the ability to achieve high-precision and high-stability control through a distributed cooperative disturbance observation and compensation mechanism. The method mainly includes the following steps: S100: state information acquisition; S200: distributed lumped disturbance estimation; S300: target trajectory determination; S400: control command calculation.

[0031] S100: Each UAV acquires its own current state information. This step is fundamental to all subsequent estimations and control. For each UAV in the hoisting system (numbered i, where i∈{1,…,N}), its own motion state needs to be acquired in real time and accurately. The current state information includes at least position information and velocity information.

[0032] Location information This represents the three-dimensional coordinates of the UAV i's center of mass in the global inertial coordinate system. This information is typically obtained through a high-precision differential global positioning system (RTK-GPS), achieving centimeter-level positioning accuracy. In environments where GPS signals are obstructed, it can also be obtained by fusing visual odometry (VIO), lidar odometry (LIO), and IMU data. Speed ​​information : Represents the three-dimensional velocity of UAV i's center of mass in the global inertial coordinate system. This information can be provided directly by RTK-GPS, or estimated by differentially analyzing continuous position information and combining it with a filter (such as a Kalman filter) to reduce the impact of measurement noise. In specific implementations of this invention, this step is typically performed at a relatively high frequency (e.g., 50Hz-100Hz) on the accompanying computer of each UAV to ensure the timeliness of the status information. The acquired... It will serve as a key input for subsequent steps S200 and S400.

[0033] S200: Running the Distributed Extended State Observer to Generate Lumped Perturbation Estimates. This step is one of the core innovations of this invention, aiming to address model uncertainty and external perturbation issues. Each UAV i in the system runs a Distributed Extended State Observer (DESO) independently and in parallel.

[0034] 1. The concept of lumped disturbance: Traditional control methods rely on accurate system dynamics models. However, accurate models for multi-UAV hoisting systems are needed. Extremely complex. Among them, This represents the internal dynamic coupling force caused by loads, cables, etc. This represents external environmental forces such as wind resistance. The precise form and parameters of these forces (such as load mass and air density) are usually unknown. This invention draws on the idea of ​​Active Disturbance Rejection Control (ADRC), treating all "undesired" or "unknown" dynamic terms as a single lumped disturbance. The translational dynamics of UAV i can be simplified to a nominal model (such as Newton's second law). ) and an unknown lumped disturbance Combinations: The lumped disturbance includes at least one of the following: external environmental disturbances (such as gusts of wind, ground effect), internal dynamic couplings of the hoisting system (such as load swaying, interaction forces between UAVs), and uncertainties in the UAV's own model (such as mass parameter errors, rotor efficiency variations, sensor bias). In this way, the modeling problem of a complex system is transformed into a problem involving unknown disturbances. The problem of real-time estimation.

[0035] 2. Construction of the Distributed Extended State Observer (DESO) to estimate this lumped disturbance Each UAV i constructs an Extended State Observer (ESO). This observer not only estimates the system's normal state (position, velocity) but also takes into account lumped disturbances. This is treated as a new, unestimated "state" of the system, thus enabling real-time observation of it. Each UAV, based on its own current state information... Each of their own control inputs (or its equivalent acceleration term) And disturbance estimation information received from at least one neighboring drone via a communication network. Run DESO. The specific form of the DESO state equation for UAV numbered i is: ,

[0036] The physical meaning and function of each variable and term are as follows: The actual position of UAV i (measured from S100). Let be the estimated position of UAV i. Here is the speed estimate for drone i. The acceleration control term for UAV i is the control command calculated in the previous cycle. Divide by the nominal mass of the drone Received. This is the known input. The lumped disturbance estimate for UAV i is the core output of this observer. It is the output error of the position measurement, and it is the core signal that drives the observer to make corrections. These are the observer gain coefficients. They determine the speed and strength at which the observer corrects its state estimate based on the output error. A detailed interpretation of the above equations follows:

[0037] First line (location estimation): This is a classic state observer form. It shows that the rate of change of the position estimate equals the rate of change of the velocity estimate, plus a correction term proportional to the position measurement error. When estimating the position... Deviation from actual position When that happens, the correction will "pull" it back to the correct value.

[0038] Second line (speed estimate): It indicates that the rate of change of the velocity estimate (i.e., the acceleration estimate) consists of three parts: the known control acceleration... The estimated disturbance acceleration And a correction term also based on position measurement error. Here It compensated for all unmodeled acceleration terms.

[0039] The third line (perturbation estimation): This is the essence of the entire DESO. The changes are driven by two parts: local update items This indicates that if the system has persistent position tracking errors... Even when control is applied and known disturbance compensation The observer will then assume that this error is caused by an incompletely estimated disturbance. Therefore, it will adjust... The rate of change, until this error is explained and absorbed, i.e. Converging to the true perturbation value. Consensus fusion term. This is the key to achieving "distributed" functionality. Let i be the set of neighboring drones, that is, the drones that are directly connected to drone i in the communication network. These are the elements of the communication topology adjacency matrix. If there is a communication link between UAVs i and j, then... (Usually set to 1), otherwise . Let be the lumped disturbance estimate received from neighboring drone j. This is the consensus gain coefficient, which adjusts the rate at which the local estimate catches up with its neighbors. The mathematical form of this term is a standard distributed consensus protocol. It calculates the perturbation estimate for drone i. Estimates of its value with all its neighbors The differences between them, and adjustments based on these differences. The change in consensus term drives the lumped disturbance estimate across all UAVs in the cluster to converge. This is crucial for collaborative tasks because even if each UAV experiences slightly different disturbances (e.g., UAVs on the windward and leeward sides), the equivalent disturbance experienced by the entire system as a whole should be cooperatively counteracted. This consensus term effectively filters out local noise from individual UAV observations, forming a more accurate and robust shared understanding of system-level disturbances. Ultimately, DESO outputs an updated lumped disturbance estimate in each control cycle. This value will be used in the calculation of control commands in step S400.

[0040] 3. Determining the Observer Gain Coefficient: DESO's performance largely depends on the gain coefficient. The choice of gain is crucial. An inappropriate gain can lead to slow estimation, overshoot, or even instability. The observer gain coefficient is based on a preset observer bandwidth. The pole placement method was used to determine this. This is a systematic and engineering-oriented tuning method that greatly simplifies the parameter tuning process. Specifically, considering the characteristic equation of a linear ESO, its poles can be placed in the same location. For a third-order system (states p, v, d), its expected characteristic equation is: By matching the coefficients of the actual characteristic equation of ESO with this expected form, the relationship between the gain coefficient and the bandwidth can be directly solved: In this way, the three gain parameters that originally needed to be adjusted independently are simplified into a single parameter with a clear physical meaning—the observer bandwidth. Adjustments. The higher the value, the faster the observer response and the stronger its ability to estimate and track disturbances, but the more sensitive it is to measurement noise. In practical applications, a suitable value is usually selected. It should be 5-10 times the controller bandwidth to ensure that disturbance estimation is much faster than the control loop response.

[0041] S300: In collaborative lifting missions, the planned trajectory is typically the desired motion trajectory of the load, not the trajectory of a single drone. Therefore, a step is needed to decompose the load's target trajectory into the target trajectories of each drone. Each drone determines its own target trajectory based on the preset load target trajectory and the geometry of the lifting system.

[0042] Target payload trajectory: Typically provided by the mission planning layer or ground station, including the expected position of the payload centroid at various time points. ,speed and acceleration and possible desired posture Geometric configuration: refers to the relative positional relationship between the drone and the load. Rigid connection: if the drone is fixed to the load by a rigid member, then the position vector of each drone i relative to the load's center of mass is... It is fixed. At this point, the target position of UAV i can be calculated as: Its target speed and target acceleration This can be obtained by differentiating the above equation. Flexible cable connection: For suspended loads, it is generally desirable for the cable to remain taut and perpendicular to the drone's plane. Assume the cable length is... Then the target position of UAV i can be simply set as an offset of the load target position in the vertical direction: Similarly, the target velocity and acceleration can be obtained by differentiation. More complex models will consider the desired cable swing angle. Through this step, the macroscopic payload task is precisely decomposed into microscopic trajectories that each drone needs to track independently. .

[0043] S400: Each UAV calculates and generates control commands to drive its own motion. This step is the closed-loop part of the control loop. Using the previously obtained state information, disturbance estimation, and target trajectory, the final control force applied to the UAV is calculated. Each UAV calculates and generates control commands based on its own target trajectory, current state information, and lumped disturbance estimate. 1. The calculation structure of the control commands is described in detail, which includes two key parts:

[0044] Disturbance feedforward compensation term: This term aims to offset the effect of the lumped disturbance estimate. Since step S200 has already estimated the unknown disturbance acting on the UAV... Therefore, when calculating control commands, a force of equal magnitude but opposite direction can be actively applied. This compensates for the disturbance in advance. This makes the drone appear to operate in a "disturbance-free" environment, greatly improving control performance and stability. Feedback control term: This term is the core of the standard controller. Its function is to drive the drone to track the target trajectory based on the error between the target trajectory and the current state information. Even after the disturbance is compensated for, the drone may still deviate from the target trajectory due to system response delays, small model errors, etc. The role of the feedback control term is to generate a corrective force based on these deviations (errors), ensuring that the drone can stably track the target. 2. The specific calculation formula for the control command is given. For the drone numbered i, its control command... The specific calculation formula is as follows: ,

[0045] The detailed explanation of this formula is as follows: Let i be the nominal mass of the drone. These are the target position, target velocity, and target acceleration of UAV i (from S300). These are the actual position and actual speed of drone i (from S100). and These are the gain matrices of the feedback controller, typically diagonal positive definite matrices. They represent the proportional gain and the derivative gain, respectively. This is the lumped disturbance estimate generated by DESO in S200. The control law can be clearly divided into two parts:

[0046] Feedback control section This is essentially a PD (proportional-derivative) controller with a feedforward term. It is an acceleration feedforward term. If the system model is perfect, only the acceleration feedforward term is applied. The force can make the drone accelerate towards the target. It is a proportional term that generates a force to "pull back" to the target position based on the current position error. This is the differential term, which generates a damping force based on the current velocity error to reduce overshoot and oscillations, making the tracking process smoother. (Disturbance feedforward compensation section) This is the disturbance feedforward compensation term. It directly subtracts the estimated disturbance force from the command calculated by the feedback controller. This simple subtraction operation is equivalent to applying an "anti-disturbance force" to the system, thereby achieving active suppression of all model uncertainties and external disturbances. Through this "feedback + feedforward" structure, the method of this invention achieves a balance between high-performance trajectory tracking (guaranteed by PD feedback) and strong stability (guaranteed by disturbance feedforward compensation).

[0047] Additional step: Control command calculated by control allocation It is a three-dimensional force vector that cannot be directly executed by the drone's actuators (i.e., multiple rotor motors). Therefore, an additional step is required to convert it into instructions that the underlying controller can understand. The method of this invention also includes, after step S400, each drone transmitting the control instructions... Decomposed into expected resultant force and expected posture .

[0048] Expecting to work together This is the total thrust required by the drone, directed along the z-axis of the aircraft. Its magnitude is equal to the control command vector. The projection of the machine along the desired z-axis. In simpler terms, it can be approximated as... Expected stance To make the total thrust direction align with the desired force vector To maintain the same orientation, the drone must tilt. The desired attitude (usually represented by a rotation matrix or quaternion) is the one that aligns the drone's z-axis. The desired orientation is then determined. This pair of (desired resultant force, desired attitude) commands is subsequently sent to the UAV's inner-loop attitude controller (typically a high-frequency PID controller integrated into the flight controller firmware). This inner-loop controller is responsible for calculating the required rotational speed for each rotor motor to accurately generate the desired resultant force and track the desired attitude, ultimately enabling the execution of commands from the upper layer. The execution of this hierarchical control architecture is a standard practice in drone control.

[0049] To further illustrate the implementation process and beneficial effects of the method of the present invention, several specific cases will be used below.

[0050] Case 1: Cooperative Transportation of a Rigid Container with Unknown Parameters by Quadrotor Drones. Scenario Description: This example simulates a typical cooperative transportation task. Four quadcopter drones are fixed to the four corners of a 1m x 1m x 0.5m container via rigid connectors. The total mass of the container is unknown (approximately 10kg), and its mass distribution is uneven, causing its center of mass to deviate from the geometric center (model uncertainty). The task requires the drone swarm to take off vertically from the ground, following a figure-eight trajectory in the air, during which it will encounter sudden lateral gusts (external environmental disturbances).

[0051] Hardware and Software Platform: Hardware Platform: Utilizing a DJIF450 rack, it houses the Pixhawk4 flight controller as the underlying attitude control unit. Each UAV is equipped with an NVIDIA Jetson Nano as a companion computer, responsible for running the S100-S400 control algorithm proposed in this invention. Positioning employs a U-blox ZED-F9PRTK-GPS module, providing centimeter-level positioning accuracy. Inter-UAV communication is achieved via a 5.8GHz Wi-Fi module, forming a fully connected communication topology. For all Software platform: The control algorithm is developed on the Ubuntu operating system, based on the Robot Operating System (ROS) framework. The S100-S400 loop runs at a frequency of 50Hz. Implementation process and corresponding technical characteristics:

[0052] Initialization: The UAV is powered on and the ROS communication network is established. The ground station sends out a figure-eight shaped target trajectory. Each UAV calculates its own target trajectory to be tracked using the S300 based on its fixed position (geometric configuration) on the container and the target trajectory. Takeoff and Hovering: The drones take off in a coordinated manner. During the hovering phase, due to the load's center of gravity offset, the entire system experiences a constant overturning moment. S100: Each drone continuously acquires its own position. and speed S200: DESO for each UAV begins operation. The asymmetric cable tension caused by the centroid offset is treated as a constant lumped perturbation (corresponding to model uncertainty). DESO's... It will quickly converge to a non-zero constant value. The consensus fusion term ensures that the four UAVs have a consistent estimate of the systematic disturbance caused by the centroid bias. S400: In calculating control commands At that time, the PD feedback term will attempt to correct the hovering position error, but more importantly, the disturbance feedforward compensation term... It actively and asymmetrically adjusts the thrust of the four drones. For example, the thrust of the drone with its center of mass shifted to one side decreases, while the thrust of the drone on the other side increases, thus generating a balancing torque that allows the enclosure to remain horizontally hovering even with an unknown center of mass. Trajectory tracking and gust resistance: The drone swarm begins tracking a figure-eight trajectory. At t=20s, a step lateral gust of 8 m / s is applied to the system (corresponding to an external environmental disturbance). S200: The actual position of the drones at the moment the gust arrives. It begins to deviate from expectations, causing the error term in DESO. A dramatic increase. This made Rapid changes, The estimated value rapidly jumps from the constant value during hovering to a new, larger value, accurately quantifying the impact of wind. Simultaneously, the consensus fusion term enables the perturbation information of the windward UAV (whose local perturbation estimation response is the fastest) to be quickly transmitted to the entire cluster, and all UAVs' perturbation information... All converge to a consistent estimate of wind force within several communication cycles. S400: Disturbance feedforward compensation term of the MPC controller. It works immediately. It applies a strong, wind-resistant compensating force before the PD feedback controller reacts to a large position error. This results in only a small trajectory deviation (approximately 10 cm) and allows the system to regain precise tracking in a very short time (approximately 2 seconds). (S400 after: Calculated force vector) The desired resultant force and attitude are decomposed and executed by the inner-loop controller. Comparative analysis: If a traditional PD or MPC controller without disturbance compensation is used, the system will exhibit steady-state error or continuous oscillation when facing an unknown center of mass. In the event of strong gusts, the system will experience significant trajectory deviation (exceeding 1 meter) and may even become uncontrollable. This invention successfully solves these problems through DESO and feedforward compensation, demonstrating superior stability and accuracy.

[0053] Case 2: Cooperative Transport of a Suspended Long Beam by Two Quadrotor UAVs to Suppress Swaying Scenario Description: This embodiment aims to verify the performance of the present invention in handling suspended loads with complex internal dynamic coupling. Two quadrotor UAVs jointly suspend a 2m long beam weighing 15kg via a 1.5m long cable. The task requires the UAV swarm to complete a maneuver of acceleration, a 90-degree turn, and deceleration. The core objective is to suppress load swaying to the greatest extent possible during this process. Implementation process and corresponding technical features:

[0054] Maneuvering Flight: The drone swarm flies along a predetermined trajectory. During acceleration and turning phases, inertial forces induce violent oscillations in the long beam. S200: The oscillation of the long beam generates complex, time-varying tensions on the drones via cables. For the simplified dynamics model of the drones, this tension represents a strong internal dynamic coupling disturbance. Each drone's DESO estimates this oscillation-induced coupling force as its lumped disturbance in real time. When the load swings forward, This will manifest as a backward pulling force; when the load swings laterally, This manifests as a lateral force. S400: Control command In the calculation, the disturbance feedforward compensation term It played a crucial role in suppressing oscillation. When DESO estimated that the forward oscillation of the load generated a backward disturbance force, the compensation term... This immediately adds a forward component to the drone's control commands. This extra force "holds" the drone, thus suppressing the forward swaying of the load via the cable. Throughout the process, the controller is essentially performing a form of "active damping" control. By estimating and compensating for the coupling force caused by the swaying in real time, it effectively adds a significant virtual damping to the load swaying system, causing it to decay rapidly.

[0055] Comparative Analysis: Using a conventional PID controller, under the same maneuvering conditions, the load would exhibit severe swaying exceeding ±15 degrees, making it difficult to control. The method of this invention, however, can suppress the sway amplitude to within ±2 degrees, significantly improving the stability and safety of the transportation process. This demonstrates the effectiveness of this invention in handling complex internally coupled dynamics. In summary, the core advantage of this multi-drone collaborative hoisting method and system lies in the creative introduction of a Distributed Extended State Observer (DESO) framework. By attributing all model uncertainties and external disturbances to "lumped disturbances," and utilizing a DESO with a consensus protocol for real-time, collaborative estimation, and then eliminating them in the controller through a simple and efficient feedforward compensation term, this invention successfully decouples high-performance control from the dependence on an accurate model. It resolves the technical contradiction in existing technologies where performance is sacrificed for stability, or vice versa. The above embodiments fully demonstrate the superior performance of this invention in handling challenging scenarios such as unknown model parameters, sudden changes in the external environment, and complex internal dynamic coupling. The method of this invention has a clear structure, simple parameter tuning (e.g., through the bandwidth method), and is easy to implement in engineering, providing a robust, efficient, and reliable control solution for multi-drone collaborative operations.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-machine collaborative hoisting method, characterized in that, The method is applicable to a hoisting system consisting of at least two drones, a load connected to the drones via cables, and a communication network between the drones. The process includes the following steps: S100: Each UAV acquires its own current status information, which includes at least position information and speed information; S2 S00: Each UAV, based on its current state information, its control input, and disturbance estimation information received from at least one neighboring UAV via the communication network, runs a distributed extended state observer to generate its own lumped disturbance estimate. The state equation of the distributed extended state observer includes a consensus fusion term to ensure that the lumped disturbance estimates generated by each UAV tend to be consistent. S300: Each UAV determines its own target trajectory based on a preset load target trajectory and the geometric configuration of the hoisting system. S400: Each UAV calculates and generates control commands to drive its own motion based on its own target trajectory, current state information, and lumped disturbance estimate.

2. The method according to claim 1, characterized in that, For the UAV numbered i, the state equation of its distributed extended state observer is as follows: ， in, The actual location of drone i Here is the estimated position of UAV i; Here is the speed estimate for drone i; For the acceleration control term of UAV i; Let be the lumped disturbance estimate for UAV i; , , The observer gain coefficient; This represents the consensus gain coefficient. Let i be the set of neighboring drones; The elements of the communication topology adjacency matrix represent the communication link status between UAV i and UAV j. Let be the lumped disturbance estimate received from neighboring drone j.

3. The method according to claim 1 or 2, characterized in that, In step S400, the calculation of the control command includes: offsetting the influence of the lumped disturbance estimate by a disturbance feedforward compensation term; and driving the UAV to track the target trajectory by a feedback control term based on the error between the target trajectory and the current state information.

4. The method according to claim 3, characterized in that, For the UAV numbered i, its control commands The specific calculation formula is as follows: ， in, For the quality of drone i; , , These are the target position, target velocity, and target acceleration of UAV i, respectively; The actual speed of drone i; and This is the gain matrix of the feedback controller; This is the lumped disturbance estimate generated by the distributed extended state observer.

5. The method according to claim 1, characterized in that, The lumped disturbance includes at least one of the following: external environmental disturbance, internal dynamic coupling of the hoisting system, and uncertainty in the UAV's own model.

6. The method according to claim 2, characterized in that, The observer gain coefficient , , It is determined by using the pole placement method based on the preset observer bandwidth.

7. The method according to claim 1, characterized in that, The method further includes: after step S400, each UAV decomposes the control command into a desired resultant force and a desired attitude, and converts the desired resultant force and desired attitude into control signals for the UAV actuators through an inner-loop attitude controller.

Images