A disturbance decoupling flight control method and system for deploying drones by an inspection robot

Abstract

This invention discloses a disturbance decoupling flight control method and system for deploying unmanned aerial vehicles (UAVs) by an inspection robot. Belonging to the field of UAV flight control technology, it addresses problems such as model mismatch caused by sudden mass changes during deployment, inner loop impact from multi-rate commands, and wasted computing power during hovering. The invention employs a multi-rate cascaded architecture with an outer loop of 20Hz and an inner loop of 1000Hz. Mass reconstruction and decoupling from horizontal wind disturbance are achieved through Z-axis push-weight balancing. A locally recursive third-order compliant device smooths attitude commands. Hysteresis switching between MPC and low-complexity hovering mode is achieved through dual-threshold dual-mode scheduling, and motor dynamics and actuator constraints are incorporated. This invention can quickly restore decoupling stability, reduce inner loop impact, and lower computing power overhead, significantly improving the accuracy and stability of the entire deployment process. It is suitable for deployment operations of power transmission line inspection robots.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control technology, specifically to a disturbance decoupling multi-rate robust flight control method and system for lifting and deploying tasks of power transmission line inspection robots. Background Technology

[0002] As the core infrastructure for power transmission, the long-term safe and stable operation of transmission lines directly affects the overall reliability of the power grid. With the rapid increase in transmission line mileage and the increasingly complex environment along the lines, traditional manual inspections are inefficient, dangerous, and costly, making them unsuitable for modern power grid operation and maintenance. In recent years, the collaborative inspection mode using drones equipped with inspection robots has gradually become the mainstream technical solution for intelligent inspection of transmission lines due to its advantages such as mobility, wide operating range, no power outages, and high safety.

[0003] In actual inspection operations, the UAV needs to perform a continuous closed-loop task of cruise transportation—fixed-point hovering—instantaneous unhooking—resumption of cruise: first, the inspection robot is hoisted to the designated guide wire position and hovered with high precision; then, the robot is instantly unhooked and deployed, allowing it to autonomously crawl and inspect the guide wire; after deployment, the UAV quickly recovers stability and continues to perform subsequent cruise tasks. This task process places extremely high demands on the UAV control system, which not only needs to maintain trajectory tracking accuracy during dynamic flight, but also must possess strong robustness, high stability, and low computational consumption capabilities in key stages such as hovering alignment, instantaneous unhooking, and disturbance recovery.

[0004] Currently, UAVs used for power transmission line inspection generally adopt a multi-rate cascaded control architecture of "outer loop Model Predictive Control (MPC) / Extended State Observer (ESO) + inner loop Incremental Nonlinear Dynamic Inverse (INDI)". The outer loop performs trajectory tracking, attitude reference generation, and disturbance estimation and compensation at a lower frequency (typically 20Hz), while the inner loop achieves rapid attitude stabilization and actuator output at a higher frequency (typically 1000Hz), balancing macroscopic trajectory planning and microscopic attitude response. This architecture has been widely used in conventional UAV flight control.

[0005] However, directly applying the above-mentioned traditional control scheme to the special working conditions of inspection robot lifting and instantaneous deployment will expose several insurmountable technical defects, as follows:

[0006] 1. Decoupling mass abrupt change leads to model mismatch and perturbation contamination.

[0007] The moment the inspection robot detaches, the total mass of the UAV system undergoes a step-like change, directly causing a severe mismatch between key parameters in the outer-loop MPC prediction model, such as thrust input gain and thrust-weight balance, and the actual object. If the controller continues to use the mass parameters before detachment, the predicted state will deviate significantly from the actual state, inducing problems such as peak position and velocity errors, large overshoot, and long oscillation duration during the recovery cruise phase. Simultaneously, existing technologies generally treat the mass mutation, aerodynamic wind disturbance, and unmodeled dynamic uniformity as lumped disturbances for estimation, resulting in the horizontal wind disturbance observation results being severely contaminated by the mass step, the failure of wind disturbance feedforward compensation, further amplifying tracking errors, and even causing attitude instability.

[0008] 2. Multi-rate cascaded structures induce high-frequency impacts and chattering in the inner loop.

[0009] The inspection and deployment task employs a multi-rate architecture with a 20Hz outer loop and a 1000Hz inner loop. The attitude reference output from the outer loop exhibits discrete step characteristics after zero-order hold. When such discontinuous commands are directly input into the high-speed INDI inner loop, they introduce severe high-order dynamic shocks at the moment of command update, manifesting as high-frequency jittering of the UAV fuselage, increased gross stimulation of control torque, and spikes in the attitude tracking curve. Such disturbances not only reduce hovering alignment accuracy but also affect deployment stability, and in severe cases, can lead to robot hook failure, wire collision damage, and threaten the safety of inspection operations. To address the multi-rate shock problem, existing technologies often employ low-pass filtering, tracking differentiators, or global trajectory smoothing. However, low-pass filtering introduces significant phase lag, reducing response speed; global trajectory smoothing requires large-scale optimization calculations within the high-frequency loop, resulting in poor real-time performance and heavy computational burden, neither of which can achieve efficient, lag-free, and computationally efficient attitude command smoothing within the 1000Hz inner loop.

[0010] 3. Wasted computing power and unstable mode switching during long-term hovering phases

[0011] Before deployment, the robot needs to hover in a fixed position for an extended period. The core objective is stable alignment within a small area, rather than complex trajectory rolling optimization. If the outer loop maintains online MPC (Multi-Process Control) solutions throughout the entire process, it will continuously consume a large amount of onboard computing resources, increasing the load on the flight control chip and system power consumption, and shortening the flight time. To reduce computing power consumption, some existing technologies adopt a single threshold mode switching strategy, switching out of MPC when the position error is less than the threshold and switching back to MPC when it is greater than the threshold. However, under the influence of wind disturbance, measurement noise, etc., the position error is prone to fluctuate slightly around the threshold, causing the control mode to frequently switch and repeatedly initialize, which destroys hovering stability and causes attitude jitter and position drift.

[0012] 4. Lack of robust engineering design for deployment scenarios

[0013] Existing control schemes mostly focus on ideal dynamic models and simulation environments, without fully considering practical engineering factors such as actuator physical constraints, motor dynamic response lag, speed limiting, and speed change rate saturation. When running on a real airborne platform, the mismatch between the ideal control quantity and the actuator output capability can easily lead to control command distortion, decreased tracking accuracy, and reduced system stability, making it difficult to reliably implement theoretical algorithms into engineering applications.

[0014] In summary, the traditional "outer loop MPC / ESO + inner loop INDI" multi-rate cascaded control scheme has significant shortcomings in addressing the unique challenges of inspection and deployment, such as instantaneous decoupling quality changes, multi-rate command impacts, long-term hovering computational optimization, and stable switching under strong disturbances. It cannot meet the comprehensive requirements of high precision, high stability, high reliability, and low computational cost for power transmission line inspection robot deployment tasks. Therefore, there is an urgent need to propose a disturbance decoupling flight control method specifically designed for the deployment conditions of inspection robots. This method aims to address core technical pain points such as model mismatch, disturbance contamination, command impacts, and wasted computational power, thereby improving the control performance and engineering practicality of UAVs throughout the entire lifting and deployment process. Summary of the Invention

[0015] The purpose of this invention is to overcome the shortcomings of existing technologies in the deployment of drones by inspection robots on power transmission lines. Addressing technical issues such as mismatch in the outer-loop prediction model due to sudden changes in decoupling quality, contamination of disturbance estimation, high-frequency impacts on the inner loop caused by multi-rate cascaded structures, wasted computing power during long-term hovering, and frequent oscillations during single-threshold switching, this invention provides a disturbance decoupling flight control method and system for deploying drones by inspection robots. Without altering the main framework of "low-rate outer loop + high-rate inner loop" cascaded control, it achieves rapid and stable recovery after decoupling, smooth transition of attitude commands, economical operation with low computing power during hovering, and stable and oscillating mode switching. This improves the control accuracy, stability, robustness, and engineering practicality of drones throughout the entire process of cruise transport, fixed-point hovering, instantaneous decoupling, and resumption of cruise.

[0016] To achieve the above objectives, this invention provides a disturbance decoupling flight control method for deploying drones by an inspection robot, and a flight control system for implementing this method. The complete description is as follows, covering both the method flow and system architecture:

[0017] (I) System Architecture Description

[0018] A disturbance-decoupled flight control system for deploying drones by an inspection robot adopts a multi-rate cascaded flight control architecture consisting of an outer loop low-frequency control and an inner loop high-frequency control. The entire system comprises the following five modules:

[0019] The reference trajectory and delivery task module is used to output the dynamic reference trajectory during the cruise transportation phase, as well as the target location and delivery timing instructions during the hovering delivery phase.

[0020] The outer loop control and disturbance observation module is used to complete trajectory tracking, disturbance observation, mass reconstruction, wind disturbance decoupling, and output desired acceleration, desired yaw angle and attitude reference commands based on the current position, speed and attitude of the UAV.

[0021] The disturbance decoupling and computing power scheduling module integrates a quality reconstruction unit, a horizontal wind disturbance decoupling unit, and a dual-threshold dual-mode scheduling unit to achieve disturbance separation, model correction, and intelligent switching of control modes.

[0022] The inner loop execution module includes a local recursive third-order compliant circuit, a kinematic inverse decoupling unit, a 1000Hz high-speed INDI attitude controller, and a control distribution unit, which are used to smooth the outer loop attitude reference and convert it into actuator control quantities.

[0023] The drone platform and sensor feedback module, including the drone body, inertial measurement unit, positioning module, and actuator motor, are used to output the real motion state and feed it back to the outer loop to form closed-loop control.

[0024] Module connection relationship:

[0025] The reference trajectory and deployment task module output is connected to the outer loop control and disturbance observation module;

[0026] The outer loop control and disturbance observation module is connected to the disturbance decoupling and computing power scheduling module;

[0027] The output of the disturbance decoupling and computing power scheduling module is connected to the inner loop execution module;

[0028] The inner loop execution module output drives the UAV platform and the sensor feedback module;

[0029] The UAV platform and sensor feedback module feed back the status to the outer loop control and disturbance observation module, forming a complete closed loop.

[0030] (II) Description of Control Methods

[0031] A disturbance decoupling flight control method for deploying drones by an inspection robot is proposed. This method is based on a multi-rate cascaded flight control architecture consisting of an outer-loop low-frequency control module and an inner-loop high-frequency control module. The overall system comprises an outer-loop control module, an inner-loop control module, and three core improvement mechanisms.

[0032] 1. A mass reconfiguration and horizontal wind disturbance decoupling mechanism based on the Z-axis thrust-weight balancing channel;

[0033] 2. A locally recursive third-order compliant circuit for the high-speed INDI inner loop input;

[0034] 3. A dual-threshold, dual-mode outer-loop scheduling strategy for the hovering phase.

[0035] With the core concept of "disturbance decoupling, attitude compliance, and computing power scheduling," without changing the original multi-rate cascaded control framework, only through local improvements to the outer loop disturbance observation layer, inner loop input layer, and outer loop scheduling layer, it achieves rapid recovery of model consistency after decoupling, suppression of the impact of outer loop discrete attitude commands on the high-speed inner loop, reduction of computing power overhead for online optimization during hovering, and improvement of mode switching stability. Ultimately, it completes stable control of the entire process of "cruise transportation - fixed-point hovering - instantaneous decoupling - resumption of cruise."

[0036] The detailed technical solutions for each item are as follows:

[0037] 1. Mass Reconfiguration and Horizontal Wind Disturbance Decoupling Mechanism Based on Z-Axis Thrust-Weight Balancing Channel

[0038] This mechanism is used to address the technical problems of mismatch in the outer ring prediction model and contamination of horizontal wind disturbance estimation caused by a sudden change in decoupling quality.

[0039] Based on the vertical lumped disturbance estimate output by the Z-axis linear extended state observer (LESO), and combined with the total thrust command, roll angle, and pitch angle, an instantaneous mass estimation formula is constructed to achieve online mass identification without additional weighing sensors.

[0040] The mean and variance of the instantaneous quality sequence are statistically analyzed using a sliding window. Segmentation judgment and updating are performed by using variance threshold and quality jump threshold to achieve stable identification and rapid updating of quality steps, and suppress false updates caused by noise and gust interference.

[0041] A fast-slow separation strategy is adopted, which uses fast update mass for Z-axis control and smooth isolation for X / Y-axis: Z-axis control uses fast update mass to achieve real-time thrust correction after decoupling; X / Y-axis observers use smoothed mass to avoid horizontal wind disturbance estimation spikes caused by mass step.

[0042] By substituting the reconstructed mass into the nominal horizontal acceleration model of the X / Y axes, the input gain deviation caused by mass change is removed from the lumped disturbance, so that the X / Y axis LESO output only reflects the real aerodynamic wind disturbance. After low-pass filtering, a pure wind disturbance compensation amount is formed, which is used to correct the outer loop MPC prediction window and improve the disturbance compensation accuracy.

[0043] 2. Locally recursive third-order compliant device for high-speed INDI inner loop

[0044] This mechanism is used to solve the technical problem of high-frequency chattering and control torque glitch caused by the direct impact of the 20Hz outer ring discrete attitude reference on the 1000Hz high-speed INDI inner ring.

[0045] With a third-order critical damping system as the core, the recursive calculation is performed at each inner loop sampling time based only on the current outer loop reference value and the previous moment's compliant state, without the need for replanning the entire trajectory, resulting in extremely low computational cost.

[0046] A semi-implicit Euler method is used for numerical updates to ensure numerical stability within the 1000Hz high-frequency loop, and outputs continuous and smooth attitude, angular velocity, and angular acceleration, realizing continuous differentiability of the attitude reference in three layers: angle, angular velocity, and angular acceleration.

[0047] The compliant output is first coupled to the angular velocity proportional loop, then converted into an inner loop angular velocity reference through inverse kinematic decoupling, and finally completed by the INDI inner loop to achieve high-speed attitude tracking. This achieves the division of labor and cooperation between "input shaping + incremental inverse compensation", which balances fast response and stable tracking.

[0048] 3. Dual-threshold dual-mode outer-loop scheduling strategy

[0049] This mechanism is used to solve the technical problems of wasted computing power during long-term hovering phases of MPC continuous online optimization and frequent mode switching caused by single threshold switching.

[0050] Set an entry threshold centered on the hovering target point. With exit threshold ,satisfy This creates a hysteresis switching range, preventing repeated switching caused by error fluctuations near the threshold.

[0051] During the dynamic cruise phase, MPC outer loop control is used to ensure trajectory tracking accuracy. When the position error is less than the entry threshold and the time lock condition is met, the system switches to PID+LESO+INDI low-complexity hovering control mode to reduce computing power consumption.

[0052] When the position error exceeds the exit threshold, switch back to MPC dynamic control mode; maintain the previous time step mode within the hysteresis interval to improve mode stability.

[0053] When switching back to MPC mode from hover mode, the optimizer is warm-started, and the initial optimization value is constructed with the current actual state. The optimizer state is not cleared, reducing switching transients and oscillations.

[0054] 4. Actuator physical constraints and motor dynamic implementation

[0055] To improve engineering feasibility, this invention incorporates actuator constraints and a real motor dynamic model:

[0056] The desired total thrust and control torque are inversely solved by the quadcopter hybrid control matrix into the square of the speed of each motor.

[0057] Limit the minimum speed, maximum speed, and maximum speed change rate of the motor, and introduce a first-order motor dynamic model to simulate the response lag of the real actuator;

[0058] The total thrust and control torque are calculated by substituting the actual motor speed back into the dynamic model of the UAV and feeding back the status, so that the control commands match the actual execution capability and improve the robustness of the system.

[0059] 5. Complete control process

[0060] The present invention is performed according to the following steps within a complete control cycle:

[0061] S1: The reference trajectory and deployment task module outputs the cruise trajectory or hovering target position;

[0062] S2: The outer loop calculates the translational state prediction and control requirements based on the current state;

[0063] S3: The Z-axis LESO completes the instantaneous quality estimation, and the reconstruction quality is updated by sliding window statistics and threshold determination;

[0064] S4: Substitute the reconstructed mass into the nominal X / Y axis model, and obtain the decoupled horizontal wind disturbance from the X / Y axis LESO;

[0065] S5: Wind disturbance estimation is low-pass filtered to form a compensation quantity, which is then input into the outer loop control and MPC prediction stages;

[0066] S6: The dual-threshold dual-mode scheduling module determines the outer loop control mode based on the position error, the current mode, and the time lock.

[0067] S7: The outer ring outputs the desired acceleration, yaw angle, and attitude reference according to the current mode;

[0068] S8: The attitude reference outputs smoothed attitude, angular velocity, and angular acceleration via a locally recursive third-order compliant filter.

[0069] S9: The desired control torque and total thrust are obtained through inverse kinematic decoupling and INDI inner loop calculation;

[0070] S10: The actual motor speed is obtained through the hybrid control matrix, speed limit, speed change rate constraint and first-order motor dynamics;

[0071] S11: Thrust and torque are calculated back to the actual rotational speed, driving the UAV to update its position, velocity, attitude, and angular velocity;

[0072] S12: After the update, the status is fed back to the outer loop observer, the quality reconstruction module and the dual-mode scheduling module, and the next cycle begins.

[0073] Beneficial effects

[0074] This invention addresses four major technical pain points in inspection and deployment scenarios—model mismatch, disturbance contamination, instruction impact, and wasted computing power—through three core improvements: quality reconstruction decoupling, attitude compliance shaping, and dual-threshold computing power scheduling. Compared to the traditional "outer ring MPC / ESO + inner ring INDI" solution, it offers the following significant advantages:

[0075] 1. Decoupling recovery performance is significantly improved, and disturbance estimation is pure and reliable.

[0076] Based on Z-axis thrust-weight balancing, online mass reconstruction is achieved, quickly correcting MPC model parameters and restoring model consistency. By separating fast and slow motion and decoupling from wind disturbance, mass mutations are separated from horizontal disturbances, preventing contamination of disturbance estimates. Experimental data shows that under typical full-process conditions, compared to the baseline solution, this invention reduces cruise RMSE by 72.4% before hovering, reduces peak error during hover transition by 71.8%, reduces peak error after uncoupling by 51.3%, shortens recovery time after uncoupling by 93.1%, and reduces cruise RMSE by 93.0%, significantly improving stability and trajectory tracking accuracy after uncoupling.

[0077] 2. The impact on the inner ring attitude was significantly suppressed, and the stability of the platform was greatly improved.

[0078] The locally recursive third-order compliant mechanism continuously shapes the discrete attitude reference of the outer ring, achieving smoothing of angle, angular velocity, and angular acceleration at three levels, thus reducing the high-order impact caused by multi-rate cascading from the source. Experimental data shows that this invention can reduce the peak control torque by 72.0%, effectively eliminating high-frequency body jitter and control torque glitch, improving wire alignment accuracy and robot deployment stability, and reducing the risk of hook failure and wire collision.

[0079] 3. The computing power consumption is significantly reduced during the hovering phase, and the mode switching is stable and oscillation-free.

[0080] Dual-threshold dual-mode scheduling enables hysteresis switching between high-precision MPC control and low-complexity hovering control. During the hovering phase, MPC online optimization is exited, significantly reducing computing power and energy consumption. Hot start and time lock further reduce switching transients. Experimental data show that compared to the pure MPC scheme, this invention reduces the average outer-loop computation time by 48.3% and the computational overhead during the hovering phase by 94.5%, while completely avoiding frequent flips caused by single-threshold switching, resulting in significantly improved hovering stability.

[0081] 4. The project is highly feasible and adaptable to deployment on real airborne platforms.

[0082] This invention fully integrates engineering processing such as actuator physical constraints, motor first-order dynamics, speed limiting, speed change rate constraints, time lock, and optimizer hot start. It does not rely on additional sensors, does not change the original multi-rate cascaded hardware architecture, has controllable computational load, and can be directly deployed on existing airborne flight control platforms. It combines theoretical advancement with engineering practicality and is suitable for mass application in power transmission line inspection UAV systems.

[0083] 5. Highly versatile and robust, adaptable to complex inspection environments.

[0084] This method can adapt to complex working conditions such as deployment of inspection robots of different weights, wind disturbances of different intensities, and hovering durations of different durations. It maintains high precision and high stability control even under conditions of sudden changes in mass, external wind disturbances, and measurement noise, and has strong robustness and wide applicability. Attached Figure Description

[0085] Figure 1 This is a general block diagram of the disturbance decoupling flight control algorithm of the present invention;

[0086] Figure 2 This is a schematic diagram of the quality mutation identification and wind disturbance decoupling mechanism based on sliding window variance evaluation;

[0087] Figure 3 A schematic diagram of the principle of a local recursive third-order compliant device for the inner loop of INDI;

[0088] Figure 4 A schematic diagram of a dual-mode outer-loop scheduling mechanism based on bidirectional hysteresis boundaries;

[0089] Figure 5 A comparison chart of average comprehensive tracking errors under normal wind disturbance and normal mass change conditions;

[0090] Figure 6 A comparison chart of the average positional error during the recovery phase after a large mass mutation;

[0091] Figure 7 A comparison chart of actual and estimated wind disturbance during the mass mutation phase;

[0092] Figure 8 A comparison chart of control torque at 1000Hz;

[0093] Figure 9 This is a comparison chart of the overall position error under single-threshold and dual-threshold modes;

[0094] Figure 10 A bar chart showing the average cost calculated for different flight phases. Detailed Implementation

[0095] 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. However, it should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0096] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0097] A disturbance decoupling flight control system for deploying drones by an inspection robot employs a multi-rate cascaded flight control architecture of "outer loop control + inner loop control," such as... Figure 1 As shown, the system consists of a reference trajectory and delivery task module, an outer loop control and disturbance observation module, an inner loop execution module, a disturbance decoupling and computing power scheduling module, and a UAV platform and sensor feedback module. The reference trajectory and delivery task module provides the dynamic reference trajectory for the cruise transport phase and the target position for the hovering delivery phase. The outer loop control and disturbance observation module generates the desired acceleration, desired yaw angle, and outer loop control variables based on the current state estimation results, and performs mass updates and wind disturbance compensation. The inner loop execution module receives the attitude reference from the outer loop, and outputs the control torque and total thrust after local recursive third-order compliance, inverse kinematic decoupling, and high-speed INDI control. The UAV platform and sensor feedback module updates the UAV's motion state based on the actual actuator output and feeds back information such as position, attitude, and velocity to the control system.

[0098] To address three key issues throughout the entire process of deploying inspection robots, three targeted improvements were made to the traditional multi-rate cascaded control architecture.

[0099] First, a mass reconstruction and horizontal wind disturbance decoupling mechanism based on the Z-axis push-weight balance channel is introduced in the outer ring disturbance observation layer to quickly update the system mass parameters after the robot is unhooked and reduce the pollution of horizontal wind disturbance estimation by mass mutation.

[0100] Second, a local recursive third-order compliant module is introduced between the outer loop output and the high-speed INDI inner loop to continuously shape the discrete attitude reference of the outer loop and reduce the impact of multi-rate handover.

[0101] Third, a dual-threshold dual-mode scheduling strategy is introduced into the outer loop control layer, enabling the system to use MPC outer loop control during the dynamic cruise phase and switch to a low-complexity hover control branch during the long-term hovering phase, thereby reducing the onboard computing burden.

[0102] Through the above design, the entire system forms a complete flight control framework with three layers of coordination: "safe recovery after decoupling - smooth terminal execution - economical hovering operation".

[0103] From a temporal perspective, within each outer loop sampling cycle, the system first completes the outer loop control and disturbance decoupling calculation based on the reference trajectory and the current state to obtain the desired acceleration, attitude reference, and thrust reference. Subsequently, within a single outer loop cycle, the high-speed inner loop repeatedly performs attitude reference shaping, attitude tracking, control allocation, and actuator dynamic updates at a higher frequency. Finally, the real motor thrust and real control torque drive the UAV platform to complete the physical response and feed the updated state back to the outer loop.

[0104] Therefore, this invention does not involve local modification of a single component, but rather achieves model consistency restoration, attitude smoothness control, and dynamic allocation of computing power within a unified framework.

[0105] A disturbance decoupling flight control method for deploying drones by an inspection robot includes:

[0106] Establish a multi-rate cascaded dynamic model for inspection drones

[0107] To uniformly describe the motion patterns of inspection UAVs throughout the entire process of cruise, hovering, disengagement, and resumption of cruise, a multi-rate cascaded dynamic model of the UAV is first established. This model includes a translational dynamic model, a rotational dynamic model, and a discrete outer-loop prediction model corresponding to control implementation. Among them, the translational dynamic model is mainly used for outer-loop position control, mass reconstruction, and decoupling from horizontal wind disturbance; the rotational dynamic model is mainly used for inner-loop attitude control, control torque generation, and actuator control allocation. Since the outer and inner loops use different control frequencies, this invention further constructs a discrete expression form suitable for outer-loop roll optimization and observer calculation based on continuous dynamics, thereby providing a unified mathematical foundation for subsequent multi-rate control implementation.

[0108] Define the ground inertial coordinate system as The body coordinate system is Let the position state of the UAV be...

[0109]

[0110] Attitude angle is

[0111]

[0112] angular velocity is

[0113]

[0114] The translational and rotational dynamics of an unmanned aerial vehicle (UAV) can be expressed as follows:

[0115]

[0116] in, For system quality, Here is the attitude rotation matrix. For total thrust, Here is the rotational inertia matrix. To control the torque, and These are translational and rotational external disturbances, respectively. Taking a single translational channel as an example, its equivalent dynamics can also be written as...

[0117]

[0118] in Given the control input, This is a lumped disturbance. This model serves as the basis for subsequent LESO, mass identification, and wind disturbance decoupling.

[0119] After abstracting a single translational channel into a standard disturbed second-order model, a linearly extended state observer can be constructed to estimate position, velocity, and lumped disturbance terms. The lumped disturbance estimate of the Z-axis channel is used as the primary basis for mass identification, while the X / Y-axis channels, after mass reconstruction, assume the function of estimating horizontal wind disturbance. The advantages of this approach are twofold: firstly, it preserves the unified expression of the three-dimensional translational system of the multi-rotor UAV; secondly, it provides a direct interface for the subsequent hierarchical design of "first identifying mass, then decoupling wind disturbance."

[0120] In terms of control implementation, the outer loop uses a lower control frequency for rolling optimization and disturbance updates, while the inner loop uses a higher control frequency for attitude tracking and actuator driving. In the current preferred implementation, the outer loop control frequency is 20Hz, and the inner loop attitude control and physics simulation frequency is 1000Hz. The desired attitude and thrust reference output by the outer loop do not directly act on the actuators, but are first continuously shaped by a locally recursive third-order compliant transformer before entering the high-speed inner loop control chain. Thus, the dynamic model not only describes the physical motion laws of the UAV, but also forms the basis of the multi-rate control framework of this invention.

[0121] Mass reconstruction and wind disturbance decoupling based on the Z-axis thrust-weight balancing channel

[0122] Under hovering and uncoupling conditions, mass changes are most directly reflected in the Z-axis thrust-weight balance relationship. Therefore, this invention selects the vertical lumped disturbance estimate from the Z-axis LESO output as the mass identification basis and constructs a mass reconstruction and horizontal wind disturbance decoupling mechanism based on sliding window variance evaluation, such as... Figure 2 As shown. Let the estimated value of the vertical lumped disturbance along the Z-axis be... The total thrust command is The roll angle and pitch angle are respectively and The instantaneous mass estimate is then written as

[0123]

[0124] in, This refers to the nominal quality of the system.

[0125] The physical meaning of the above instantaneous mass estimation formula is that, before and after the robot detaches, if the total thrust command and the current attitude angle are known, the lumped disturbance in the vertical direction that is not explained by the existing model actually mainly corresponds to the deviation in the thrust-weight relationship caused by the change in system mass. In other words, when the robot suddenly detaches, if the controller still outputs thrust according to the mass before detachment, the Z-axis observer will observe a significant "redundant disturbance" within a short period, which can be reinterpreted as a change in mass. In this way, the present invention can complete online mass identification based on existing control and state variables without adding additional weighing sensors.

[0126] To suppress the effects of gusts and measurement noise, the construction length is [length missing]. A sliding window is used to statistically analyze the instantaneous quality series, with the window mean and variance being respectively...

[0127]

[0128] In this invention, the sliding window mean and variance serve the functions of "estimating the new quality level" and "determining whether the system is currently in a stable transition state," respectively. Specifically, when the window variance is large, it indicates that the instantaneous quality estimate is still in a state of significant fluctuation. At this time, the system may have just experienced the moment of decoupling or been affected by strong gusts of wind. If the current estimated value is directly used to update the system quality, the quality parameter will fluctuate rapidly with the noise. When the window variance is small but the mean deviation is large, it indicates that the system has transitioned from the original quality level to a new stable quality level. At this time, the quality should be updated quickly to restore the consistency of the outer loop model as soon as possible. When the window variance is small and the mean deviation is also small, it indicates that there is no significant quality jump, only estimation steady-state error or slow drift. At this time, low-pass smoothing correction is sufficient. Through the above three state determinations, this invention ensures the speed of quality update after decoupling while avoiding erroneous updates caused by observation noise.

[0129] Based on the window variance and quality deviation, the reconstruction quality is... The update is performed in segments: when the variance is large, the quality estimate from the previous time step is kept unchanged; when the variance is small and the mean deviation exceeds the quality jump threshold, a window mean is used for rapid updating; when the variance is small and the mean deviation does not exceed the threshold, a low-pass filter is used for smooth correction. This update law can be written as follows:

[0130] in, The variance threshold, For quality jump threshold, This is a smoothing factor.

[0131] In the code implementation, a "fast-slow separation" strategy was further adopted: the Z-axis control directly uses the quality obtained from the fast update. This allows for immediate thrust correction after uncoupling; while the X / Y axis observer uses the smoothed mass. As known inputs, this avoids direct spikes in horizontal observations caused by abrupt quality changes. The code uses a sliding window, mean, variance, and threshold for quality assessment, and employs smooth quality feeding for the X / Y axis observers. This implementation further enhances decoupling stability.

[0132] The "fast-slow separation" strategy is a further refinement of this invention in engineering implementation. Its core idea is as follows: For Z-axis control, the most critical aspect after decoupling is to quickly correct the vertical thrust to avoid maintaining excessively large thrust commands due to mass changes. Therefore, the Z-axis control chain prioritizes the rapidly updated mass parameters. For X / Y-axis disturbance observation, directly inputting the step-updated mass into the horizontal observer can easily trigger a sudden change in equivalent nominal acceleration at the moment of mass switching, causing a spike in horizontal wind disturbance estimation. Therefore, the X / Y-axis observation chain uses smoothed mass parameters as known input. Through this "fast Z-axis, stable X / Y-axis" approach, both rapid thrust correction after decoupling and stability of horizontal disturbance estimation can be simultaneously achieved.

[0133] In obtaining reconstruction quality Then, it is substituted back into the nominal dynamic model of the horizontal axis to remove the parameter perturbations caused by mass changes from the lumped disturbance, thus obtaining estimates of wind disturbances in the X and Y directions. Neglecting wind disturbances, the nominal horizontal acceleration is...

[0134]

[0135] After resubmitting the reconstructed mass into the nominal horizontal acceleration model in the X / Y directions, the "input gain bias caused by mass change" that was originally mixed in with the lumped disturbance is preferentially absorbed by the nominal model. The remaining unmodeled part in the observer output mainly corresponds to the real aerodynamic wind disturbance. In other words, this invention does not directly "make a stronger estimate" of the wind disturbance, but first eliminates some parameter perturbations that should not originally be attributed to the wind disturbance through mass reconstruction, and then observes and filters the real horizontal wind disturbance. Therefore, the observation results are purer and more suitable for use as an outer loop compensation quantity.

[0136] Will and After being used as known inputs for the X / Y axis LESO, the observer output primarily reflects aerodynamic wind disturbance. The final wind disturbance compensation is then obtained through a first-order low-pass filter.

[0137]

[0138] The decoupled and filtered horizontal wind disturbance compensation can be used to correct the reference state or state offset in the outer loop's future prediction window. This allows the outer loop MPC to consider the impact of wind disturbance in advance during rolling optimization, rather than passively correcting it after the error has occurred. Thus, the outer loop obtains not a disturbance estimate contaminated by mass step errors, but horizontal wind disturbance compensation information consistent with the current mass parameters, thereby improving model prediction consistency and control stability during the recovery cruise phase.

[0139] Local recursive third-order compliant circuit for INDI inner loop input

[0140] Since the outer loop output attitude reference exhibits a discrete step form after zero-order hold, directly inputting it into the 1000Hz high-speed INDI inner loop can easily trigger high-order dynamic shocks at the transition moment. Therefore, this invention employs a locally recursive third-order compliant circuit facing the inner loop input, the principle of which is as follows: Figure 3 As shown. The "local recursion" described in this invention does not involve offline replanning of the entire reference trajectory. Instead, it refers to generating a smooth attitude reference for the next moment in real time at each inner loop sampling moment, based solely on the current outer loop discrete reference value, the previous moment's compliance state, and the current recursion result. Compared to the whole trajectory smoothing method, this method does not require solving a large-scale optimization problem in the 1000Hz high-frequency control loop, and can achieve continuous transitions of angle, angular velocity, and angular acceleration with extremely low computational cost. Therefore, it is more suitable for embedding in high-speed attitude control chains.

[0141] any attitude channel For example, let the outer ring discrete reference be... The compliant states are respectively smooth posture. Smooth angular velocity and smooth angular acceleration Then, the following third-order critical damping system can be constructed:

[0142]

[0143] To ensure numerical stability at a microstep size of 1000Hz, a semi-implicit Euler method is used for recursive updates:

[0144]

[0145] in, This is the sampling period for the inner loop. In the above recursive process, the smoothing angle... Directly used as an attitude reference for entering the inner loop, smoothing angular velocity. Used as feedforward velocity information for the inner loop to smooth angular acceleration. This ensures that the reference does not undergo abrupt changes at the higher-order derivative level. Since all three are derived recursively from the same third-order critical damping system, they can maintain a consistent dynamic transition relationship. With this treatment, even if the outer-loop attitude reference is updated in a step manner at a frequency of 20Hz, the reference trajectory entering the 1000Hz inner loop is still a continuous and differentiable reference trajectory.

[0146] The code implementation further provides three engineering details: (1) The roll, pitch, and yaw channels all execute the third-order recursion independently; (2) The softener bandwidth uses a fixed parameter. (3) After deriving the smooth attitude reference, it is also combined with the angular velocity proportional loop and kinematic inverse decoupling to form the inner loop final reference. This makes the compliance mechanism not only theoretically sound, but also has a direct implementation method in the 1000Hz inner loop.

[0147] In a preferred embodiment of the invention, the compliance unit output is not directly fed into the control torque calculation. Instead, it is first coupled with the angular velocity proportional loop, then converted into an inner loop angular velocity reference via inverse kinematic decoupling, and finally the control torque is generated by the INDI inner loop. In other words, the compliance unit performs the "input shaping" function, while the INDI is responsible for "high-speed attitude tracking and incremental compensation." The two are interconnected in the control chain: the former smooths out the outer loop discrete commands, while the latter executes the smoothed reference quickly and accurately. Through this division of labor, the invention can simultaneously ensure both the speed and smoothness of the inner loop.

[0148] Dual-threshold dual-mode outer-loop scheduling strategy

[0149] The primary objective of the long-term hovering phase is to maintain stable alignment of the drone and lifting robot near the target area, rather than continuously performing complex trajectory optimization. Therefore, this invention employs a dual-threshold, dual-mode outer-loop scheduling strategy, such as... Figure 4 As shown, hovering the target point Centered on the target point, the system's operating area can be divided into three parts: the innermost layer is the PID hovering mode operating area, the middle layer is the hysteresis interval enclosed by the entry and exit thresholds, and the outermost layer is the MPC dynamic control mode operating area. When the system approaches the target point from the outside in, it is only allowed to switch to the low-complexity hovering mode after entering the inner threshold; when the system deviates from the target point from the inside out, it is only forced to switch back to MPC mode after exceeding the outer threshold. This dual-boundary design avoids repeated switching back and forth around a single threshold. Let the current position be... Hovering target point is The hovering error is defined as follows:

[0150]

[0151] Set entry thresholds around the target point. and exit threshold and satisfy

[0152]

[0153] Let pattern variables Indicates the current control mode, where Indicates MPC dynamic mode, To indicate a low-complexity hovering mode, the switching logic is written as follows:

[0154]

[0155] In the aforementioned switching logic, the first branch corresponds to the trigger condition of "transitioning from dynamic control to hover control"; the second branch corresponds to the trigger condition of "exiting hover control and resuming dynamic control"; and the third branch is used to maintain the control mode from the previous moment within the hysteresis interval, thereby enhancing the mode-keeping capability near the boundary. Compared with traditional single-threshold triggering, the dual-threshold logic of this invention will not continuously reset the mode due to small fluctuations in error around a certain single threshold, thus making it more suitable for inspection and deployment scenarios with wind disturbance and measurement noise.

[0156] Based on this, the outer-loop control law can be expressed as:

[0157]

[0158] in, This is the current state. For dynamic reference state, The target is in a hovering state. and These represent the reconstruction quality and the filtered perturbation estimate, respectively.

[0159] In the code implementation, the dual-mode scheduling also incorporates two implementation details that further enhance stability: (1) a time lock is set for entering hover mode, allowing switching to low-complexity mode only within a specified time window and when the position error is sufficiently small; (2) when switching back to MPC from hover mode, the optimizer initial value is not cleared, but instead constructed using the current actual state. Enables warm starts. These details help reduce mode switching transients and avoid accidental switching.

[0160] The time lock prevents the system from prematurely switching to a low-complexity mode before entering the predetermined hovering phase due to a decrease in short-term errors, thus preventing premature exit from optimization control during the dynamic phase. The warm start mechanism ensures that when the system switches back from hovering mode to MPC mode, the optimizer does not start from zero but constructs new initial values ​​using the current actual state, allowing for a smoother recovery of the optimization solution. While these two details pertain to implementation-level optimization, they play a crucial role in reducing mode-switching transients and preventing erroneous switches.

[0161] Actuator physical constraints and motor dynamic implementation

[0162] To ensure the engineering feasibility of this invention, actuator constraints and a motor dynamic model are introduced into the control implementation. The code sets the minimum and maximum motor speeds, maximum speed change rate, first-order response time constant, and maximum total thrust. The desired total thrust and torque are inversely solved using a quadcopter hybrid control matrix to obtain the squares of the motor speeds. The speeds and speed change rates are then limited, and the actual motor speeds are obtained through first-order motor dynamics. Subsequently, the actual total thrust and torque are calculated back from the actual motor speeds for use in dynamic simulations. This design ensures that the control input no longer remains a purely ideal control quantity but reflects the physical limitations of the actuator.

[0163] In the quadcopter control distribution, the speeds of the four motors are set as follows: The thrust coefficient is The reverse torque coefficient is The length of the arm is Then the total thrust and the three-axis control torque satisfy the mixed control relationship with the square of the speed of each motor.

[0164]

[0165] Therefore, after obtaining the desired total thrust and desired control torque, the desired square of the speed of each motor can be obtained by inverse solving the hybrid control matrix, and then the desired motor speed can be further obtained.

[0166] To reflect the true dynamics of the actuator, this invention does not assume that the motor speed instantaneously reaches the desired value, but instead introduces a first-order dynamic model of the motor and a speed change rate constraint. Let the... The commanded speed of each motor is The actual motor speed is The motor time constant is Then it can be written as

[0167]

[0168] Simultaneously, minimum speed, maximum speed, and maximum rate of change of speed constraints are set for each motor to ensure that the actual actuator output meets the physical capability boundaries. Finally, the actual motor speed is used to calculate the actual total thrust and actual control torque, and fed back into the dynamics simulation and state estimation chain. Through this process, the control input of this invention is no longer an ideal quantity, but a real, executable quantity consistent with the actuator's capability.

[0169] The aforementioned physical constraints of the actuator and the dynamic implementation of the motor extend the technical solution of this invention from "feasibility at the control law level" to "feasibility at the execution level." In other words, this invention does not only prove that it can work under ideal thrust and ideal torque conditions, but also maintains the effectiveness of the overall control framework even when considering motor saturation, speed change rate limitations, and the dynamic lag of the actual motor. This is of great significance for subsequent engineering applications.

[0170] The complete execution flow of the method of this invention integrates the above modules. Within a complete control cycle, the method of this invention executes according to the following steps: Step S1, the reference trajectory and deployment task module outputs the dynamic reference trajectory for the cruise transportation phase or the target position for the hovering deployment phase based on the current task phase; Step S2, the outer loop calculates the translational state prediction and outer loop control requirements based on the current position, velocity, attitude, and control state of the previous moment; Step S3, the Z-axis LESO constructs an instantaneous mass estimate based on the lumped disturbance estimate in the vertical direction, and updates the current reconstructed mass by determining the mean and variance of the sliding window; Step S4, the reconstructed mass is substituted into the nominal dynamic model in the X / Y direction to calculate the nominal horizontal acceleration, and the decoupled horizontal wind disturbance estimate is obtained from the X / Y-axis LESO; Step S5, the decoupled horizontal wind disturbance estimate is low-pass filtered to form the final wind disturbance compensation information, and input into the outer loop control or future prediction window correction stage; Step S6, the dual-threshold dual-mode scheduling module calculates the translational state prediction and outer loop control requirements based on the current position and the hovering target point. The system determines whether the outer loop of the current cycle adopts MPC dynamic mode or hovering low-complexity mode based on the error, current mode, time lock state, and exit condition. Step S7: The outer loop outputs the desired acceleration, desired yaw angle, and attitude reference according to the current mode. Step S8: The attitude reference enters a local recursive third-order compliant mechanism, which recursively derives smooth attitude, smooth angular velocity, and smooth angular acceleration at each high-speed inner loop sampling moment. Step S9: After kinematic inverse decoupling and high-speed INDI inner loop calculation, the desired control torque and desired total thrust are obtained. Step S10: The desired total thrust and desired control torque are inversely solved into the desired speed of each motor through the quadcopter hybrid control matrix, and the actual motor speed is obtained after speed limiting, speed change rate limiting, and first-order motor dynamics. Step S11: The actual total thrust and actual control torque are obtained by substituting the actual motor speed back into the actual motor speed, and the UAV dynamics model is driven to update its position, velocity, attitude, and angular velocity states. Step S12: The updated states are fed back to the outer loop observer, mass reconstruction module, and dual-mode scheduling module, entering the next control cycle. Thus, this invention forms a complete closed-loop process from reference generation, disturbance decoupling, mode switching, attitude shaping, control execution to state feedback.

[0171] The advantages of this invention do not arise in isolation, but are derived from the corresponding solutions to the aforementioned technical problems.

[0172] First, because this invention establishes a mass reconstruction mechanism through the Z-axis thrust-weight balance channel and re-substitutes the reconstructed mass into the horizontal axis nominal dynamic model, it can quickly restore the consistency between the outer loop model and the real object after the robot disengages, while simultaneously separating the mass mutation from the horizontal wind disturbance estimation. Therefore, this invention can reduce the peak recovery error after disengagement and improve the stability of recovery cruise. The results of the paper show that, under representative full-process conditions, compared with ESO-MPC&INDI, the proposed method reduces the cruise RMSE before hovering, the peak error of entering hovering transition, the peak error after disengagement, the recovery time after disengagement, and the recovery cruise RMSE by 72.4%, 71.8%, 51.3%, 93.1%, and 93.0%, respectively. The relevant full-process tracking error, the recovery stage error after a large mass mutation, and the comparison between the actual wind disturbance and the estimated wind disturbance during the mass mutation stage are as follows: Figure 5 , Figure 6 and Figure 7 As shown.

[0173] Secondly, because this invention introduces a locally recursive third-order compliant mechanism between the outer loop and the high-speed inner loop, it makes the discrete attitude reference of the outer loop continuous in terms of angle, angular velocity, and angular acceleration. Therefore, it can reduce the impact of reference handover under multi-rate cascading, reduce high-frequency fluctuations in attitude response and control torque glitches, thereby improving the platform stability during conductor alignment and deployment. The ablation results given in the abstract show that the complete frame can reduce the peak control torque by 72.0% compared to the baseline group. The corresponding 1000Hz high-speed inner loop control torque comparison is as follows: Figure 8 As shown.

[0174] Third, because this invention employs dual-threshold dual-mode scheduling, the system maintains the MPC outer loop during the dynamic phase and switches to a low-complexity control branch during the hovering phase. Furthermore, it suppresses frequent boundary switching through hysteresis boundaries, thus reducing the online optimization burden while ensuring hovering control performance. The ablation results presented in the paper abstract show that the complete framework reduces the average computation time of the outer loop by 48.3% compared to the baseline group. The corresponding comparisons of the integrated position error under single-threshold and dual-threshold mode switching, as well as the average computational overhead results for different stages, are as follows: Figure 9 and Figure 10 As shown.

[0175] Fourth, because the code implementation also incorporates engineering processes such as actuator constraints, first-order motor dynamics, time locks, and hot start, this solution is not only theoretically sound, but also feasible and stable under the conditions of considering the upper limit of motor speed, the limit of speed change rate, and the replacement of actual thrust.

[0176] Alternative solutions:

[0177] 1. In the quality reconstruction section, besides the sliding window mean / variance evaluation, exponential moving average, steady-state detector, or other equivalent statistical stability determination methods can also be used. As long as the quality step can be identified and the updated quality parameters can be output, a similar purpose can be achieved. This substitution does not change the core idea of ​​"identifying quality changes using the vertical push-weight relationship". Based on the code and paper implementation, the preferred method in this invention is currently sliding window statistical determination.

[0178] 2. The low-pass filter in the horizontal wind disturbance estimation can be replaced by a first-order inertial filter, Kalman filter or other equivalent filter, as long as it can reduce high-frequency noise without destroying the wind disturbance decoupling logic.

[0179] 3. The local recursive third-order smoother can be replaced with other higher-order smoothers that satisfy the continuity of attitude angle, angular velocity and angular acceleration, such as an equivalent higher-order tracker or higher-order reference model. As long as it can be embedded in the high-speed inner loop and the impact of discrete attitude reference is reduced, similar goals can be achieved.

[0180] 4. In dual-threshold dual-mode scheduling, the switching trigger can be a combination of position error and velocity error, or a composite criterion of position error and dwell time, in addition to the Euclidean distance between the current position and the target point. The code already embodies the implementation idea of ​​"distance threshold + time lock," which is an optional alternative.

[0181] 5. The low-complexity control branch during the hovering phase is currently preferably PID+ESO+INDI, but it can also be replaced with other low-complexity hovering controllers, as long as it can still form a dual-mode scheduling relationship with the MPC outer loop during the dynamic phase and maintain switching stability.

[0182] 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 or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A disturbance decoupling flight control method for deploying drones by an inspection robot, characterized in that, A multi-rate cascaded flight control architecture based on outer-loop control + inner-loop control includes: Mass reconstruction and horizontal wind disturbance decoupling based on Z-axis thrust-weight balance channel, local recursive third-order compliant device for high-speed INDI inner loop, and dual-threshold dual-mode outer loop scheduling for hovering phase. Through the above three mechanisms, we can achieve model consistency restoration after decoupling, smooth transition of attitude commands, computing power optimization during hovering, and stable mode switching.

2. The disturbance decoupling flight control method for deploying drones by an inspection robot according to claim 1, characterized in that, The mass reconstruction and horizontal wind disturbance decoupling include: Based on the vertical lumped disturbance estimate output by Z-axis LESO, the instantaneous mass estimate is calculated in combination with the total thrust command, roll angle and pitch angle. A sliding window is used to statistically analyze the mean and variance of instantaneous quality, and the quality is updated and reconstructed in segments according to the variance threshold and the quality jump threshold. Substituting the reconstructed mass into the nominal dynamic model of the X / Y axis, the mass change perturbation is extracted from the lumped disturbance to obtain a pure horizontal wind disturbance estimate, which is then used for outer loop feedforward compensation.

3. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 2, characterized in that, The refactoring quality update rules are as follows: When the window variance is greater than or equal to the variance threshold, the quality of the previous time step is maintained. When the window variance is less than the variance threshold and the mean deviation is greater than the quality jump threshold, the window mean is updated quickly. When the window variance is less than the variance threshold and the mean deviation is less than or equal to the quality jump threshold, it is corrected using low-pass smoothing.

4. The disturbance decoupling flight control method for deploying drones by an inspection robot according to claim 1, characterized in that, A fast-slow separation strategy is adopted, which uses fast update mass for Z-axis control to correct thrust in real time; and smoothed mass for X / Y axis observers to avoid horizontal wind disturbance estimation spikes caused by abrupt mass changes.

5. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 2, characterized in that, Horizontal wind disturbance decoupling includes: Substitute the reconstructed mass into the nominal acceleration model of the X / Y axes to absorb the input gain bias; The actual aerodynamic wind disturbance is output from the X / Y axis LESO, and then filtered by a low-pass filter to form a wind disturbance compensation amount, which is used to correct the outer loop MPC prediction window.

6. The disturbance decoupling flight control method for deploying drones by an inspection robot according to claim 1, characterized in that, The locally recursive third-order compliant circuit is used for continuous shaping of the discrete attitude reference. High-speed INDI inner loop: With a third-order critical damping system as the core, the inner loop sampling time is based solely on the current outer loop reference and the previous moment's compliance state. A semi-implicit Euler method is used to ensure numerical stability, and the output of continuous and smooth attitude, angular velocity and angular acceleration are used as the inner loop reference of INDI.

7. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 6, characterized in that, The compliance output is first coupled to the angular velocity proportional loop, then converted into an inner loop angular velocity reference through inverse kinematic decoupling, and finally the control torque is generated by the INDI inner loop.

8. The disturbance decoupling flight control method for deploying drones by an inspection robot according to claim 1, characterized in that, The dual-threshold dual-mode outer loop scheduling includes: setting an entry threshold centered on the hovering target point. With exit threshold , A hysteresis interval is formed; dynamic cruise uses the MPC outer loop. When the position error is less than the entry threshold and the time lock is met, it switches to the low-complexity hovering mode; when the position error is greater than the exit threshold, it switches back to MPC and maintains the previous mode within the hysteresis interval.

9. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 8, characterized in that, When switching back to MPC from hover mode, the optimizer is warm-started to construct the initial optimization values ​​based on the current state, reducing the switching transient.

10. A disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 8, characterized in that, The low-complexity hovering mode adopts a PID+LESO+INDI control architecture.

11. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 1, characterized in that, It also includes actuator physical constraints and motor dynamic implementation: The desired total thrust and control torque are inversely solved by the quadcopter hybrid control matrix into the square of the speed of each motor. Limit the motor speed and rate of change of speed, and introduce a first-order motor dynamic model to obtain the actual speed; The total thrust and control torque are calculated by substituting the actual rotational speed, driving the UAV and feeding back the status to the control closed loop.

12. The disturbance decoupling flight control method for deploying unmanned aerial vehicles by an inspection robot according to claim 1, characterized in that, A complete control cycle includes the following steps: S1: Output cruise trajectory or hovering target position; S2: Calculate the translational state prediction and control requirements for the outer loop; S3: Z-axis LESO completes instantaneous quality estimation and reconstructed quality update; S4: Substitute the reconstructed mass and obtain the decoupled horizontal wind disturbance from the X / Y axis LESO; S5: Wind disturbance filtering and input outer loop compensation; S6: Dual-threshold dual-mode scheduling determines the outer loop control mode; S7: Outer loop output desired acceleration, yaw angle and attitude reference; S8: Local recursive third-order compliant output smooth attitude reference; S9: The desired torque and thrust are obtained through inverse kinematic decoupling and INDI inner loop; S10: The actual motor speed is obtained through mixed control, limiting, and motor dynamics; S11: Real thrust and torque drive the UAV to update its status; S12: Status feedback is sent to the outer loop, and the next cycle begins.

13. A disturbance decoupling flight control system for an inspection robot deploying a drone, characterized in that, include: Reference trajectory and deployment task module, outer loop control and disturbance observation module, disturbance decoupling and computing power scheduling module, inner loop execution module, UAV platform and sensor feedback module; The system executes the method described in any one of claims 1-12 to complete the entire process of stable control during cruise, hovering, disengagement, and resumption of cruise.

Images